forked from tinygrad/tinygrad
Compare commits
1
Commits
| Author | SHA1 | Date | |
|---|---|---|---|
 github-actions[bot]
|
02318048e9 |
@@ -1,538 +0,0 @@
|
||||
name: Benchmarks
|
||||
env:
|
||||
# TODO: this rescheduling makes gpt2, mixtral and llama unjitted slower
|
||||
# TODO: very slow for llama 70B and resnet training 6 GPU
|
||||
RUN_PROCESS_REPLAY: "1"
|
||||
ASSERT_PROCESS_REPLAY: "0"
|
||||
PYTHONPATH: .
|
||||
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
||||
|
||||
on:
|
||||
push:
|
||||
branches:
|
||||
- master
|
||||
- update_benchmark
|
||||
- update_benchmark_staging
|
||||
workflow_dispatch:
|
||||
inputs:
|
||||
run_process_replay:
|
||||
description: "Run process replay tests"
|
||||
required: false
|
||||
default: false
|
||||
type: boolean
|
||||
|
||||
jobs:
|
||||
testmacbenchmark:
|
||||
name: Mac Benchmark
|
||||
runs-on: [self-hosted, macOS]
|
||||
timeout-minutes: 20
|
||||
defaults:
|
||||
run:
|
||||
shell: bash -o pipefail {0}
|
||||
if: github.repository_owner == 'tinygrad'
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Symlink models and datasets
|
||||
run: |
|
||||
mkdir -p weights
|
||||
ln -s ~/tinygrad/extra/disassemblers/applegpu extra/disassemblers/applegpu
|
||||
ln -s ~/tinygrad/weights/sd-v1-4.ckpt weights/sd-v1-4.ckpt
|
||||
ln -s ~/tinygrad/weights/bpe_simple_vocab_16e6.txt.gz weights/bpe_simple_vocab_16e6.txt.gz
|
||||
ln -s ~/tinygrad/weights/LLaMA weights/LLaMA
|
||||
ln -s ~/tinygrad/extra/datasets/cifar-10-python.tar.gz extra/datasets/cifar-10-python.tar.gz
|
||||
- name: setup staging db
|
||||
if: github.ref == 'refs/heads/update_benchmark_staging'
|
||||
run: |
|
||||
echo "CACHEDB=/tmp/staging.db" >> $GITHUB_ENV
|
||||
rm -f /tmp/staging.db /tmp/staging.db-shm /tmp/staging.db-wal
|
||||
- name: reset process replay
|
||||
run: python3.11 test/external/process_replay/reset.py
|
||||
- name: Run Stable Diffusion
|
||||
run: JIT=1 python3.11 examples/stable_diffusion.py --fp16 --seed 0 --noshow --timing | tee sd.txt
|
||||
- name: Run Stable Diffusion without fp16
|
||||
run: JIT=1 python3.11 examples/stable_diffusion.py --seed 0 --noshow --timing | tee sd_no_fp16.txt
|
||||
- name: Run Stable Diffusion v2
|
||||
run: JIT=1 python3.11 examples/sdv2.py --fp16 --seed 0 --noshow --timing | tee sdv2.txt
|
||||
- name: Run SDXL
|
||||
run: JIT=1 python3.11 examples/sdxl.py --seed 0 --noshow --timing | tee sdxl.txt
|
||||
- name: Run model inference benchmark
|
||||
run: METAL=1 python3.11 test/external/external_model_benchmark.py
|
||||
- name: Test speed vs torch
|
||||
run: BIG=2 MPS=1 python3.11 test/test_speed_v_torch.py | tee torch_speed.txt
|
||||
- name: Test tensor cores
|
||||
run: METAL=1 python3.11 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
- name: Run Tensor Core GEMM (float)
|
||||
run: DEBUG=2 python3.11 extra/gemm/simple_matmul.py | tee matmul.txt
|
||||
- name: Run Tensor Core GEMM (half)
|
||||
run: DEBUG=2 HALF=1 python3.11 extra/gemm/simple_matmul.py | tee matmul_half.txt
|
||||
- name: Run Tensor Core GEMM (bfloat16)
|
||||
run: DEBUG=2 BFLOAT16=1 python3.11 extra/gemm/simple_matmul.py | tee matmul_bfloat16.txt
|
||||
- name: Fuzz Padded Tensor Core GEMM
|
||||
run: METAL=1 M_START=6 M_STOP=10 M_STEP=1 N_START=6 N_STOP=10 N_STEP=1 K_START=6 K_STOP=24 K_STEP=1 TC_OPT=2 DEBUG=2 python3.11 ./extra/gemm/fuzz_matmul.py
|
||||
- name: Run LLaMA
|
||||
run: |
|
||||
JIT=0 python3.11 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_unjitted.txt
|
||||
JIT=1 python3.11 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_jitted.txt
|
||||
- name: Run LLaMA with BEAM
|
||||
run: JITBEAM=2 IGNORE_BEAM_CACHE=1 python3.11 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_beam.txt
|
||||
- name: Run quantized LLaMA
|
||||
run: |
|
||||
python3.11 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing --quantize int8 | tee llama_int8.txt
|
||||
python3.11 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing --quantize nf4 | tee llama_nf4.txt
|
||||
- name: Run LLaMA 7B on 4 (virtual) GPUs
|
||||
run: python3.11 examples/llama.py --gen 1 --size 7B --shard 4 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_four_gpu.txt
|
||||
- name: Run GPT2
|
||||
run: |
|
||||
JIT=0 python3.11 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_unjitted.txt
|
||||
JIT=1 python3.11 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_jitted.txt
|
||||
- name: Run GPT2 w HALF
|
||||
run: HALF=1 python3.11 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half.txt
|
||||
- name: Run GPT2 w HALF/BEAM
|
||||
run: HALF=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 CAST_BEFORE_VIEW=0 python3.11 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
|
||||
- name: Train MNIST
|
||||
run: time PYTHONPATH=. TARGET_EVAL_ACC_PCT=96.0 python3.11 examples/beautiful_mnist.py | tee beautiful_mnist.txt
|
||||
- name: Run 10 CIFAR training steps
|
||||
run: JIT=1 STEPS=10 python3.11 examples/hlb_cifar10.py | tee train_cifar.txt
|
||||
- name: Run 10 CIFAR training steps w HALF
|
||||
run: JIT=2 STEPS=10 DEFAULT_FLOAT=HALF python3.11 examples/hlb_cifar10.py | tee train_cifar_half.txt
|
||||
#- name: Run 10 CIFAR training steps w BF16
|
||||
# run: STEPS=10 DEFAULT_FLOAT=BFLOAT16 python3.11 examples/hlb_cifar10.py | tee train_cifar_bf16.txt
|
||||
- name: Run 10 CIFAR training steps w winograd
|
||||
run: JIT=1 WINO=1 STEPS=10 python3.11 examples/hlb_cifar10.py | tee train_cifar_wino.txt
|
||||
- uses: actions/upload-artifact@v4
|
||||
with:
|
||||
name: Speed (Mac)
|
||||
path: |
|
||||
onnx_inference_speed.csv
|
||||
torch_speed.txt
|
||||
llama_unjitted.txt
|
||||
llama_jitted.txt
|
||||
llama_beam.txt
|
||||
llama_int8.txt
|
||||
llama_nf4.txt
|
||||
llama_four_gpu.txt
|
||||
gpt2_unjitted.txt
|
||||
gpt2_jitted.txt
|
||||
gpt2_half.txt
|
||||
gpt2_half_beam.txt
|
||||
matmul.txt
|
||||
matmul_half.txt
|
||||
matmul_bfloat16.txt
|
||||
sd.txt
|
||||
sd_no_fp16.txt
|
||||
sdv2.txt
|
||||
sdxl.txt
|
||||
beautiful_mnist.txt
|
||||
train_cifar.txt
|
||||
train_cifar_half.txt
|
||||
train_cifar_bf16.txt
|
||||
train_cifar_wino.txt
|
||||
- name: Run process replay tests
|
||||
run: cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3.11 process_replay.py
|
||||
|
||||
testnvidiabenchmark:
|
||||
name: tinybox green Benchmark
|
||||
runs-on: [self-hosted, Linux, tinyboxgreen]
|
||||
timeout-minutes: 20
|
||||
defaults:
|
||||
run:
|
||||
shell: bash -o pipefail {0}
|
||||
if: github.repository_owner == 'tinygrad'
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Print nvidia-smi
|
||||
run: nvidia-smi
|
||||
- name: Symlink models and datasets
|
||||
run: |
|
||||
mkdir -p weights
|
||||
ln -s ~/tinygrad/weights/LLaMA weights/LLaMA
|
||||
ln -s /raid/weights/mixtral-8x7b-32kseqlen weights/mixtral-8x7b-32kseqlen
|
||||
ln -s /raid/weights/LLaMA-2 weights/LLaMA-2
|
||||
ln -s /raid/weights/LLaMA-3 weights/LLaMA-3
|
||||
mkdir -p extra/datasets
|
||||
ln -s /raid/datasets/imagenet extra/datasets/imagenet
|
||||
- name: setup staging db
|
||||
if: github.ref == 'refs/heads/update_benchmark_staging'
|
||||
run: |
|
||||
echo "CACHEDB=/tmp/staging.db" >> $GITHUB_ENV
|
||||
rm -f /tmp/staging.db /tmp/staging.db-shm /tmp/staging.db-wal
|
||||
- name: reset process replay
|
||||
run: test/external/process_replay/reset.py
|
||||
- name: Run model inference benchmark
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 NOCLANG=1 python3 test/external/external_model_benchmark.py
|
||||
- name: Test speed vs torch
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 HALF=1 BIG=2 TORCHCUDA=1 python3 test/test_speed_v_torch.py | tee torch_speed.txt
|
||||
- name: Test speed vs theoretical
|
||||
run: NV=1 IGNORE_BEAM_CACHE=1 BEAM_DEBUG=1 DEBUG=1 python -m pytest -rA test/external/speed_v_theoretical.py --durations=20
|
||||
- name: Test tensor cores
|
||||
run: |
|
||||
NV=1 python3 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
PTX=1 NV=1 python3 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
- name: Run Tensor Core GEMM (CUDA)
|
||||
run: |
|
||||
CUDA=1 HALF=1 DEBUG=2 python3 extra/gemm/simple_matmul.py | tee matmul.txt
|
||||
CUDA=1 BFLOAT16=1 DEBUG=2 python3 extra/gemm/simple_matmul.py | tee matmul_bfloat16.txt
|
||||
- name: Run Tensor Core GEMM (PTX)
|
||||
run: NV=1 PTX=1 HALF=1 DEBUG=2 python3 extra/gemm/simple_matmul.py | tee matmul_ptx.txt
|
||||
- name: Run Tensor Core GEMM (NV)
|
||||
run: NV=1 HALF=1 DEBUG=2 python3 extra/gemm/simple_matmul.py | tee matmul_nv.txt
|
||||
- name: Test NV=1
|
||||
run: DEBUG=2 NV=1 python -m pytest -rA test/test_tiny.py
|
||||
- name: Test CUDA=1
|
||||
run: DEBUG=2 CUDA=1 python -m pytest -rA test/test_tiny.py
|
||||
- name: Run Stable Diffusion
|
||||
run: NV=1 python3 examples/stable_diffusion.py --fp16 --seed 0 --noshow --timing | tee sd.txt
|
||||
- name: Run SDXL
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 python3 examples/sdxl.py --seed 0 --noshow --timing | tee sdxl.txt
|
||||
- name: Run LLaMA
|
||||
run: |
|
||||
NV=1 JIT=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_unjitted.txt
|
||||
NV=1 JIT=1 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_jitted.txt
|
||||
- name: Run LLaMA with BEAM
|
||||
run: NV=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_beam.txt
|
||||
# - name: Run LLaMA 7B on 4 GPUs
|
||||
# run: NV=1 RUN_PROCESS_REPLAY=0 python3 examples/llama.py --gen 1 --size 7B --shard 4 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_four_gpu.txt
|
||||
# - name: Run LLaMA 7B on 6 GPUs
|
||||
# run: NV=1 RUN_PROCESS_REPLAY=0 python3 examples/llama.py --gen 1 --size 7B --shard 6 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_six_gpu.txt
|
||||
- name: Run LLaMA-3 8B BEAM
|
||||
run: NV=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 python3 examples/llama3.py --size 8B --model weights/LLaMA-3/8B-SF-DPO/ --benchmark --temperature 0 | tee llama3_beam.txt
|
||||
- name: Run LLaMA-3 8B on 4 GPUs
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 python3 examples/llama3.py --size 8B --shard 4 --model weights/LLaMA-3/8B-SF-DPO/ --benchmark --temperature 0 | tee llama3_four_gpu.txt
|
||||
- name: Run LLaMA-3 8B on 6 GPUs
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 python3 examples/llama3.py --size 8B --shard 6 --model weights/LLaMA-3/8B-SF-DPO/ --benchmark --temperature 0 | tee llama3_six_gpu.txt
|
||||
- name: Run LLaMA-2 70B
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 MAX_CONTEXT=256 python3 examples/llama.py --gen 2 --size 70B --shard 6 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_2_70B.txt
|
||||
- name: Run Mixtral 8x7B
|
||||
run: time NV=1 RUN_PROCESS_REPLAY=0 python3 examples/mixtral.py --temperature 0 --count 10 --timing | tee mixtral.txt
|
||||
- name: Run GPT2
|
||||
run: |
|
||||
NV=1 JIT=0 python3 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_unjitted.txt
|
||||
NV=1 JIT=1 python3 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_jitted.txt
|
||||
- name: Run GPT2 w HALF
|
||||
run: NV=1 HALF=1 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half.txt
|
||||
- name: Run GPT2 w HALF/BEAM
|
||||
run: NV=1 HALF=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 CAST_BEFORE_VIEW=0 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
|
||||
- uses: actions/upload-artifact@v4
|
||||
with:
|
||||
name: Speed (NVIDIA)
|
||||
path: |
|
||||
onnx_inference_speed.csv
|
||||
torch_speed.txt
|
||||
matmul.txt
|
||||
matmul_bfloat16.txt
|
||||
matmul_ptx.txt
|
||||
matmul_nv.txt
|
||||
sd.txt
|
||||
sdxl.txt
|
||||
llama_unjitted.txt
|
||||
llama_jitted.txt
|
||||
llama_beam.txt
|
||||
llama3_beam.txt
|
||||
llama3_four_gpu.txt
|
||||
llama3_six_gpu.txt
|
||||
llama_2_70B.txt
|
||||
mixtral.txt
|
||||
gpt2_unjitted.txt
|
||||
gpt2_jitted.txt
|
||||
gpt2_half.txt
|
||||
gpt2_half_beam.txt
|
||||
- name: Run process replay tests
|
||||
run: cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
testmorenvidiabenchmark:
|
||||
name: tinybox green Training Benchmark
|
||||
runs-on: [self-hosted, Linux, tinyboxgreen]
|
||||
timeout-minutes: 20
|
||||
defaults:
|
||||
run:
|
||||
shell: bash -o pipefail {0}
|
||||
if: github.repository_owner == 'tinygrad'
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Symlink models and datasets
|
||||
run: |
|
||||
mkdir -p weights
|
||||
ln -s ~/tinygrad/weights/bpe_simple_vocab_16e6.txt.gz weights/bpe_simple_vocab_16e6.txt.gz
|
||||
ln -s ~/tinygrad/weights/LLaMA weights/LLaMA
|
||||
ln -s ~/tinygrad/extra/datasets/cifar-10-python.tar.gz extra/datasets/cifar-10-python.tar.gz
|
||||
ln -s /raid/weights/mixtral-8x7b-32kseqlen weights/mixtral-8x7b-32kseqlen
|
||||
ln -s /raid/weights/LLaMA-2 weights/LLaMA-2
|
||||
mkdir -p extra/datasets
|
||||
ln -s /raid/datasets/imagenet extra/datasets/imagenet
|
||||
- name: setup staging db
|
||||
if: github.ref == 'refs/heads/update_benchmark_staging'
|
||||
run: |
|
||||
echo "CACHEDB=/tmp/staging.db" >> $GITHUB_ENV
|
||||
rm -f /tmp/staging.db /tmp/staging.db-shm /tmp/staging.db-wal
|
||||
- name: reset process replay
|
||||
run: test/external/process_replay/reset.py
|
||||
- name: Fuzz Padded Tensor Core GEMM (NV)
|
||||
run: NV=1 M_START=12 M_STOP=20 M_STEP=1 N_START=6 N_STOP=10 N_STEP=1 K_START=28 K_STOP=36 K_STEP=1 HALF=1 TC_OPT=2 python3 ./extra/gemm/fuzz_matmul.py
|
||||
- name: Fuzz Padded Tensor Core GEMM (PTX)
|
||||
run: NV=1 PTX=1 M_START=12 M_STOP=20 M_STEP=1 N_START=6 N_STOP=10 N_STEP=1 K_START=28 K_STOP=36 K_STEP=1 HALF=1 TC_OPT=2 python3 ./extra/gemm/fuzz_matmul.py
|
||||
- name: Train MNIST
|
||||
run: time PYTHONPATH=. NV=1 TARGET_EVAL_ACC_PCT=96.0 python3 examples/beautiful_mnist.py | tee beautiful_mnist.txt
|
||||
- name: Run 10 CIFAR training steps
|
||||
run: NV=1 STEPS=10 python3 examples/hlb_cifar10.py | tee train_cifar.txt
|
||||
- name: Run 10 CIFAR training steps w HALF
|
||||
run: NV=1 STEPS=10 DEFAULT_FLOAT=HALF python3 examples/hlb_cifar10.py | tee train_cifar_half.txt
|
||||
- name: Run 10 CIFAR training steps w BF16
|
||||
run: NV=1 STEPS=10 DEFAULT_FLOAT=BFLOAT16 python3 examples/hlb_cifar10.py | tee train_cifar_bf16.txt
|
||||
- name: Run 10 CIFAR training steps w winograd
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 WINO=1 STEPS=10 DEFAULT_FLOAT=HALF python3 examples/hlb_cifar10.py | tee train_cifar_wino.txt
|
||||
- name: Run full CIFAR training w 1 GPU
|
||||
run: time NV=1 DEFAULT_FLOAT=HALF LATEWINO=1 STEPS=1000 TARGET_EVAL_ACC_PCT=93.2 python3 examples/hlb_cifar10.py | tee train_cifar_one_gpu.txt
|
||||
- name: Run full CIFAR training steps w 6 GPUS
|
||||
run: time RUN_PROCESS_REPLAY=0 NV=1 DEFAULT_FLOAT=HALF STEPS=350 BS=1536 GPUS=6 TARGET_EVAL_ACC_PCT=93.2 python3 examples/hlb_cifar10.py | tee train_cifar_six_gpu.txt
|
||||
- name: Run MLPerf resnet eval on training data
|
||||
run: time NV=1 MODEL=resnet python3 examples/mlperf/model_eval.py
|
||||
- name: Run 10 MLPerf ResNet50 training steps (1 gpu)
|
||||
run: NV=1 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=256 GPUS=1 MODEL=resnet python3 examples/mlperf/model_train.py | tee train_resnet_one_gpu.txt
|
||||
- name: Run 10 MLPerf ResNet50 training steps (6 gpu)
|
||||
run: NV=1 RUN_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=1536 GPUS=6 MODEL=resnet python3 examples/mlperf/model_train.py | tee train_resnet.txt
|
||||
- uses: actions/upload-artifact@v4
|
||||
with:
|
||||
name: Speed (NVIDIA Training)
|
||||
path: |
|
||||
beautiful_mnist.txt
|
||||
train_cifar.txt
|
||||
train_cifar_half.txt
|
||||
train_cifar_bf16.txt
|
||||
train_cifar_wino.txt
|
||||
train_cifar_one_gpu.txt
|
||||
train_resnet.txt
|
||||
train_resnet_one_gpu.txt
|
||||
train_cifar_six_gpu.txt
|
||||
- name: Run process replay tests
|
||||
run: cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
testamdbenchmark:
|
||||
name: tinybox red Benchmark
|
||||
runs-on: [self-hosted, Linux, tinybox]
|
||||
timeout-minutes: 20
|
||||
defaults:
|
||||
run:
|
||||
shell: bash -o pipefail {0}
|
||||
if: github.repository_owner == 'tinygrad'
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Symlink models and datasets
|
||||
run: |
|
||||
mkdir -p weights
|
||||
ln -s ~/tinygrad/weights/bpe_simple_vocab_16e6.txt.gz weights/bpe_simple_vocab_16e6.txt.gz
|
||||
ln -s ~/tinygrad/weights/LLaMA weights/LLaMA
|
||||
ln -s ~/tinygrad/extra/datasets/cifar-10-python.tar.gz extra/datasets/cifar-10-python.tar.gz
|
||||
ln -s /raid/weights/mixtral-8x7b-32kseqlen weights/mixtral-8x7b-32kseqlen
|
||||
ln -s /raid/weights/LLaMA-2 weights/LLaMA-2
|
||||
ln -s /raid/weights/LLaMA-3 weights/LLaMA-3
|
||||
mkdir -p extra/datasets
|
||||
ln -s /raid/datasets/imagenet extra/datasets/imagenet
|
||||
- name: setup staging db
|
||||
if: github.ref == 'refs/heads/update_benchmark_staging'
|
||||
run: |
|
||||
echo "CACHEDB=/tmp/staging.db" >> $GITHUB_ENV
|
||||
rm -f /tmp/staging.db /tmp/staging.db-shm /tmp/staging.db-wal
|
||||
- name: reset process replay
|
||||
run: test/external/process_replay/reset.py
|
||||
- name: setup perflevel
|
||||
run: |
|
||||
examples/mlperf/training_submission_v4.1/tinycorp/benchmarks/bert/implementations/tinybox_red/setup.sh
|
||||
rocm-smi
|
||||
- name: Show off tinybox
|
||||
run: /opt/rocm/bin/rocm-bandwidth-test
|
||||
# TODO: unstable on AMD
|
||||
#- name: Run model inference benchmark
|
||||
# run: LD_PRELOAD="/opt/rocm/lib/libhsa-runtime64.so" HSA=1 NOCLANG=1 python3 test/external/external_model_benchmark.py
|
||||
# TODO: unstable on AMD
|
||||
#- name: Test speed vs torch
|
||||
# run: |
|
||||
# python3 -c "import torch; print(torch.__version__)"
|
||||
# LD_PRELOAD="/opt/rocm/lib/libhsa-runtime64.so" HSA=1 BIG=2 TORCHCUDA=1 python3 test/test_speed_v_torch.py | tee torch_speed.txt
|
||||
- name: Test speed vs theoretical
|
||||
run: AMD=1 IGNORE_BEAM_CACHE=1 BEAM_DEBUG=1 DEBUG=1 python -m pytest -rA test/external/speed_v_theoretical.py --durations=20
|
||||
- name: Test tensor cores
|
||||
run: AMD=1 python3 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
- name: Run Tensor Core GEMM (AMD)
|
||||
run: AMD=1 HALF=1 DEBUG=2 python3 extra/gemm/simple_matmul.py | tee matmul_amd.txt
|
||||
- name: Test AMD=1
|
||||
run: DEBUG=2 AMD=1 python -m pytest -rA test/test_tiny.py
|
||||
- name: Test HIP=1
|
||||
run: DEBUG=2 HIP=1 python -m pytest -rA test/test_tiny.py
|
||||
# TODO: AMD compiler bug causes this to fail
|
||||
#- name: Fuzz Padded Tensor Core GEMM
|
||||
# run: HSA=1 M_START=12 M_STOP=20 M_STEP=1 N_START=12 N_STOP=20 N_STEP=1 K_START=28 K_STOP=36 K_STEP=1 HALF=1 TC_OPT=2 DEBUG=2 python3 ./extra/gemm/fuzz_matmul.py
|
||||
- name: Run Stable Diffusion
|
||||
run: AMD=1 python3 examples/stable_diffusion.py --fp16 --seed 0 --noshow --timing | tee sd.txt
|
||||
- name: Run SDXL
|
||||
run: AMD=1 python3 examples/sdxl.py --seed 0 --noshow --timing | tee sdxl.txt
|
||||
- name: Run LLaMA 7B
|
||||
run: |
|
||||
AMD=1 JIT=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_unjitted.txt
|
||||
AMD=1 JIT=1 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_jitted.txt
|
||||
- name: Run LLaMA 7B with BEAM
|
||||
run: AMD=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_beam.txt
|
||||
# - name: Run LLaMA 7B on 4 GPUs
|
||||
# run: AMD=1 RUN_PROCESS_REPLAY=0 python3 examples/llama.py --gen 1 --size 7B --shard 4 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_four_gpu.txt
|
||||
# - name: Run LLaMA 7B on 6 GPUs
|
||||
# run: AMD=1 RUN_PROCESS_REPLAY=0 python3 examples/llama.py --gen 1 --size 7B --shard 6 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_six_gpu.txt
|
||||
- name: Run LLaMA-3 8B BEAM
|
||||
run: AMD=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 python3 examples/llama3.py --size 8B --model weights/LLaMA-3/8B-SF-DPO/ --benchmark --temperature 0 | tee llama3_beam.txt
|
||||
- name: Run LLaMA-3 8B on 4 GPUs
|
||||
run: AMD=1 RUN_PROCESS_REPLAY=0 python3 examples/llama3.py --size 8B --shard 4 --model weights/LLaMA-3/8B-SF-DPO/ --benchmark --temperature 0 | tee llama3_four_gpu.txt
|
||||
- name: Run LLaMA-3 8B on 6 GPUs
|
||||
run: AMD=1 RUN_PROCESS_REPLAY=0 python3 examples/llama3.py --size 8B --shard 6 --model weights/LLaMA-3/8B-SF-DPO/ --benchmark --temperature 0 | tee llama3_six_gpu.txt
|
||||
- name: Run LLaMA-2 70B
|
||||
run: AMD=1 RUN_PROCESS_REPLAY=0 python3 examples/llama.py --gen 2 --size 70B --shard 6 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_2_70B.txt
|
||||
- name: Run Mixtral 8x7B
|
||||
run: time AMD=1 python3 examples/mixtral.py --temperature 0 --count 10 --timing | tee mixtral.txt
|
||||
- name: Run GPT2
|
||||
run: |
|
||||
AMD=1 JIT=0 python3 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_unjitted.txt
|
||||
AMD=1 JIT=1 python3 examples/gpt2.py --prompt "Hello." --count 10 --temperature 0 --timing | tee gpt2_jitted.txt
|
||||
- name: Run GPT2 w HALF
|
||||
run: AMD=1 HALF=1 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half.txt
|
||||
- name: Run GPT2 w HALF/BEAM
|
||||
run: AMD=1 HALF=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 CAST_BEFORE_VIEW=0 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
|
||||
- uses: actions/upload-artifact@v4
|
||||
with:
|
||||
name: Speed (AMD)
|
||||
path: |
|
||||
onnx_inference_speed.csv
|
||||
torch_speed.txt
|
||||
llama_unjitted.txt
|
||||
llama_jitted.txt
|
||||
llama_beam.txt
|
||||
llama3_beam.txt
|
||||
llama3_four_gpu.txt
|
||||
llama3_six_gpu.txt
|
||||
llama_2_70B.txt
|
||||
gpt2_unjitted.txt
|
||||
gpt2_jitted.txt
|
||||
gpt2_half.txt
|
||||
gpt2_half_beam.txt
|
||||
matmul.txt
|
||||
matmul_amd.txt
|
||||
sd.txt
|
||||
sdxl.txt
|
||||
mixtral.txt
|
||||
- name: Run process replay tests
|
||||
run: cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
testmoreamdbenchmark:
|
||||
name: tinybox red Training Benchmark
|
||||
runs-on: [self-hosted, Linux, tinybox]
|
||||
timeout-minutes: 20
|
||||
defaults:
|
||||
run:
|
||||
shell: bash -o pipefail {0}
|
||||
if: github.repository_owner == 'tinygrad'
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Symlink models and datasets
|
||||
run: |
|
||||
mkdir -p weights
|
||||
ln -s ~/tinygrad/weights/bpe_simple_vocab_16e6.txt.gz weights/bpe_simple_vocab_16e6.txt.gz
|
||||
ln -s ~/tinygrad/weights/LLaMA weights/LLaMA
|
||||
ln -s ~/tinygrad/extra/datasets/cifar-10-python.tar.gz extra/datasets/cifar-10-python.tar.gz
|
||||
ln -s /raid/weights/mixtral-8x7b-32kseqlen weights/mixtral-8x7b-32kseqlen
|
||||
ln -s /raid/weights/LLaMA-2 weights/LLaMA-2
|
||||
mkdir -p extra/datasets
|
||||
ln -s /raid/datasets/imagenet extra/datasets/imagenet
|
||||
- name: setup staging db
|
||||
if: github.ref == 'refs/heads/update_benchmark_staging'
|
||||
run: |
|
||||
echo "CACHEDB=/tmp/staging.db" >> $GITHUB_ENV
|
||||
rm -f /tmp/staging.db /tmp/staging.db-shm /tmp/staging.db-wal
|
||||
- name: reset process replay
|
||||
run: test/external/process_replay/reset.py
|
||||
- name: setup perflevel
|
||||
run: |
|
||||
examples/mlperf/training_submission_v4.1/tinycorp/benchmarks/bert/implementations/tinybox_red/setup.sh
|
||||
rocm-smi
|
||||
- name: Train MNIST
|
||||
run: time PYTHONPATH=. AMD=1 TARGET_EVAL_ACC_PCT=96.0 python3 examples/beautiful_mnist.py | tee beautiful_mnist.txt
|
||||
- name: Run 10 CIFAR training steps
|
||||
run: AMD=1 STEPS=10 python3 examples/hlb_cifar10.py | tee train_cifar.txt
|
||||
- name: Run 10 CIFAR training steps w HALF
|
||||
run: AMD=1 STEPS=10 DEFAULT_FLOAT=HALF python3 examples/hlb_cifar10.py | tee train_cifar_half.txt
|
||||
- name: Run 10 CIFAR training steps w BF16
|
||||
run: AMD=1 STEPS=10 DEFAULT_FLOAT=BFLOAT16 python3 examples/hlb_cifar10.py | tee train_cifar_bf16.txt
|
||||
- name: Run 10 CIFAR training steps w winograd
|
||||
run: AMD=1 WINO=1 STEPS=10 DEFAULT_FLOAT=HALF python3 examples/hlb_cifar10.py | tee train_cifar_wino.txt
|
||||
- name: Run full CIFAR training w 1 GPU
|
||||
run: time AMD=1 DEFAULT_FLOAT=HALF LATEWINO=1 STEPS=1000 TARGET_EVAL_ACC_PCT=93.2 python3 examples/hlb_cifar10.py | tee train_cifar_one_gpu.txt
|
||||
- name: Run full CIFAR training steps w 6 GPUS
|
||||
run: time AMD=1 DEFAULT_FLOAT=HALF STEPS=350 BS=1536 GPUS=6 TARGET_EVAL_ACC_PCT=93.2 python3 examples/hlb_cifar10.py | tee train_cifar_six_gpu.txt
|
||||
- name: Run MLPerf resnet eval
|
||||
run: time AMD=1 MODEL=resnet python3 examples/mlperf/model_eval.py
|
||||
- name: Run 10 MLPerf ResNet50 training steps (1 gpu)
|
||||
run: AMD=1 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=256 GPUS=1 MODEL=resnet python3 examples/mlperf/model_train.py | tee train_resnet_one_gpu.txt
|
||||
- name: Run 10 MLPerf ResNet50 training steps (6 gpu)
|
||||
run: AMD=1 RUN_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=1536 GPUS=6 MODEL=resnet python3 examples/mlperf/model_train.py | tee train_resnet.txt
|
||||
- uses: actions/upload-artifact@v4
|
||||
with:
|
||||
name: Speed (AMD Training)
|
||||
path: |
|
||||
beautiful_mnist.txt
|
||||
train_cifar.txt
|
||||
train_cifar_half.txt
|
||||
train_cifar_bf16.txt
|
||||
train_cifar_wino.txt
|
||||
train_cifar_one_gpu.txt
|
||||
train_resnet.txt
|
||||
train_resnet_one_gpu.txt
|
||||
train_cifar_six_gpu.txt
|
||||
- name: Run process replay tests
|
||||
run: cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
testqualcommbenchmark:
|
||||
name: comma Benchmark
|
||||
runs-on: [self-hosted, Linux, comma]
|
||||
timeout-minutes: 20
|
||||
defaults:
|
||||
run:
|
||||
shell: bash -o pipefail {0}
|
||||
if: github.repository_owner == 'tinygrad'
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: setup staging db
|
||||
if: github.ref == 'refs/heads/update_benchmark_staging'
|
||||
run: |
|
||||
echo "CACHEDB=/tmp/staging.db" >> $GITHUB_ENV
|
||||
rm -f /tmp/staging.db /tmp/staging.db-shm /tmp/staging.db-wal
|
||||
- name: reset process replay
|
||||
run: test/external/process_replay/reset.py
|
||||
- name: validate openpilot 0.9.7
|
||||
run: PYTHONPATH=. FLOAT16=0 IMAGE=2 QCOM=1 taskset -c 4-7 python3 test/external/external_benchmark_openpilot.py https://github.com/commaai/openpilot/raw/v0.9.7/selfdrive/modeld/models/supercombo.onnx | tee openpilot_image_0_9_7.txt
|
||||
- name: benchmark openpilot 0.9.4
|
||||
run: PYTHONPATH=. QCOM=1 taskset -c 4-7 python3 test/external/external_benchmark_openpilot.py https://github.com/commaai/openpilot/raw/v0.9.4/selfdrive/modeld/models/supercombo.onnx | tee openpilot_0_9_4.txt
|
||||
- name: benchmark openpilot 0.9.7
|
||||
run: PYTHONPATH=. QCOM=1 taskset -c 4-7 python3 test/external/external_benchmark_openpilot.py https://github.com/commaai/openpilot/raw/v0.9.7/selfdrive/modeld/models/supercombo.onnx | tee openpilot_0_9_7.txt
|
||||
- name: benchmark openpilot w IMAGE=2 0.9.4
|
||||
run: PYTHONPATH=. NOLOCALS=1 FLOAT16=1 IMAGE=2 QCOM=1 taskset -c 4-7 python3 test/external/external_benchmark_openpilot.py https://github.com/commaai/openpilot/raw/v0.9.4/selfdrive/modeld/models/supercombo.onnx | tee openpilot_image_0_9_4.txt
|
||||
- name: benchmark openpilot w IMAGE=2 0.9.7
|
||||
run: PYTHONPATH=. NOLOCALS=1 FLOAT16=1 IMAGE=2 QCOM=1 taskset -c 4-7 python3 test/external/external_benchmark_openpilot.py https://github.com/commaai/openpilot/raw/v0.9.7/selfdrive/modeld/models/supercombo.onnx | tee openpilot_image_0_9_7.txt
|
||||
- name: openpilot compile3 0.9.7
|
||||
run: PYTHONPATH="." QCOM=1 taskset -c 4-7 python3 examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/v0.9.7/selfdrive/modeld/models/supercombo.onnx
|
||||
- name: openpilot compile3 0.9.7+ tomb raider
|
||||
run: PYTHONPATH="." QCOM=1 taskset -c 4-7 python3 examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/e8bea2c78ffa92685ece511e9b554122aaf1a79d/selfdrive/modeld/models/supercombo.onnx
|
||||
- name: openpilot dmonitoring compile3 0.9.7
|
||||
run: PYTHONPATH="." QCOM=1 taskset -c 4-7 python3 examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/v0.9.7/selfdrive/modeld/models/dmonitoring_model.onnx
|
||||
- name: Run process replay tests
|
||||
run: cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
- uses: actions/upload-artifact@v4
|
||||
with:
|
||||
name: Speed (comma)
|
||||
path: |
|
||||
openpilot_compile_0_9_4.txt
|
||||
openpilot_compile_0_9_7.txt
|
||||
openpilot_0_9_4.txt
|
||||
openpilot_0_9_7.txt
|
||||
openpilot_image_0_9_4.txt
|
||||
openpilot_image_0_9_7.txt
|
||||
@@ -1,30 +0,0 @@
|
||||
name: Deploy Docs
|
||||
on:
|
||||
push:
|
||||
branches:
|
||||
- master
|
||||
- mkdocs
|
||||
permissions:
|
||||
contents: write
|
||||
jobs:
|
||||
deploy:
|
||||
runs-on: ubuntu-latest
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
- name: Configure Git Credentials
|
||||
run: |
|
||||
git config user.name github-actions[bot]
|
||||
git config user.email 41898282+github-actions[bot]@users.noreply.github.com
|
||||
- uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.x
|
||||
- run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV
|
||||
- uses: actions/cache@v4
|
||||
with:
|
||||
key: mkdocs-material-${{ env.cache_id }}
|
||||
path: .cache
|
||||
restore-keys: |
|
||||
mkdocs-material-
|
||||
- run: pip install -e .[docs]
|
||||
- run: mkdocs build --strict
|
||||
- run: mkdocs gh-deploy --force
|
||||
@@ -1,30 +0,0 @@
|
||||
# This workflows will upload a Python Package using Twine when a release is created
|
||||
# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries
|
||||
|
||||
name: Upload Python Package
|
||||
|
||||
on:
|
||||
release:
|
||||
types: [published]
|
||||
workflow_dispatch:
|
||||
|
||||
jobs:
|
||||
deploy:
|
||||
runs-on: ubuntu-latest
|
||||
steps:
|
||||
- uses: actions/checkout@v4
|
||||
- name: Set up Python
|
||||
uses: actions/setup-python@v2
|
||||
with:
|
||||
python-version: '3.x'
|
||||
- name: Install dependencies
|
||||
run: |
|
||||
python -m pip install --upgrade pip
|
||||
pip install setuptools wheel twine
|
||||
- name: Build and publish
|
||||
env:
|
||||
TWINE_USERNAME: ${{ secrets.PYPI_USERNAME }}
|
||||
TWINE_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
|
||||
run: |
|
||||
python setup.py sdist bdist_wheel
|
||||
twine upload dist/*
|
||||
@@ -1,98 +0,0 @@
|
||||
name: Check Line Counts
|
||||
on:
|
||||
pull_request_target:
|
||||
|
||||
# Cancel the workflow in progress in newer build is about to start.
|
||||
concurrency:
|
||||
group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
|
||||
cancel-in-progress: true
|
||||
|
||||
jobs:
|
||||
checkbranch:
|
||||
name: Check PR Branch status
|
||||
runs-on: ubuntu-latest
|
||||
outputs:
|
||||
branchstat: ${{ steps.brstat.outputs.stat}}
|
||||
steps:
|
||||
- name: Check code from PR branch
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
repository: ${{ github.event.pull_request.head.repo.full_name }}
|
||||
ref: ${{ github.event.pull_request.head.sha }}
|
||||
fetch-depth: 0
|
||||
- name: Check whether branch is up-to-date
|
||||
id: brstat
|
||||
run: |
|
||||
git remote add tinygrad https://github.com/tinygrad/tinygrad
|
||||
git fetch tinygrad master
|
||||
echo "${{ github.event.pull_request.head.sha }}"
|
||||
git rev-list --left-right --count tinygrad/master...${{ github.event.pull_request.head.sha }} | awk '{print "Behind "$1" - Ahead "$2""}'
|
||||
count=$(git rev-list --left-right --count tinygrad/master...${{ github.event.pull_request.head.sha }} | awk '{print $1}')
|
||||
if [ $count -gt 0 ]
|
||||
then
|
||||
echo "Current branch is behind tinygrad master branch!"
|
||||
echo "stat=true" >> "$GITHUB_OUTPUT"
|
||||
else
|
||||
echo "stat=false" >> "$GITHUB_OUTPUT"
|
||||
fi
|
||||
|
||||
szdiff:
|
||||
name: Core Library Line Difference
|
||||
permissions:
|
||||
contents: read
|
||||
pull-requests: write
|
||||
runs-on: ubuntu-latest
|
||||
needs: checkbranch
|
||||
if: needs.checkbranch.outputs.branchstat == 'false'
|
||||
steps:
|
||||
- name: Checkout code from PR branch
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
repository: ${{ github.event.pull_request.head.repo.full_name }}
|
||||
ref: ${{ github.event.pull_request.head.sha }}
|
||||
path: pr
|
||||
# the base default to tinygrad master and cannot be other fork branch for security purpose
|
||||
- name: Checkout code from tinygrad master
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
path: base
|
||||
- name: Set up Python 3.10
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: '3.10'
|
||||
- name: Count Line Diff
|
||||
run: |
|
||||
pip install tabulate
|
||||
BASE="$GITHUB_WORKSPACE/base"
|
||||
PR="$GITHUB_WORKSPACE/pr"
|
||||
cp "$BASE/sz.py" .
|
||||
echo "loc_content<<EOF" >> "$GITHUB_ENV"
|
||||
python sz.py "$BASE" "$PR" >> "$GITHUB_ENV"
|
||||
echo "EOF" >> "$GITHUB_ENV"
|
||||
- name: Comment Code Line Diff
|
||||
continue-on-error: false
|
||||
uses: marocchino/sticky-pull-request-comment@v2
|
||||
with:
|
||||
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
||||
ignore_empty: true
|
||||
skip_unchanged: true
|
||||
recreate: true
|
||||
message: ${{ env.loc_content }}
|
||||
|
||||
rebase:
|
||||
name: Core Library Line Difference
|
||||
permissions:
|
||||
pull-requests: write
|
||||
runs-on: ubuntu-latest
|
||||
needs: checkbranch
|
||||
if: needs.checkbranch.outputs.branchstat == 'true'
|
||||
steps:
|
||||
- name: Comment Rebase
|
||||
continue-on-error: false
|
||||
uses: marocchino/sticky-pull-request-comment@v2
|
||||
with:
|
||||
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
||||
skip_unchanged: true
|
||||
recreate: true
|
||||
message: |
|
||||
This branch currently is behind tinygrad/master. The line count difference bot is disabled.
|
||||
@@ -1,594 +0,0 @@
|
||||
name: Unit Tests
|
||||
env:
|
||||
# increment this when downloads substantially change to avoid the internet
|
||||
DOWNLOAD_CACHE_VERSION: '8'
|
||||
RUN_PROCESS_REPLAY: 1
|
||||
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
||||
PYTHONPATH: .
|
||||
|
||||
on:
|
||||
push:
|
||||
branches:
|
||||
- master
|
||||
pull_request:
|
||||
workflow_dispatch:
|
||||
|
||||
jobs:
|
||||
autogen:
|
||||
name: Autogen+Docs
|
||||
runs-on: ubuntu-latest
|
||||
timeout-minutes: 20
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Set up Python 3.12
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.12
|
||||
- name: Install docs dependencies (no cache)
|
||||
run: pip install -e '.[docs]'
|
||||
- name: Use as an external package
|
||||
run: |
|
||||
mkdir $HOME/test_external_dir
|
||||
cd $HOME/test_external_dir
|
||||
python -m venv venv
|
||||
source venv/bin/activate
|
||||
pip install $GITHUB_WORKSPACE
|
||||
python -c "from tinygrad.tensor import Tensor; print(Tensor([1,2,3,4,5]))"
|
||||
pip install mypy
|
||||
mypy -c "from tinygrad.tensor import Tensor; print(Tensor([1,2,3,4,5]))"
|
||||
- name: Run beautiful_mnist with tinygrad only
|
||||
run: |
|
||||
mkdir $GITHUB_WORKSPACE/test_dir
|
||||
cd $GITHUB_WORKSPACE/test_dir
|
||||
python -m venv venv
|
||||
source venv/bin/activate
|
||||
pip install $GITHUB_WORKSPACE
|
||||
cp $GITHUB_WORKSPACE/examples/beautiful_mnist.py .
|
||||
PYTHONPATH=$GITHUB_WORKSPACE BS=2 STEPS=10 python beautiful_mnist.py
|
||||
- name: Test Docs Build
|
||||
run: python -m mkdocs build --strict
|
||||
- name: Test Docs
|
||||
run: |
|
||||
python docs/abstractions2.py
|
||||
python docs/abstractions3.py
|
||||
- name: Test Quickstart
|
||||
run: awk '/```python/{flag=1;next}/```/{flag=0}flag' docs/quickstart.md > quickstart.py && PYTHONPATH=. python quickstart.py
|
||||
- name: Test DEBUG
|
||||
run: DEBUG=100 python3 -c "from tinygrad import Tensor; N = 1024; a, b = Tensor.rand(N, N), Tensor.rand(N, N); c = (a.reshape(N, 1, N) * b.T.reshape(1, N, N)).sum(axis=2); print((c.numpy() - (a.numpy() @ b.numpy())).mean())"
|
||||
- name: Install OpenCL
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
echo "deb [ allow-insecure=yes ] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
|
||||
sudo apt update || true
|
||||
sudo apt install --allow-unauthenticated -y --no-install-recommends opencl-headers \
|
||||
intel-oneapi-runtime-openmp=2023.2.1-16 intel-oneapi-runtime-compilers-common=2023.2.1-16 intel-oneapi-runtime-compilers=2023.2.1-16 \
|
||||
intel-oneapi-runtime-dpcpp-sycl-opencl-cpu=2023.2.1-16 intel-oneapi-runtime-tbb-common=2021.10.0-49541 \
|
||||
intel-oneapi-runtime-tbb=2021.10.0-49541 intel-oneapi-runtime-opencl=2023.2.1-16
|
||||
- name: Install packages (cuda)
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
sudo apt update -y || true
|
||||
sudo apt install -y --no-install-recommends git g++ cmake ninja-build llvm-15-dev zlib1g-dev libglew-dev \
|
||||
flex bison libfl-dev libboost-thread-dev libboost-filesystem-dev nvidia-cuda-toolkit-gcc libzstd-dev
|
||||
- name: Install packages (amd)
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
|
||||
sudo tee /etc/apt/sources.list.d/rocm.list <<'EOF'
|
||||
deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/6.1.2 jammy main
|
||||
EOF
|
||||
echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600
|
||||
sudo apt update || true
|
||||
sudo apt install --no-install-recommends --allow-unauthenticated -y hsa-rocr comgr hsa-rocr-dev liburing-dev libc6-dev
|
||||
curl -s https://api.github.com/repos/Qazalin/remu/releases/latest | \
|
||||
jq -r '.assets[] | select(.name == "libremu.so").browser_download_url' | \
|
||||
sudo xargs curl -L -o /usr/local/lib/libremu.so
|
||||
sudo tee --append /etc/ld.so.conf.d/rocm.conf <<'EOF'
|
||||
/opt/rocm/lib
|
||||
/opt/rocm/lib64
|
||||
EOF
|
||||
sudo ldconfig
|
||||
- name: Compile EfficientNet to C and test it
|
||||
run: |
|
||||
CLANG=1 PYTHONPATH="." python examples/compile_efficientnet.py > recognize.c
|
||||
clang -O2 recognize.c -lm -o recognize
|
||||
cat test/models/efficientnet/Chicken.jpg | ./recognize | grep cock
|
||||
- name: Verify OpenCL autogen
|
||||
run: |
|
||||
cp tinygrad/runtime/autogen/opencl.py /tmp/opencl.py.bak
|
||||
./autogen_stubs.sh opencl
|
||||
diff /tmp/opencl.py.bak tinygrad/runtime/autogen/opencl.py
|
||||
- name: Verify CUDA autogen
|
||||
run: |
|
||||
cp tinygrad/runtime/autogen/cuda.py /tmp/cuda.py.bak
|
||||
cp tinygrad/runtime/autogen/nv_gpu.py /tmp/nv_gpu.py.bak
|
||||
./autogen_stubs.sh cuda
|
||||
./autogen_stubs.sh nv
|
||||
diff /tmp/cuda.py.bak tinygrad/runtime/autogen/cuda.py
|
||||
diff /tmp/nv_gpu.py.bak tinygrad/runtime/autogen/nv_gpu.py
|
||||
- name: Verify AMD autogen
|
||||
run: |
|
||||
cp tinygrad/runtime/autogen/hsa.py /tmp/hsa.py.bak
|
||||
cp tinygrad/runtime/autogen/comgr.py /tmp/comgr.py.bak
|
||||
cp tinygrad/runtime/autogen/amd_gpu.py /tmp/amd_gpu.py.bak
|
||||
./autogen_stubs.sh hsa
|
||||
./autogen_stubs.sh comgr
|
||||
./autogen_stubs.sh amd
|
||||
diff /tmp/hsa.py.bak tinygrad/runtime/autogen/hsa.py
|
||||
diff /tmp/comgr.py.bak tinygrad/runtime/autogen/comgr.py
|
||||
diff /tmp/amd_gpu.py.bak tinygrad/runtime/autogen/amd_gpu.py
|
||||
- name: Verify Linux autogen
|
||||
run: |
|
||||
cp tinygrad/runtime/autogen/libc.py /tmp/libc.py.bak
|
||||
cp tinygrad/runtime/autogen/io_uring.py /tmp/io_uring.py.bak
|
||||
./autogen_stubs.sh libc
|
||||
./autogen_stubs.sh io_uring
|
||||
diff /tmp/libc.py.bak tinygrad/runtime/autogen/libc.py
|
||||
diff /tmp/io_uring.py.bak tinygrad/runtime/autogen/io_uring.py
|
||||
|
||||
uops:
|
||||
name: uops tests
|
||||
runs-on: ubuntu-latest
|
||||
timeout-minutes: 20
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Set up Python 3.12
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.12
|
||||
- name: Cache python packages
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ${{ env.Python3_ROOT_DIR }}/lib/python3.12/site-packages
|
||||
key: uops-packages-${{ hashFiles('**/setup.py') }}-3.12
|
||||
- name: Install dependencies
|
||||
run: pip install -e '.[testing]' --extra-index-url https://download.pytorch.org/whl/cpu
|
||||
- name: Test IMAGE=2 support
|
||||
run: |
|
||||
IMAGE=2 PYTHON=1 python3 test/test_ops.py TestOps.test_gemm
|
||||
IMAGE=2 PYTHON=1 python3 test/test_ops.py TestOps.test_simple_conv2d
|
||||
- name: Test emulated METAL tensor cores
|
||||
run: |
|
||||
DEBUG=2 EMULATE_METAL=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_ops.py TestOps.test_big_gemm
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_METAL=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
- name: Test emulated AMX tensor cores
|
||||
run: PYTHONPATH=. DEBUG=2 AMX=1 EMULATE_AMX=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_ops.py TestOps.test_gemm
|
||||
- name: Test emulated AMD tensor cores
|
||||
run: |
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 FORWARD_ONLY=1 PYTHON=1 N=16 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 FORWARD_ONLY=1 PYTHON=1 N=64 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 FORWARD_ONLY=1 PYTHON=1 N=16 HALF=1 ACC_HALF=1 python3 ./extra/gemm/simple_matmul.py
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 FORWARD_ONLY=1 PYTHON=1 N=64 HALF=1 ACC_HALF=1 python3 ./extra/gemm/simple_matmul.py
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
- name: Test emulated CUDA tensor cores
|
||||
run: |
|
||||
DEBUG=2 EMULATE_CUDA=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_ops.py TestOps.test_gemm_fp16
|
||||
PYTHONPATH="." DEBUG=2 EMULATE_CUDA=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_linearizer.py TestLinearizer.test_tensor_cores TestLinearizer.test_tensor_cores_padded
|
||||
- name: Test emulated INTEL OpenCL tensor cores
|
||||
run: DEBUG=2 EMULATE_INTEL=1 FORWARD_ONLY=1 PYTHON=1 HALF=1 N=64 python3 ./extra/gemm/simple_matmul.py
|
||||
- name: Full test tensor cores
|
||||
run: |
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_METAL=1 FORWARD_ONLY=1 PYTHON=1 python3 ./test/test_linearizer.py TestLinearizer.test_tensor_cores
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 FORWARD_ONLY=1 PYTHON=1 python3 ./test/test_linearizer.py TestLinearizer.test_tensor_cores
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_CUDA=1 FORWARD_ONLY=1 PYTHON=1 python3 ./test/test_linearizer.py TestLinearizer.test_tensor_cores
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_INTEL=1 FORWARD_ONLY=1 PYTHON=1 python3 ./test/test_linearizer.py TestLinearizer.test_tensor_cores
|
||||
PYTHONPATH=. DEBUG=2 AMX=1 EMULATE_AMX=1 FORWARD_ONLY=1 PYTHON=1 python3 ./test/test_linearizer.py TestLinearizer.test_tensor_cores
|
||||
- name: Test tensor cores (TC=3)
|
||||
run: |
|
||||
TC=3 DEBUG=3 EMULATE_METAL=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_ops.py TestOps.test_gemm
|
||||
TC=3 PYTHONPATH=. DEBUG=3 EMULATE_AMD=1 PYTHON=1 N=16 HALF=1 ACC_HALF=0 python3 ./extra/gemm/simple_matmul.py
|
||||
TC=3 DEBUG=3 EMULATE_CUDA=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_ops.py TestOps.test_gemm_fp16
|
||||
TC=3 PYTHONPATH=. DEBUG=3 EMULATE_INTEL=1 PYTHON=1 N=16 HALF=1 python3 ./extra/gemm/simple_matmul.py
|
||||
TC=3 PYTHONPATH=. DEBUG=3 AMX=1 EMULATE_AMX=1 FORWARD_ONLY=1 PYTHON=1 python3 test/test_ops.py TestOps.test_gemm
|
||||
- name: Test device flop counts
|
||||
run: |
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_METAL=1 PYTHON=1 python3 ./test/test_uops_stats.py TestUOpsStatsMatmulHalf
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_AMD=1 PYTHON=1 python3 ./test/test_uops_stats.py TestUOpsStatsMatmulHalf
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_CUDA=1 PYTHON=1 python3 ./test/test_uops_stats.py TestUOpsStatsMatmulHalf
|
||||
PYTHONPATH=. DEBUG=2 EMULATE_INTEL=1 PYTHON=1 python3 ./test/test_uops_stats.py TestUOpsStatsMatmulHalf
|
||||
PYTHONPATH=. DEBUG=2 AMX=1 EMULATE_AMX=1 PYTHON=1 python3 ./test/test_uops_stats.py TestUOpsStats.test_simple_matmul
|
||||
- name: Test dtype with Python emulator
|
||||
run: DEBUG=1 PYTHONPATH=. PYTHON=1 python3 -m pytest -n=auto test/test_dtype.py test/test_dtype_alu.py
|
||||
- name: Test ops with Python emulator
|
||||
run: DEBUG=2 PYTHON=1 python3 -m pytest -n=auto test/test_ops.py -k "not (test_split or test_simple_cumsum or test_cumsum or test_einsum or test_dot or test_dot_1d or test_big_gemm or test_broadcastdot or test_multidot or test_var_axis or test_std_axis or test_broadcast_full or test_broadcast_partial or test_simple_conv3d or test_dilated_conv_transpose2d or test_simple_conv_transpose3d or test_large_input_conv2d or test_max_pool2d or test_max_pool2d_simple or test_max_pool2d_bigger_stride or test_avg_pool2d or test_cat or test_scaled_product_attention or test_scaled_product_attention_causal or test_slice_fancy_indexing_dim_inject_none or test_slice_fancy_indexing_list_indices or test_slice_fancy_indexing_no_dim_collapse or test_slice_fancy_indexing_tuple_indices or test_slice_fancy_indexing_list_with_tensors or test_slice_fancy_indexing_dim_collapse_int or test_interpolate_bilinear or test_interpolate_bilinear_corners_aligned)" --durations=20
|
||||
- name: Test uops with Python emulator
|
||||
run: PYTHON=1 python3 -m pytest test/test_uops.py --durations=20
|
||||
- name: Test symbolic with Python emulator
|
||||
run: PYTHONPATH=. PYTHON=1 python3 test/test_symbolic_ops.py
|
||||
- name: test_linearizer_failures with Python emulator
|
||||
run: PYTHONPATH=. PYTHON=1 python3 -m pytest -rA test/test_linearizer_failures.py::TestLinearizerFailures::test_failure_1
|
||||
|
||||
linter:
|
||||
name: Linters+fuzz+unit Tests
|
||||
runs-on: ubuntu-latest
|
||||
timeout-minutes: 20
|
||||
|
||||
# TODO: run the pre-commit hook to replace a lot of this
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Set up Python 3.10
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: "3.10"
|
||||
- name: Cache python packages
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ${{ env.Python3_ROOT_DIR }}/lib/python3.10/site-packages
|
||||
key: linting-packages-${{ hashFiles('**/setup.py') }}-3.10
|
||||
- name: Install dependencies
|
||||
run: pip install -e '.[linting,testing]' --extra-index-url https://download.pytorch.org/whl/cpu
|
||||
- name: Lint bad-indentation and trailing-whitespace with pylint
|
||||
run: python -m pylint --disable=all -e W0311 -e C0303 --jobs=0 --indent-string=' ' --recursive=y .
|
||||
- name: Lint with ruff
|
||||
run: |
|
||||
pip3 install --upgrade --force-reinstall ruff
|
||||
python3 -m ruff check .
|
||||
- name: Lint tinygrad with pylint
|
||||
run: python -m pylint tinygrad/
|
||||
- name: Run mypy
|
||||
run: python -m mypy --strict-equality --lineprecision-report . && cat lineprecision.txt
|
||||
- name: Test README
|
||||
run: awk '/```python/{flag=1;next}/```/{flag=0}flag' README.md > README.py && PYTHONPATH=. python README.py
|
||||
- name: Run unit tests
|
||||
run: PYTHONPATH="." python -m pytest -n=auto test/unit/
|
||||
- name: Fuzz Test symbolic
|
||||
run: python test/external/fuzz_symbolic.py
|
||||
- name: Fuzz Test shapetracker
|
||||
run: |
|
||||
PYTHONPATH="." python test/external/fuzz_shapetracker.py
|
||||
PYTHONPATH="." python test/external/fuzz_shapetracker_math.py
|
||||
- name: Repo line count <= 9999 lines
|
||||
run: MAX_LINE_COUNT=9999 python sz.py
|
||||
|
||||
testopencl:
|
||||
strategy:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
task: [optimage, onnx]
|
||||
|
||||
name: ${{ matrix.task=='optimage'&&'GPU IMAGE+compile Tests' || matrix.task=='onnx'&&'ONNX+Optimization Tests' }}
|
||||
runs-on: ubuntu-20.04
|
||||
timeout-minutes: 20
|
||||
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Install OpenCL
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
echo "deb [ allow-insecure=yes ] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
|
||||
sudo apt update || true
|
||||
sudo apt install --allow-unauthenticated -y --no-install-recommends \
|
||||
intel-oneapi-runtime-openmp=2023.2.1-16 intel-oneapi-runtime-compilers-common=2023.2.1-16 intel-oneapi-runtime-compilers=2023.2.1-16 \
|
||||
intel-oneapi-runtime-dpcpp-sycl-opencl-cpu=2023.2.1-16 intel-oneapi-runtime-tbb-common=2021.10.0-49541 \
|
||||
intel-oneapi-runtime-tbb=2021.10.0-49541 intel-oneapi-runtime-opencl=2023.2.1-16
|
||||
- name: Set up Python 3.11
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.11
|
||||
- name: Cache python packages
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ${{ env.Python3_ROOT_DIR }}/lib/python3.11/site-packages
|
||||
key: testing-packages-${{ hashFiles('**/setup.py') }}
|
||||
- name: Cache downloads
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ~/.cache/tinygrad/downloads/
|
||||
key: downloads-cache-${{ matrix.task }}-${{ env.DOWNLOAD_CACHE_VERSION }}
|
||||
- name: Install Dependencies
|
||||
run: pip install -e '.[testing,testing_tf]' --extra-index-url https://download.pytorch.org/whl/cpu
|
||||
- if: ${{ matrix.task == 'optimage' }}
|
||||
name: Run Kernel Count Test
|
||||
run: PYTHONPATH="." GPU=1 python -m pytest -n=auto test/external/external_test_opt.py
|
||||
- if: ${{ matrix.task == 'optimage'}}
|
||||
name: Test WINO=1
|
||||
run: GPU=1 DEBUG=2 WINO=1 python3 test/test_ops.py TestOps.test_simple_conv2d
|
||||
- if: ${{ matrix.task == 'optimage'}}
|
||||
name: Test GPU IMAGE=2 ops + training
|
||||
run: |
|
||||
PYTHONPATH="." GPU=1 IMAGE=2 python -m pytest -n=auto test/test_ops.py --durations=20
|
||||
PYTHONPATH="." GPU=1 IMAGE=2 python3 test/models/test_end2end.py TestEnd2End.test_linear_mnist
|
||||
- if: ${{ matrix.task == 'optimage' }}
|
||||
name: Test openpilot model kernel count and gate usage
|
||||
run: |
|
||||
PYTHONPATH="." ALLOWED_KERNEL_COUNT=208 ALLOWED_READ_IMAGE=2131 ALLOWED_GATED_READ_IMAGE=13 FLOAT16=0 GPU=1 IMAGE=2 python examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/v0.9.4/selfdrive/modeld/models/supercombo.onnx
|
||||
- if: ${{ matrix.task == 'optimage' }}
|
||||
name: Test openpilot alt model correctness (float32)
|
||||
run: PYTHONPATH="." FLOAT16=0 DEBUGCL=1 GPU=1 IMAGE=2 python examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/3799fe46b3a629e491d4b8498b8ae83e4c88c304/selfdrive/modeld/models/supercombo.onnx
|
||||
- if: ${{ matrix.task == 'optimage' }}
|
||||
name: Test openpilot fastvits model correctness (float32)
|
||||
run: PYTHONPATH="." FLOAT16=0 DEBUGCL=1 GPU=1 IMAGE=2 python examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/9118973ed03c1ae1d40cf69a29507ec2cc78efd7/selfdrive/modeld/models/supercombo.onnx
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test ONNX (GPU)
|
||||
run: GPU=1 python -m pytest -n=auto test/external/external_test_onnx_backend.py --durations=20
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test ONNX (CLANG)
|
||||
run: CLANG=1 python -m pytest -n=auto test/external/external_test_onnx_backend.py --durations=20
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Run CLOUD=1 Test
|
||||
run: |
|
||||
CLOUDDEV=CLANG CLOUD=1 python3 test/test_tiny.py
|
||||
CLOUDDEV=GPU CLOUD=1 python3 test/test_tiny.py
|
||||
CLOUDDEV=GPU IMAGE=2 CLOUD=1 python3 test/test_tiny.py
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test Optimization Helpers
|
||||
run: PYTHONPATH="." DEBUG=1 python3 extra/optimization/test_helpers.py
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test Action Space
|
||||
run: PYTHONPATH="." DEBUG=1 GPU=1 python3 extra/optimization/get_action_space.py
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test Beam Search
|
||||
run: PYTHONPATH="." GPU=1 IGNORE_BEAM_CACHE=1 python3 -m pytest extra/optimization/test_beam_search.py
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test MLPerf optimizers
|
||||
run: GPU=1 python -m pytest -n=auto test/external/external_test_optim.py --durations=20
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test MLPerf losses
|
||||
run: GPU=1 python -m pytest -n=auto test/external/external_test_losses.py --durations=20
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test MLPerf metrics
|
||||
run: GPU=1 python -m pytest -n=auto test/external/external_test_metrics.py --durations=20
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Test MLPerf datasets
|
||||
run: GPU=1 python -m pytest -n=auto test/external/external_test_datasets.py --durations=20
|
||||
- if: ${{ matrix.task == 'onnx' }}
|
||||
name: Run handcode_opt
|
||||
run: PYTHONPATH=. MODEL=resnet GPU=1 DEBUG=1 BS=4 HALF=0 python3 examples/handcode_opt.py
|
||||
- name: Run process replay tests
|
||||
run: |
|
||||
export PR_TITLE=$(jq -r .pull_request.title "$GITHUB_EVENT_PATH")
|
||||
export COMMIT_MESSAGE=$(git show -s --format=%B ${{ github.event.pull_request.head.sha }})
|
||||
cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
testwebgpu:
|
||||
name: WebGPU Tests
|
||||
runs-on: ubuntu-latest
|
||||
timeout-minutes: 20
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
- name: Set up Python 3.11
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.11
|
||||
- name: Cache python packages
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ~/.local/lib/python3.11/site-packages
|
||||
key: webgpu-testing-user3-packages-${{ hashFiles('**/setup.py') }}
|
||||
- name: Install Dependencies
|
||||
run: pip install --user -e '.[webgpu,testing]' --extra-index-url https://download.pytorch.org/whl/cpu
|
||||
- name: Install dependencies for software-based vulkan
|
||||
run: |
|
||||
sudo apt update -y || true
|
||||
sudo apt install -y libegl1-mesa libgl1-mesa-dri libxcb-xfixes0-dev mesa-vulkan-drivers
|
||||
- name: Cache downloads
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ~/.cache/tinygrad/downloads/
|
||||
key: downloads-cache-webgpu-${{ env.DOWNLOAD_CACHE_VERSION }}
|
||||
- name: Check Device.DEFAULT (WEBGPU) and print some source
|
||||
run: |
|
||||
WEBGPU=1 python -c "from tinygrad import Device; assert Device.DEFAULT == 'WEBGPU', Device.DEFAULT"
|
||||
WEBGPU=1 DEBUG=4 FORWARD_ONLY=1 python3 test/test_ops.py TestOps.test_add
|
||||
- name: Run selected webgpu tests
|
||||
run: |
|
||||
WEBGPU=1 WGPU_BACKEND_TYPE=Vulkan python3 -m pytest -n=auto test/ --ignore=test/external --ignore=test/models --ignore=test/unit \
|
||||
--ignore=test/test_copy_speed.py --ignore=test/test_rearrange_einops.py --ignore=test/test_speed_v_torch.py --ignore=test/test_transcendental.py \
|
||||
--ignore=test/test_fuzz_shape_ops.py --ignore=test/test_linearizer_failures.py --durations=20
|
||||
- name: Run process replay tests
|
||||
run: |
|
||||
export PR_TITLE=$(jq -r .pull_request.title "$GITHUB_EVENT_PATH")
|
||||
export COMMIT_MESSAGE=$(git show -s --format=%B ${{ github.event.pull_request.head.sha }})
|
||||
cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
testmetal:
|
||||
name: Metal Tests
|
||||
runs-on: macos-14
|
||||
timeout-minutes: 20
|
||||
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
fetch-depth: 2 # NOTE: this fetches the HEAD commit of the PR
|
||||
- name: Set up Python 3.11
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.11
|
||||
- name: Cache python packages
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: /Users/runner/Library/Python/3.11/lib/python/site-packages
|
||||
key: metal-m1-testing-user3-packages-${{ hashFiles('**/setup.py') }}
|
||||
- name: Install Dependencies
|
||||
run: pip install --user -e '.[webgpu,testing]' --extra-index-url https://download.pytorch.org/whl/cpu
|
||||
- name: Cache downloads
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ~/Library/Caches/tinygrad/downloads/
|
||||
key: downloads-cache-metal-only-${{ env.DOWNLOAD_CACHE_VERSION }}
|
||||
- name: Check Device.DEFAULT (METAL) and print some source
|
||||
run: |
|
||||
METAL=1 python -c "from tinygrad import Device; assert Device.DEFAULT == 'METAL', Device.DEFAULT"
|
||||
METAL=1 DEBUG=4 FORWARD_ONLY=1 python3 test/test_ops.py TestOps.test_add
|
||||
- name: Run metal test
|
||||
run: JIT=2 METAL=1 python -m pytest -n=auto test/ --ignore=test/external --ignore=test/models --ignore=test/unit --durations=20
|
||||
- name: Run real world test
|
||||
run: JIT=2 METAL=1 python -m pytest -n=auto test/models/test_real_world.py --durations=20
|
||||
- name: Run ONNX
|
||||
run: JIT=2 METAL=1 python -m pytest -n=auto test/external/external_test_onnx_backend.py --durations=20
|
||||
- name: Test tensor core ops (fake)
|
||||
run: TC=2 METAL=1 DEBUG=3 python test/test_ops.py TestOps.test_gemm
|
||||
- name: Test tensor core ops (real)
|
||||
run: METAL=1 DEBUG=3 python test/test_ops.py TestOps.test_big_gemm
|
||||
- name: Test LLaMA compile speed
|
||||
run: PYTHONPATH="." METAL=1 python test/external/external_test_speed_llama.py
|
||||
- name: Test Beam Search
|
||||
run: PYTHONPATH="." METAL=1 IGNORE_BEAM_CACHE=1 python3 -m pytest extra/optimization/test_beam_search.py
|
||||
- name: Fuzz Test linearizer
|
||||
run: PYTHONPATH="." METAL=1 DEPTH=4 FUZZ_N=50 FUZZ_MAX_SIZE=1000000 python test/external/fuzz_linearizer.py
|
||||
# - name: Fuzz Test models schedule
|
||||
# run: FUZZ_SCHEDULE=1 FUZZ_SCHEDULE_MAX_PATHS=5 python -m pytest test/models/test_train.py test/models/test_end2end.py
|
||||
- name: Run TRANSCENDENTAL math
|
||||
run: TRANSCENDENTAL=2 python -m pytest -n=auto test/test_ops.py::TestOps::test_sin test/test_ops.py::TestOps::test_cos test/test_ops.py::TestOps::test_tan test/test_ops.py::TestOps::test_exp test/test_ops.py::TestOps::test_log --durations=20
|
||||
# WebGPU e2e tests
|
||||
- name: Build WEBGPU Efficientnet
|
||||
run: WEBGPU=1 WGPU_BACKEND_TYPE=Metal python3 -m examples.compile_efficientnet
|
||||
- name: Clean npm cache
|
||||
run: npm cache clean --force
|
||||
- name: Install Puppeteer
|
||||
run: npm install puppeteer
|
||||
- name: Run WEBGPU Efficientnet
|
||||
run: node test/web/test_webgpu.js
|
||||
- name: Run process replay tests
|
||||
run: |
|
||||
export PR_TITLE=$(jq -r .pull_request.title "$GITHUB_EVENT_PATH")
|
||||
export COMMIT_MESSAGE=$(git show -s --format=%B ${{ github.event.pull_request.head.sha }})
|
||||
cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
tests:
|
||||
strategy:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
backend: [llvm, clang, gpu, ptx, amd, nv] #, triton]
|
||||
|
||||
name: Tests on (${{ matrix.backend }})
|
||||
runs-on: ubuntu-22.04
|
||||
timeout-minutes: 20
|
||||
|
||||
steps:
|
||||
- name: Checkout Code
|
||||
uses: actions/checkout@v4
|
||||
with:
|
||||
fetch-depth: 2 # NOTE: this fetches the HEAD commit of the PR
|
||||
- name: Set up Python 3.12
|
||||
uses: actions/setup-python@v5
|
||||
with:
|
||||
python-version: 3.12
|
||||
- name: Cache python packages
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ${{ env.Python3_ROOT_DIR }}/lib/python3.12/site-packages
|
||||
key: ${{ matrix.backend }}-packages-${{ hashFiles('**/setup.py') }}
|
||||
- name: Cache downloads
|
||||
uses: actions/cache@v4
|
||||
with:
|
||||
path: ~/.cache/tinygrad/downloads/
|
||||
key: downloads-cache-${{ matrix.backend }}-${{ env.DOWNLOAD_CACHE_VERSION }}
|
||||
- name: Set env
|
||||
run: printf "${{ matrix.backend == 'llvm' && 'LLVM=1' || matrix.backend == 'clang' && 'CLANG=1' || matrix.backend == 'gpu' && 'GPU=1' || matrix.backend == 'PTX' && 'FORWARD_ONLY=1\nJIT=1\nOPT=2\nNV=1\nPTX=1\nMOCKGPU=1' || matrix.backend == 'triton' && 'FORWARD_ONLY=1\nJIT=1\nOPT=2\nNV=1\nMOCKGPU=1\nTRITON=1\nTRITON_PTXAS_PATH=/usr/bin/ptxas' || matrix.backend == 'amd' && 'AMD=1\nMOCKGPU=1\nFORWARD_ONLY=1' || matrix.backend == 'nv' && 'NV=1\nMOCKGPU=1\nFORWARD_ONLY=1' }}" >> $GITHUB_ENV
|
||||
- name: Install OpenCL
|
||||
if: matrix.backend == 'gpu'
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
echo "deb [ allow-insecure=yes ] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
|
||||
sudo apt update || true
|
||||
sudo apt install --allow-unauthenticated -y --no-install-recommends opencl-headers \
|
||||
intel-oneapi-runtime-openmp=2023.2.1-16 intel-oneapi-runtime-compilers-common=2023.2.1-16 intel-oneapi-runtime-compilers=2023.2.1-16 \
|
||||
intel-oneapi-runtime-dpcpp-sycl-opencl-cpu=2023.2.1-16 intel-oneapi-runtime-tbb-common=2021.10.0-49541 \
|
||||
intel-oneapi-runtime-tbb=2021.10.0-49541 intel-oneapi-runtime-opencl=2023.2.1-16
|
||||
- name: Install packages (cuda)
|
||||
if: matrix.backend == 'ptx' || matrix.backend == 'triton' || matrix.backend == 'nv'
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
sudo apt update -y || true
|
||||
sudo apt install -y --no-install-recommends git g++ cmake ninja-build llvm-15-dev zlib1g-dev libglew-dev \
|
||||
flex bison libfl-dev libboost-thread-dev libboost-filesystem-dev nvidia-cuda-toolkit-gcc libzstd-dev
|
||||
- name: Cache gpuocelot
|
||||
if: matrix.backend == 'ptx' || matrix.backend == 'triton' || matrix.backend == 'nv'
|
||||
id: cache-build
|
||||
uses: actions/cache@v4
|
||||
env:
|
||||
cache-name: cache-gpuocelot-build
|
||||
with:
|
||||
path: ${{ github.workspace }}/gpuocelot/ocelot
|
||||
key: ubuntu22.04-gpuocelot-4524e34adb7eaccc6f71262f2e21d7052bb17c2f-rebuild-9
|
||||
- name: Clone/compile gpuocelot
|
||||
if: (matrix.backend == 'ptx' || matrix.backend == 'triton' || matrix.backend == 'nv') && steps.cache-build.outputs.cache-hit != 'true'
|
||||
run: |
|
||||
git clone --recurse-submodules https://github.com/gpuocelot/gpuocelot.git ${{ github.workspace }}/gpuocelot
|
||||
cd ${{ github.workspace }}/gpuocelot/ocelot
|
||||
git checkout 4524e34adb7eaccc6f71262f2e21d7052bb17c2f
|
||||
mkdir build
|
||||
cd build
|
||||
cmake .. -Wno-dev -G Ninja -DOCELOT_BUILD_TOOLS=OFF -DCMAKE_BUILD_ALWAYS=0 -DBUILD_TESTS_CUDA=OFF
|
||||
ninja
|
||||
- name: Install gpuocelot
|
||||
if: matrix.backend == 'ptx' || matrix.backend == 'triton' || matrix.backend == 'nv'
|
||||
run: |
|
||||
cd ${{ github.workspace }}/gpuocelot/ocelot/build
|
||||
sudo cp libgpuocelot.so /usr/lib/libgpuocelot.so
|
||||
- name: Install packages (amd)
|
||||
if: matrix.backend == 'amd'
|
||||
run: |
|
||||
echo 'Acquire::http::Pipeline-Depth "5";' | sudo tee -a /etc/apt/apt.conf.d/99parallel
|
||||
wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
|
||||
sudo tee /etc/apt/sources.list.d/rocm.list <<'EOF'
|
||||
deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/6.1.2 jammy main
|
||||
EOF
|
||||
echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600
|
||||
sudo apt update || true
|
||||
sudo apt install --no-install-recommends --allow-unauthenticated -y hsa-rocr comgr hsa-rocr-dev liburing-dev libc6-dev
|
||||
curl -s https://api.github.com/repos/Qazalin/remu/releases/latest | \
|
||||
jq -r '.assets[] | select(.name == "libremu.so").browser_download_url' | \
|
||||
sudo xargs curl -L -o /usr/local/lib/libremu.so
|
||||
sudo tee --append /etc/ld.so.conf.d/rocm.conf <<'EOF'
|
||||
/opt/rocm/lib
|
||||
/opt/rocm/lib64
|
||||
EOF
|
||||
sudo ldconfig
|
||||
- name: Install dependencies
|
||||
run: pip install -e '.[testing${{matrix.backend=='llvm'&&',llvm'||matrix.backend=='ptx'&&',cuda'||matrix.backend=='triton'&&',triton'||''}}]' --extra-index-url https://download.pytorch.org/whl/cpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/Triton-Nightly/pypi/simple/
|
||||
- name: Check Device.DEFAULT and print some source
|
||||
run: |
|
||||
PYTHONPATH=${{ github.workspace }} python3 -c "from tinygrad import Device; assert Device.DEFAULT in ['LLVM','CLANG','CUDA','GPU','AMD','NV'], Device.DEFAULT"
|
||||
DEBUG=5 PYTHONPATH=${{ github.workspace }} FORWARD_ONLY=1 python3 test/test_ops.py TestOps.test_add
|
||||
- name: Run pytest (not cuda or amd)
|
||||
if: matrix.backend!='ptx' && matrix.backend!='triton' && matrix.backend != 'amd' && matrix.backend != 'nv'
|
||||
run: python -m pytest -n=auto test/ --ignore=test/unit --durations=20
|
||||
- name: Run ONNX (only LLVM)
|
||||
if: matrix.backend == 'llvm'
|
||||
run: python -m pytest -n=auto test/external/external_test_onnx_backend.py --durations=20
|
||||
- name: Run pytest (cuda)
|
||||
if: matrix.backend=='ptx'||matrix.backend=='triton'||matrix.backend=='nv'
|
||||
run: python -m pytest -n=auto test/ -k 'not (half or test_efficientnet_safetensors)' --ignore=test/external --ignore=test/models --ignore=test/unit --ignore test/test_gc.py --durations=20
|
||||
- name: Run pytest (amd)
|
||||
if: matrix.backend=='amd'
|
||||
run: python -m pytest -n=auto test/test_ops.py test/test_dtype.py test/test_dtype_alu.py test/test_linearizer.py test/test_randomness.py test/imported/test_indexing.py test/test_hcq.py --durations=20
|
||||
- name: Run TRANSCENDENTAL math
|
||||
run: TRANSCENDENTAL=2 python -m pytest -n=auto test/test_ops.py::TestOps::test_sin test/test_ops.py::TestOps::test_cos test/test_ops.py::TestOps::test_tan test/test_ops.py::TestOps::test_exp test/test_ops.py::TestOps::test_log --durations=20
|
||||
- name: Run process replay tests
|
||||
run: |
|
||||
export PR_TITLE=$(jq -r .pull_request.title "$GITHUB_EVENT_PATH")
|
||||
export COMMIT_MESSAGE=$(git show -s --format=%B ${{ github.event.pull_request.head.sha }})
|
||||
cp test/external/process_replay/process_replay.py ./process_replay.py && git fetch origin master && git -c advice.detachedHead=false checkout origin/master && PYTHONPATH=. python3 process_replay.py
|
||||
|
||||
#testunicorn:
|
||||
# name: ARM64 unicorn Test
|
||||
# runs-on: ubuntu-latest
|
||||
# timeout-minutes: 20
|
||||
# steps:
|
||||
# - name: Checkout Code
|
||||
# uses: actions/checkout@v4
|
||||
# - name: Set up Python 3.11
|
||||
# uses: actions/setup-python@v5
|
||||
# with:
|
||||
# python-version: 3.11
|
||||
# - name: Cache python packages
|
||||
# uses: actions/cache@v4
|
||||
# with:
|
||||
# path: ${{ env.Python3_ROOT_DIR }}/lib/python3.11/site-packages
|
||||
# key: testing-arm-packages-${{ hashFiles('**/setup.py') }}
|
||||
# - name: Install cross-assembler
|
||||
# run: |
|
||||
# sudo apt update -y
|
||||
# sudo apt install -y --no-install-recommends gcc-aarch64-linux-gnu
|
||||
# - name: Install dependencies
|
||||
# run: pip install -e '.[testing,arm]' --extra-index-url https://download.pytorch.org/whl/cpu
|
||||
# - name: Test arm
|
||||
# run: CI=1 ARM64=1 CLANG=1 python -m pytest -n=auto test/ -k 'not (test_nn.py and (test_conv_transpose2d or test_conv2d))' --ignore=test/models --ignore=test/test_speed_v_torch.py --ignore=test/test_net_speed.py --ignore=test/test_specific_conv.py --ignore=test/unit/test_disk_tensor.py
|
||||
-59
@@ -1,59 +0,0 @@
|
||||
__pycache__
|
||||
.venv/
|
||||
.venv-*/
|
||||
.vscode
|
||||
.DS_Store
|
||||
notebooks
|
||||
.*.swp
|
||||
.*.swo
|
||||
*.pyc
|
||||
*.so
|
||||
*.txt
|
||||
build
|
||||
/dist
|
||||
*.egg-info
|
||||
/env
|
||||
a.out
|
||||
boxes.jpg
|
||||
pandecode.dump
|
||||
vertex.bin
|
||||
recognize*
|
||||
.idea
|
||||
*.prof
|
||||
extra/disassemblers/applegpu
|
||||
extra/datasets/cifar-10-python.tar.gz
|
||||
extra/datasets/librispeech/
|
||||
extra/datasets/imagenet/
|
||||
extra/datasets/wiki/
|
||||
extra/datasets/kits19
|
||||
extra/datasets/kits19/
|
||||
extra/datasets/squad/
|
||||
extra/datasets/img_align_celeba*
|
||||
extra/datasets/open-images-v6-mlperf
|
||||
extra/datasets/kits/
|
||||
extra/datasets/COCO/
|
||||
extra/datasets/audio*
|
||||
extra/weights
|
||||
venv
|
||||
examples/**/**/net*.*[js,json]
|
||||
examples/**/*.safetensors
|
||||
examples/webgpu/stable_diffusion/*.*[js,mjs]
|
||||
node_modules
|
||||
package.json
|
||||
package-lock.json
|
||||
temp
|
||||
*.csv
|
||||
.coverage
|
||||
coverage.xml
|
||||
htmlcov
|
||||
outputs_yolov8
|
||||
wandb
|
||||
model.safetensors
|
||||
quickstart.py
|
||||
.hypothesis
|
||||
weights
|
||||
*.lprof
|
||||
comgr_*
|
||||
*.pkl
|
||||
site/
|
||||
master_schedule.py
|
||||
@@ -1,46 +0,0 @@
|
||||
# on Windows -- $env:SKIP="tests,example"
|
||||
repos:
|
||||
- repo: local
|
||||
hooks:
|
||||
- id: ruff
|
||||
name: ruff
|
||||
entry: python3 -m ruff check .
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
- id: tiny
|
||||
name: tiny tests
|
||||
entry: python3 -m pytest test/test_tiny.py
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
- id: mypy
|
||||
name: mypy
|
||||
entry: python3 -m mypy tinygrad/ --strict-equality
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
- id: devicetests
|
||||
name: select GPU tests
|
||||
entry: env GPU=1 PYTHONPATH="." python3 -m pytest test/test_uops.py test/test_search.py
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
- id: tests
|
||||
name: subset of tests
|
||||
entry: env PYTHONPATH="." python3 -m pytest -n=4 test/unit/ test/test_ops.py test/test_dtype.py test/test_schedule.py test/test_assign.py test/test_symbolic_shapetracker.py
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
- id: example
|
||||
name: multi device tests
|
||||
entry: python3 test/external/external_test_example.py
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
- id: pylint
|
||||
name: pylint
|
||||
entry: python3 -m pylint tinygrad/
|
||||
language: system
|
||||
always_run: true
|
||||
pass_filenames: false
|
||||
@@ -1,472 +0,0 @@
|
||||
[MASTER]
|
||||
|
||||
# A comma-separated list of package or module names from where C extensions may
|
||||
# be loaded. Extensions are loading into the active Python interpreter and may
|
||||
# run arbitrary code
|
||||
extension-pkg-whitelist=scipy,cereal.messaging.messaging_pyx,PyQt5,av
|
||||
|
||||
# Add files or directories to the blacklist. They should be base names, not
|
||||
# paths.
|
||||
ignore=CVS,autogen,msm_kgsl.py
|
||||
|
||||
# Add files or directories matching the regex patterns to the blacklist. The
|
||||
# regex matches against base names, not paths.
|
||||
ignore-patterns=
|
||||
|
||||
# Python code to execute, usually for sys.path manipulation such as
|
||||
# pygtk.require().
|
||||
#init-hook=
|
||||
|
||||
# Use multiple processes to speed up Pylint.
|
||||
jobs=4
|
||||
|
||||
# List of plugins (as comma separated values of python modules names) to load,
|
||||
# usually to register additional checkers.
|
||||
load-plugins=
|
||||
|
||||
# Pickle collected data for later comparisons.
|
||||
persistent=yes
|
||||
|
||||
# Specify a configuration file.
|
||||
#rcfile=
|
||||
|
||||
# When enabled, pylint would attempt to guess common misconfiguration and emit
|
||||
# user-friendly hints instead of false-positive error messages
|
||||
suggestion-mode=yes
|
||||
|
||||
# Allow loading of arbitrary C extensions. Extensions are imported into the
|
||||
# active Python interpreter and may run arbitrary code.
|
||||
unsafe-load-any-extension=no
|
||||
|
||||
|
||||
[MESSAGES CONTROL]
|
||||
|
||||
# Only show warnings with the listed confidence levels. Leave empty to show
|
||||
# all. Valid levels: HIGH, INFERENCE, INFERENCE_FAILURE, UNDEFINED
|
||||
confidence=
|
||||
|
||||
# Disable the message, report, category or checker with the given id(s). You
|
||||
# can either give multiple identifiers separated by comma (,) or put this
|
||||
# option multiple times (only on the command line, not in the configuration
|
||||
# file where it should appear only once).You can also use "--disable=all" to
|
||||
# disable everything first and then reenable specific checks. For example, if
|
||||
# you want to run only the similarities checker, you can use "--disable=all
|
||||
# --enable=similarities". If you want to run only the classes checker, but have
|
||||
# no Warning level messages displayed, use"--disable=all --enable=classes
|
||||
# --disable=W"
|
||||
disable=C,R,W0613,W0511,W0212,W0201,W0106,W0603,W0621,W0703,W1201,W1203,E1136,W1514,E1101,W0221,W0105,E0401,abstract-method
|
||||
# E1101 for function binding
|
||||
# W0221 for Function class
|
||||
# W0105 for comment strings
|
||||
# E0401 for missing imports
|
||||
|
||||
# Enable the message, report, category or checker with the given id(s). You can
|
||||
# either give multiple identifier separated by comma (,) or put this option
|
||||
# multiple time (only on the command line, not in the configuration file where
|
||||
# it should appear only once). See also the "--disable" option for examples.
|
||||
enable=c-extension-no-member,use-a-generator, no-else-return
|
||||
|
||||
|
||||
[REPORTS]
|
||||
|
||||
# Python expression which should return a note less than 10 (10 is the highest
|
||||
# note). You have access to the variables errors warning, statement which
|
||||
# respectively contain the number of errors / warnings messages and the total
|
||||
# number of statements analyzed. This is used by the global evaluation report
|
||||
# (RP0004).
|
||||
evaluation=10.0 - ((float(5 * error + warning + refactor + convention) / statement) * 10)
|
||||
|
||||
# Template used to display messages. This is a python new-style format string
|
||||
# used to format the message information. See doc for all details
|
||||
#msg-template=
|
||||
|
||||
# Set the output format. Available formats are text, parseable, colorized, json
|
||||
# and msvs (visual studio).You can also give a reporter class, eg
|
||||
# mypackage.mymodule.MyReporterClass.
|
||||
output-format=text
|
||||
|
||||
# Tells whether to display a full report or only the messages
|
||||
reports=no
|
||||
|
||||
# Activate the evaluation score.
|
||||
score=yes
|
||||
|
||||
|
||||
[REFACTORING]
|
||||
|
||||
# Maximum number of nested blocks for function / method body
|
||||
max-nested-blocks=5
|
||||
|
||||
# Complete name of functions that never returns. When checking for
|
||||
# inconsistent-return-statements if a never returning function is called then
|
||||
# it will be considered as an explicit return statement and no message will be
|
||||
# printed.
|
||||
never-returning-functions=optparse.Values,sys.exit
|
||||
|
||||
|
||||
[LOGGING]
|
||||
|
||||
# Logging modules to check that the string format arguments are in logging
|
||||
# function parameter format
|
||||
logging-modules=logging
|
||||
|
||||
|
||||
[SPELLING]
|
||||
|
||||
# Limits count of emitted suggestions for spelling mistakes
|
||||
max-spelling-suggestions=4
|
||||
|
||||
# Spelling dictionary name. Available dictionaries: none. To make it working
|
||||
# install python-enchant package.
|
||||
spelling-dict=
|
||||
|
||||
# List of comma separated words that should not be checked.
|
||||
spelling-ignore-words=
|
||||
|
||||
# A path to a file that contains private dictionary; one word per line.
|
||||
spelling-private-dict-file=
|
||||
|
||||
# Tells whether to store unknown words to indicated private dictionary in
|
||||
# --spelling-private-dict-file option instead of raising a message.
|
||||
spelling-store-unknown-words=no
|
||||
|
||||
|
||||
[MISCELLANEOUS]
|
||||
|
||||
# List of note tags to take in consideration, separated by a comma.
|
||||
notes=FIXME,
|
||||
XXX,
|
||||
TODO
|
||||
|
||||
|
||||
[SIMILARITIES]
|
||||
|
||||
# Ignore comments when computing similarities.
|
||||
ignore-comments=yes
|
||||
|
||||
# Ignore docstrings when computing similarities.
|
||||
ignore-docstrings=yes
|
||||
|
||||
# Ignore imports when computing similarities.
|
||||
ignore-imports=no
|
||||
|
||||
# Minimum lines number of a similarity.
|
||||
min-similarity-lines=4
|
||||
|
||||
|
||||
[TYPECHECK]
|
||||
|
||||
# List of decorators that produce context managers, such as
|
||||
# contextlib.contextmanager. Add to this list to register other decorators that
|
||||
# produce valid context managers.
|
||||
contextmanager-decorators=contextlib.contextmanager
|
||||
|
||||
# List of members which are set dynamically and missed by pylint inference
|
||||
# system, and so shouldn't trigger E1101 when accessed. Python regular
|
||||
# expressions are accepted.
|
||||
generated-members=capnp.* cereal.* pygame.* zmq.* setproctitle.* smbus2.* usb1.* serial.* cv2.* ft4222.* carla.*
|
||||
|
||||
# Tells whether missing members accessed in mixin class should be ignored. A
|
||||
# mixin class is detected if its name ends with "mixin" (case insensitive).
|
||||
ignore-mixin-members=yes
|
||||
|
||||
# This flag controls whether pylint should warn about no-member and similar
|
||||
# checks whenever an opaque object is returned when inferring. The inference
|
||||
# can return multiple potential results while evaluating a Python object, but
|
||||
# some branches might not be evaluated, which results in partial inference. In
|
||||
# that case, it might be useful to still emit no-member and other checks for
|
||||
# the rest of the inferred objects.
|
||||
ignore-on-opaque-inference=yes
|
||||
|
||||
# List of class names for which member attributes should not be checked (useful
|
||||
# for classes with dynamically set attributes). This supports the use of
|
||||
# qualified names.
|
||||
ignored-classes=optparse.Values,thread._local,_thread._local
|
||||
|
||||
# List of module names for which member attributes should not be checked
|
||||
# (useful for modules/projects where namespaces are manipulated during runtime
|
||||
# and thus existing member attributes cannot be deduced by static analysis. It
|
||||
# supports qualified module names, as well as Unix pattern matching.
|
||||
ignored-modules=flask setproctitle usb1 flask.ext.socketio smbus2 usb1.*
|
||||
|
||||
# Show a hint with possible names when a member name was not found. The aspect
|
||||
# of finding the hint is based on edit distance.
|
||||
missing-member-hint=yes
|
||||
|
||||
# The minimum edit distance a name should have in order to be considered a
|
||||
# similar match for a missing member name.
|
||||
missing-member-hint-distance=1
|
||||
|
||||
# The total number of similar names that should be taken in consideration when
|
||||
# showing a hint for a missing member.
|
||||
missing-member-max-choices=1
|
||||
|
||||
|
||||
[VARIABLES]
|
||||
|
||||
# List of additional names supposed to be defined in builtins. Remember that
|
||||
# you should avoid to define new builtins when possible.
|
||||
additional-builtins=
|
||||
|
||||
# Tells whether unused global variables should be treated as a violation.
|
||||
allow-global-unused-variables=yes
|
||||
|
||||
# List of strings which can identify a callback function by name. A callback
|
||||
# name must start or end with one of those strings.
|
||||
callbacks=cb_,
|
||||
_cb
|
||||
|
||||
# A regular expression matching the name of dummy variables (i.e. expectedly
|
||||
# not used).
|
||||
dummy-variables-rgx=_+$|(_[a-zA-Z0-9_]*[a-zA-Z0-9]+?$)|dummy|^ignored_|^unused_
|
||||
|
||||
# Argument names that match this expression will be ignored. Default to name
|
||||
# with leading underscore
|
||||
ignored-argument-names=_.*|^ignored_|^unused_
|
||||
|
||||
# Tells whether we should check for unused import in __init__ files.
|
||||
init-import=no
|
||||
|
||||
# List of qualified module names which can have objects that can redefine
|
||||
# builtins.
|
||||
redefining-builtins-modules=six.moves,past.builtins,future.builtins
|
||||
|
||||
|
||||
[FORMAT]
|
||||
|
||||
# Expected format of line ending, e.g. empty (any line ending), LF or CRLF.
|
||||
expected-line-ending-format=
|
||||
|
||||
# Regexp for a line that is allowed to be longer than the limit.
|
||||
ignore-long-lines=^\s*(# )?<?https?://\S+>?$
|
||||
|
||||
# Number of spaces of indent required inside a hanging or continued line.
|
||||
indent-after-paren=4
|
||||
|
||||
# String used as indentation unit. This is usually " " (4 spaces) or "\t" (1
|
||||
# tab).
|
||||
indent-string=' '
|
||||
|
||||
# Maximum number of characters on a single line.
|
||||
max-line-length=150
|
||||
|
||||
# Maximum number of lines in a module
|
||||
max-module-lines=1000
|
||||
|
||||
# Allow the body of a class to be on the same line as the declaration if body
|
||||
# contains single statement.
|
||||
single-line-class-stmt=no
|
||||
|
||||
# Allow the body of an if to be on the same line as the test if there is no
|
||||
# else.
|
||||
single-line-if-stmt=no
|
||||
|
||||
|
||||
[BASIC]
|
||||
|
||||
# Naming style matching correct argument names
|
||||
argument-naming-style=snake_case
|
||||
|
||||
# Regular expression matching correct argument names. Overrides argument-
|
||||
# naming-style
|
||||
#argument-rgx=
|
||||
|
||||
# Naming style matching correct attribute names
|
||||
attr-naming-style=snake_case
|
||||
|
||||
# Regular expression matching correct attribute names. Overrides attr-naming-
|
||||
# style
|
||||
#attr-rgx=
|
||||
|
||||
# Bad variable names which should always be refused, separated by a comma
|
||||
bad-names=foo,
|
||||
bar,
|
||||
baz,
|
||||
toto,
|
||||
tutu,
|
||||
tata
|
||||
|
||||
# Naming style matching correct class attribute names
|
||||
class-attribute-naming-style=any
|
||||
|
||||
# Regular expression matching correct class attribute names. Overrides class-
|
||||
# attribute-naming-style
|
||||
#class-attribute-rgx=
|
||||
|
||||
# Naming style matching correct class names
|
||||
class-naming-style=PascalCase
|
||||
|
||||
# Regular expression matching correct class names. Overrides class-naming-style
|
||||
#class-rgx=
|
||||
|
||||
# Naming style matching correct constant names
|
||||
const-naming-style=UPPER_CASE
|
||||
|
||||
# Regular expression matching correct constant names. Overrides const-naming-
|
||||
# style
|
||||
#const-rgx=
|
||||
|
||||
# Minimum line length for functions/classes that require docstrings, shorter
|
||||
# ones are exempt.
|
||||
docstring-min-length=-1
|
||||
|
||||
# Naming style matching correct function names
|
||||
function-naming-style=snake_case
|
||||
|
||||
# Regular expression matching correct function names. Overrides function-
|
||||
# naming-style
|
||||
#function-rgx=
|
||||
|
||||
# Good variable names which should always be accepted, separated by a comma
|
||||
good-names=i,
|
||||
j,
|
||||
k,
|
||||
ex,
|
||||
Run,
|
||||
_
|
||||
|
||||
# Include a hint for the correct naming format with invalid-name
|
||||
include-naming-hint=no
|
||||
|
||||
# Naming style matching correct inline iteration names
|
||||
inlinevar-naming-style=any
|
||||
|
||||
# Regular expression matching correct inline iteration names. Overrides
|
||||
# inlinevar-naming-style
|
||||
#inlinevar-rgx=
|
||||
|
||||
# Naming style matching correct method names
|
||||
method-naming-style=snake_case
|
||||
|
||||
# Regular expression matching correct method names. Overrides method-naming-
|
||||
# style
|
||||
#method-rgx=
|
||||
|
||||
# Naming style matching correct module names
|
||||
module-naming-style=snake_case
|
||||
|
||||
# Regular expression matching correct module names. Overrides module-naming-
|
||||
# style
|
||||
#module-rgx=
|
||||
|
||||
# Colon-delimited sets of names that determine each other's naming style when
|
||||
# the name regexes allow several styles.
|
||||
name-group=
|
||||
|
||||
# Regular expression which should only match function or class names that do
|
||||
# not require a docstring.
|
||||
no-docstring-rgx=^_
|
||||
|
||||
# List of decorators that produce properties, such as abc.abstractproperty. Add
|
||||
# to this list to register other decorators that produce valid properties.
|
||||
property-classes=abc.abstractproperty
|
||||
|
||||
# Naming style matching correct variable names
|
||||
variable-naming-style=snake_case
|
||||
|
||||
# Regular expression matching correct variable names. Overrides variable-
|
||||
# naming-style
|
||||
#variable-rgx=
|
||||
|
||||
|
||||
[DESIGN]
|
||||
|
||||
# Maximum number of arguments for function / method
|
||||
max-args=5
|
||||
|
||||
# Maximum number of attributes for a class (see R0902).
|
||||
max-attributes=7
|
||||
|
||||
# Maximum number of boolean expressions in a if statement
|
||||
max-bool-expr=5
|
||||
|
||||
# Maximum number of branch for function / method body
|
||||
max-branches=12
|
||||
|
||||
# Maximum number of locals for function / method body
|
||||
max-locals=15
|
||||
|
||||
# Maximum number of parents for a class (see R0901).
|
||||
max-parents=7
|
||||
|
||||
# Maximum number of public methods for a class (see R0904).
|
||||
max-public-methods=20
|
||||
|
||||
# Maximum number of return / yield for function / method body
|
||||
max-returns=6
|
||||
|
||||
# Maximum number of statements in function / method body
|
||||
max-statements=50
|
||||
|
||||
# Minimum number of public methods for a class (see R0903).
|
||||
min-public-methods=2
|
||||
|
||||
|
||||
[CLASSES]
|
||||
|
||||
# List of method names used to declare (i.e. assign) instance attributes.
|
||||
defining-attr-methods=__init__,
|
||||
__new__,
|
||||
setUp
|
||||
|
||||
# List of member names, which should be excluded from the protected access
|
||||
# warning.
|
||||
exclude-protected=_asdict,
|
||||
_fields,
|
||||
_replace,
|
||||
_source,
|
||||
_make
|
||||
|
||||
# List of valid names for the first argument in a class method.
|
||||
valid-classmethod-first-arg=cls
|
||||
|
||||
# List of valid names for the first argument in a metaclass class method.
|
||||
valid-metaclass-classmethod-first-arg=mcs
|
||||
|
||||
|
||||
[IMPORTS]
|
||||
|
||||
# Allow wildcard imports from modules that define __all__.
|
||||
allow-wildcard-with-all=no
|
||||
|
||||
# Analyse import fallback blocks. This can be used to support both Python 2 and
|
||||
# 3 compatible code, which means that the block might have code that exists
|
||||
# only in one or another interpreter, leading to false positives when analysed.
|
||||
analyse-fallback-blocks=no
|
||||
|
||||
# Deprecated modules which should not be used, separated by a comma
|
||||
deprecated-modules=regsub,
|
||||
TERMIOS,
|
||||
Bastion,
|
||||
rexec
|
||||
|
||||
# Create a graph of external dependencies in the given file (report RP0402 must
|
||||
# not be disabled)
|
||||
ext-import-graph=
|
||||
|
||||
# Create a graph of every (i.e. internal and external) dependencies in the
|
||||
# given file (report RP0402 must not be disabled)
|
||||
import-graph=
|
||||
|
||||
# Create a graph of internal dependencies in the given file (report RP0402 must
|
||||
# not be disabled)
|
||||
int-import-graph=
|
||||
|
||||
# Force import order to recognize a module as part of the standard
|
||||
# compatibility libraries.
|
||||
known-standard-library=
|
||||
|
||||
# Force import order to recognize a module as part of a third party library.
|
||||
known-third-party=enchant
|
||||
|
||||
[STRING]
|
||||
|
||||
# This flag controls whether the implicit-str-concat should generate a warning
|
||||
# on implicit string concatenation in sequences defined over several lines.
|
||||
check-str-concat-over-line-jumps=yes
|
||||
|
||||
[EXCEPTIONS]
|
||||
|
||||
# Exceptions that will emit a warning when being caught. Defaults to
|
||||
# "Exception"
|
||||
overgeneral-exceptions=builtins.Exception
|
||||
@@ -1,7 +0,0 @@
|
||||
Copyright (c) 2024, the tiny corp
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
||||
@@ -1,182 +0,0 @@
|
||||
<div align="center">
|
||||
|
||||
<picture>
|
||||
<source media="(prefers-color-scheme: light)" srcset="/docs/logo_tiny_light.svg">
|
||||
<img alt="tiny corp logo" src="/docs/logo_tiny_dark.svg" width="50%" height="50%">
|
||||
</picture>
|
||||
|
||||
tinygrad: For something between [PyTorch](https://github.com/pytorch/pytorch) and [karpathy/micrograd](https://github.com/karpathy/micrograd). Maintained by [tiny corp](https://tinygrad.org).
|
||||
|
||||
<h3>
|
||||
|
||||
[Homepage](https://github.com/tinygrad/tinygrad) | [Documentation](https://docs.tinygrad.org/) | [Discord](https://discord.gg/ZjZadyC7PK)
|
||||
|
||||
</h3>
|
||||
|
||||
[](https://github.com/tinygrad/tinygrad/stargazers)
|
||||
[](https://github.com/tinygrad/tinygrad/actions/workflows/test.yml)
|
||||
[](https://discord.gg/ZjZadyC7PK)
|
||||
|
||||
</div>
|
||||
|
||||
---
|
||||
|
||||
This may not be the best deep learning framework, but it is a deep learning framework.
|
||||
|
||||
Due to its extreme simplicity, it aims to be the easiest framework to add new accelerators to, with support for both inference and training. If XLA is CISC, tinygrad is RISC.
|
||||
|
||||
tinygrad is still alpha software, but we [raised some money](https://geohot.github.io/blog/jekyll/update/2023/05/24/the-tiny-corp-raised-5M.html) to make it good. Someday, we will tape out chips.
|
||||
|
||||
## Features
|
||||
|
||||
### LLaMA and Stable Diffusion
|
||||
|
||||
tinygrad can run [LLaMA](/docs/showcase.md#llama) and [Stable Diffusion](/docs/showcase.md#stable-diffusion)!
|
||||
|
||||
### Laziness
|
||||
|
||||
Try a matmul. See how, despite the style, it is fused into one kernel with the power of laziness.
|
||||
|
||||
```sh
|
||||
DEBUG=3 python3 -c "from tinygrad import Tensor;
|
||||
N = 1024; a, b = Tensor.rand(N, N), Tensor.rand(N, N);
|
||||
c = (a.reshape(N, 1, N) * b.T.reshape(1, N, N)).sum(axis=2);
|
||||
print((c.numpy() - (a.numpy() @ b.numpy())).mean())"
|
||||
```
|
||||
|
||||
And we can change `DEBUG` to `4` to see the generated code.
|
||||
|
||||
### Neural networks
|
||||
|
||||
As it turns out, 90% of what you need for neural networks are a decent autograd/tensor library.
|
||||
Throw in an optimizer, a data loader, and some compute, and you have all you need.
|
||||
|
||||
```python
|
||||
from tinygrad import Tensor, nn
|
||||
|
||||
class LinearNet:
|
||||
def __init__(self):
|
||||
self.l1 = Tensor.kaiming_uniform(784, 128)
|
||||
self.l2 = Tensor.kaiming_uniform(128, 10)
|
||||
def __call__(self, x:Tensor) -> Tensor:
|
||||
return x.flatten(1).dot(self.l1).relu().dot(self.l2)
|
||||
|
||||
model = LinearNet()
|
||||
optim = nn.optim.Adam([model.l1, model.l2], lr=0.001)
|
||||
|
||||
x, y = Tensor.rand(4, 1, 28, 28), Tensor([2,4,3,7]) # replace with real mnist dataloader
|
||||
|
||||
with Tensor.train():
|
||||
for i in range(10):
|
||||
optim.zero_grad()
|
||||
loss = model(x).sparse_categorical_crossentropy(y).backward()
|
||||
optim.step()
|
||||
print(i, loss.item())
|
||||
```
|
||||
|
||||
See [examples/beautiful_mnist.py](examples/beautiful_mnist.py) for the full version that gets 98% in ~5 seconds
|
||||
|
||||
## Accelerators
|
||||
|
||||
tinygrad already supports numerous accelerators, including:
|
||||
|
||||
- [x] [GPU (OpenCL)](tinygrad/runtime/ops_gpu.py)
|
||||
- [x] [CLANG (C Code)](tinygrad/runtime/ops_clang.py)
|
||||
- [x] [LLVM](tinygrad/runtime/ops_llvm.py)
|
||||
- [x] [METAL](tinygrad/runtime/ops_metal.py)
|
||||
- [x] [CUDA](tinygrad/runtime/ops_cuda.py)
|
||||
- [x] [AMD](tinygrad/runtime/ops_amd.py)
|
||||
- [x] [NV](tinygrad/runtime/ops_nv.py)
|
||||
- [x] [QCOM](tinygrad/runtime/ops_qcom.py)
|
||||
- [x] [WEBGPU](tinygrad/runtime/ops_webgpu.py)
|
||||
|
||||
And it is easy to add more! Your accelerator of choice only needs to support a total of ~25 low level ops.
|
||||
|
||||
To check default accelerator run: `python3 -c "from tinygrad import Device; print(Device.DEFAULT)"`
|
||||
|
||||
## Installation
|
||||
|
||||
The current recommended way to install tinygrad is from source.
|
||||
|
||||
### From source
|
||||
|
||||
```sh
|
||||
git clone https://github.com/tinygrad/tinygrad.git
|
||||
cd tinygrad
|
||||
python3 -m pip install -e .
|
||||
```
|
||||
|
||||
### Direct (master)
|
||||
|
||||
```sh
|
||||
python3 -m pip install git+https://github.com/tinygrad/tinygrad.git
|
||||
```
|
||||
|
||||
## Documentation
|
||||
|
||||
Documentation along with a quick start guide can be found on the [docs website](https://docs.tinygrad.org/) built from the [docs/](/docs) directory.
|
||||
|
||||
### Quick example comparing to PyTorch
|
||||
|
||||
```python
|
||||
from tinygrad import Tensor
|
||||
|
||||
x = Tensor.eye(3, requires_grad=True)
|
||||
y = Tensor([[2.0,0,-2.0]], requires_grad=True)
|
||||
z = y.matmul(x).sum()
|
||||
z.backward()
|
||||
|
||||
print(x.grad.tolist()) # dz/dx
|
||||
print(y.grad.tolist()) # dz/dy
|
||||
```
|
||||
|
||||
The same thing but in PyTorch:
|
||||
```python
|
||||
import torch
|
||||
|
||||
x = torch.eye(3, requires_grad=True)
|
||||
y = torch.tensor([[2.0,0,-2.0]], requires_grad=True)
|
||||
z = y.matmul(x).sum()
|
||||
z.backward()
|
||||
|
||||
print(x.grad.tolist()) # dz/dx
|
||||
print(y.grad.tolist()) # dz/dy
|
||||
```
|
||||
|
||||
## Contributing
|
||||
|
||||
There has been a lot of interest in tinygrad lately. Following these guidelines will help your PR get accepted.
|
||||
|
||||
We'll start with what will get your PR closed with a pointer to this section:
|
||||
|
||||
- No code golf! While low line count is a guiding light of this project, anything that remotely looks like code golf will be closed. The true goal is reducing complexity and increasing readability, and deleting `\n`s does nothing to help with that.
|
||||
- All docs and whitespace changes will be closed unless you are a well-known contributor. The people writing the docs should be those who know the codebase the absolute best. People who have not demonstrated that shouldn't be messing with docs. Whitespace changes are both useless *and* carry a risk of introducing bugs.
|
||||
- Anything you claim is a "speedup" must be benchmarked. In general, the goal is simplicity, so even if your PR makes things marginally faster, you have to consider the tradeoff with maintainablity and readablity.
|
||||
- In general, the code outside the core `tinygrad/` folder is not well tested, so unless the current code there is broken, you shouldn't be changing it.
|
||||
- If your PR looks "complex", is a big diff, or adds lots of lines, it won't be reviewed or merged. Consider breaking it up into smaller PRs that are individually clear wins. A common pattern I see is prerequisite refactors before adding new functionality. If you can (cleanly) refactor to the point that the feature is a 3 line change, this is great, and something easy for us to review.
|
||||
|
||||
Now, what we want:
|
||||
|
||||
- Bug fixes (with a regression test) are great! This library isn't 1.0 yet, so if you stumble upon a bug, fix it, write a test, and submit a PR, this is valuable work.
|
||||
- Solving bounties! tinygrad [offers cash bounties](https://docs.google.com/spreadsheets/d/1WKHbT-7KOgjEawq5h5Ic1qUWzpfAzuD_J06N1JwOCGs/edit?usp=sharing) for certain improvements to the library. All new code should be high quality and well tested.
|
||||
- Features. However, if you are adding a feature, consider the line tradeoff. If it's 3 lines, there's less of a bar of usefulness it has to meet over something that's 30 or 300 lines. All features must have regression tests. In general with no other constraints, your feature's API should match torch or numpy.
|
||||
- Refactors that are clear wins. In general, if your refactor isn't a clear win it will be closed. But some refactors are amazing! Think about readability in a deep core sense. A whitespace change or moving a few functions around is useless, but if you realize that two 100 line functions can actually use the same 110 line function with arguments while also improving readability, this is a big win. Refactors should pass [process replay](#process-replay-tests).
|
||||
- Tests/fuzzers. If you can add tests that are non brittle, they are welcome. We have some fuzzers in here too, and there's a plethora of bugs that can be found with them and by improving them. Finding bugs, even writing broken tests (that should pass) with `@unittest.expectedFailure` is great. This is how we make progress.
|
||||
- Dead code removal from core `tinygrad/` folder. We don't care about the code in extra, but removing dead code from the core library is great. Less for new people to read and be confused by.
|
||||
|
||||
### Running tests
|
||||
|
||||
You should install the pre-commit hooks with `pre-commit install`. This will run the linter, mypy, and a subset of the tests on every commit.
|
||||
|
||||
For more examples on how to run the full test suite please refer to the [CI workflow](.github/workflows/test.yml).
|
||||
|
||||
Some examples of running tests locally:
|
||||
```sh
|
||||
python3 -m pip install -e '.[testing]' # install extra deps for testing
|
||||
python3 test/test_ops.py # just the ops tests
|
||||
python3 -m pytest test/ # whole test suite
|
||||
```
|
||||
|
||||
#### Process replay tests
|
||||
|
||||
[Process replay](https://github.com/tinygrad/tinygrad/blob/master/test/external/process_replay/README.md) compares your PR's generated kernels against master. If your PR is a refactor or speedup without any expected behavior change, It should include [pr] in the pull request title.
|
||||
@@ -0,0 +1,54 @@
|
||||
# abstractions2 goes from back to front, here we will go from front to back
|
||||
|
||||
# *****
|
||||
# 0. Load mnist on the device
|
||||
|
||||
from tinygrad.nn.datasets import mnist
|
||||
X_train, Y_train, _, _ = mnist()
|
||||
X_train = X_train.float()
|
||||
X_train -= X_train.mean()
|
||||
|
||||
# *****
|
||||
# 1. Define an MNIST model.
|
||||
|
||||
from tinygrad import Tensor, Context
|
||||
|
||||
l1 = Tensor.kaiming_uniform(128, 784)
|
||||
l2 = Tensor.kaiming_uniform(10, 128)
|
||||
def model(x): return x.flatten(1).dot(l1.T).relu().dot(l2.T)
|
||||
l1n, l2n = l1.numpy(), l2.numpy()
|
||||
|
||||
# *****
|
||||
# 2. Choose a batch for training and do the backward pass.
|
||||
|
||||
from tinygrad.nn.optim import SGD
|
||||
optim = SGD([l1, l2])
|
||||
|
||||
with Context(TRAINING=1):
|
||||
X, Y = X_train[(samples:=Tensor.randint(128, high=X_train.shape[0]))], Y_train[samples]
|
||||
optim.zero_grad()
|
||||
model(X).sparse_categorical_crossentropy(Y).backward()
|
||||
optim.schedule_step() # this will step the optimizer without running realize
|
||||
|
||||
# *****
|
||||
# 3. Create a schedule (linear uop).
|
||||
|
||||
# The weight Tensors have been assigned to, but not yet realized. Everything is still lazy at this point
|
||||
# l1.uop and l2.uop define a computation graph
|
||||
|
||||
from tinygrad.engine.realize import run_linear
|
||||
linear = Tensor.schedule_linear(l1, l2)
|
||||
|
||||
print(f"The schedule contains {len(linear.src)} items.")
|
||||
for call in linear.src: print(str(call)[:80])
|
||||
|
||||
# *****
|
||||
# 4. Lower and run the schedule (linear uop).
|
||||
|
||||
run_linear(linear)
|
||||
|
||||
# *****
|
||||
# 5. Print the weight change
|
||||
|
||||
print("first weight change\n", l1.numpy()-l1n)
|
||||
print("second weight change\n", l2.numpy()-l2n)
|
||||
@@ -0,0 +1,251 @@
|
||||
# tinygrad allows you to write kernels at many different abstractions levels.
|
||||
# This is for RDNA3, but if you don't have one you can run with the emulator
|
||||
# PYTHONPATH="." DEV=MOCKPCI+AMD
|
||||
|
||||
from tinygrad import Tensor, Context, GlobalCounters, UOp, Device
|
||||
from tinygrad.helpers import DEV, DEBUG, getenv
|
||||
from tinygrad.uop.ops import AxisType, KernelInfo, Ops
|
||||
from tinygrad.dtype import AddrSpace, dtypes
|
||||
from tinygrad.runtime.autogen.amd.rdna3.ins import *
|
||||
|
||||
def eval_harness(name, tensor, fxn, check=None):
|
||||
print(f"***** {name}")
|
||||
GlobalCounters.reset()
|
||||
with Context(DEBUG=max(DEBUG.value, 2)): out = fxn(tensor).item()
|
||||
assert check is None or abs(out - check) < abs(check) * 1e-3, f"out was wrong {out}, expected {check}, off by {out/check}x"
|
||||
print(f"computed in {GlobalCounters.time_sum_s*1000:.2f} ms, {(a.nbytes()/1e9)/GlobalCounters.time_sum_s:.2f} GB/s")
|
||||
return out
|
||||
|
||||
SZ = 256*1024 if DEV.interface.startswith("MOCK") else 1024*1024*1024
|
||||
|
||||
def example_2_hip(a:Tensor, correct):
|
||||
GLOBALS = 1024
|
||||
THREADS = 256
|
||||
def hip_reduce_sum(out:UOp, buf:UOp) -> UOp:
|
||||
assert SZ % (GLOBALS * THREADS) == 0
|
||||
CHUNK = SZ // (GLOBALS * THREADS)
|
||||
# NOTE: tinygrad doesn't populate HIP hidden kernargs, so blockDim.x/gridDim.x read as 0.
|
||||
# We hardcode block/grid sizes as constexpr to avoid any dependency on those builtins.
|
||||
code = f"""
|
||||
#include <hip/hip_runtime.h>
|
||||
constexpr unsigned int BLOCK = {THREADS};
|
||||
constexpr unsigned int CHUNK = {CHUNK};
|
||||
extern "C" __global__ void hip_reduce_sum_kernel(float* __restrict__ block_sums, const float* __restrict__ x) {{
|
||||
__shared__ float sdata[BLOCK];
|
||||
|
||||
unsigned int tid = threadIdx.x;
|
||||
unsigned int gid = blockIdx.x * BLOCK + tid;
|
||||
|
||||
// Each thread sums CHUNK consecutive elements from its own region
|
||||
float sum = 0.0f;
|
||||
const float* base = x + gid * CHUNK;
|
||||
#pragma unroll 16
|
||||
for (unsigned int k = 0; k < CHUNK; k++) {{
|
||||
sum += base[k];
|
||||
}}
|
||||
|
||||
sdata[tid] = sum;
|
||||
__syncthreads();
|
||||
|
||||
// Block reduction in shared memory
|
||||
for (unsigned int s = BLOCK / 2; s > 0; s >>= 1) {{
|
||||
if (tid < s) {{
|
||||
sdata[tid] += sdata[tid + s];
|
||||
}}
|
||||
__syncthreads();
|
||||
}}
|
||||
|
||||
// One partial sum per block
|
||||
if (tid == 0) {{
|
||||
block_sums[blockIdx.x] = sdata[0];
|
||||
}}
|
||||
}}"""
|
||||
|
||||
# TODO: remove the need for the compiler here, you should just be able to remove Ops.BINARY
|
||||
from tinygrad.runtime.support.compiler_amd import HIPCCCompiler
|
||||
lib = HIPCCCompiler(Device[Device.DEFAULT].renderer.target.arch, []).compile_cached(code)
|
||||
# the sink specifies the GLOBAL and LOCAL sizes, along with the input buffers and name
|
||||
sink = UOp.sink(UOp.special(GLOBALS, 'gidx0'), UOp.special(THREADS, 'lidx0'), out, buf,
|
||||
arg=KernelInfo(name="hip_reduce_sum_kernel"))
|
||||
return UOp(Ops.PROGRAM, src=(sink, UOp(Ops.LINEAR, src=(*sink.src, sink)), UOp(Ops.SOURCE, arg=code), UOp(Ops.BINARY, arg=lib)))
|
||||
eval_harness("HIP kernel", a, lambda x: Tensor.empty(GLOBALS).custom_kernel(x, fxn=hip_reduce_sum)[0].sum(), check=correct)
|
||||
|
||||
def example_3_custom_uop(a:Tensor, correct):
|
||||
# This GPU has 32 CUs, keep them all busy
|
||||
CU_COUNT = 32
|
||||
def custom_sum(out:UOp, buf:UOp) -> UOp:
|
||||
LCLS = 256
|
||||
buf = buf.reshape(CU_COUNT, -1, LCLS)
|
||||
|
||||
glbl = UOp.range(CU_COUNT, 0, AxisType.GLOBAL)
|
||||
lane = UOp.range(LCLS, 1, AxisType.LOCAL)
|
||||
|
||||
# accumulate the globals into a per lane accumulator
|
||||
reduce_loop = UOp.range(buf.shape[1], 2, AxisType.REDUCE)
|
||||
acc = UOp.placeholder((1,), dtypes.float, slot=6, addrspace=AddrSpace.REG)
|
||||
acc = acc.after(acc.store(0))
|
||||
acc = acc.after(acc[0].store(acc.after(reduce_loop)[0] + buf[glbl, reduce_loop, lane]).end(reduce_loop))
|
||||
|
||||
# store all the per lane accumulators to LOCAL
|
||||
local_accs = UOp.placeholder((LCLS,), dtypes.float, slot=0, addrspace=AddrSpace.LOCAL)
|
||||
local_accs = local_accs.after(local_accs[lane].store(acc[0]))
|
||||
|
||||
# accumulate LOCALs into a single per CU accumulator
|
||||
late_reduce_loop = UOp.range(LCLS, 3, AxisType.REDUCE)
|
||||
acc2 = UOp.placeholder((1,), dtypes.float, slot=7, addrspace=AddrSpace.REG)
|
||||
acc2 = acc2.after(acc2.store(0))
|
||||
acc2 = acc2.after(acc2[0].store(acc2.after(late_reduce_loop)[0] + local_accs[late_reduce_loop]).end(late_reduce_loop))[0]
|
||||
|
||||
# store (NOTE: since the address doesn't depend on the warp, this will be automatically gated)
|
||||
return out[glbl].store(acc2).end(lane, glbl).sink(arg=KernelInfo(opts_to_apply=()))
|
||||
|
||||
eval_harness("custom UOp kernel", a, lambda x: Tensor.empty(CU_COUNT).custom_kernel(x, fxn=custom_sum)[0].sum(), check=correct)
|
||||
|
||||
def example_5_custom_assembly(a:Tensor, correct):
|
||||
# Kernel class copied from amd_asm_matmul
|
||||
class Kernel:
|
||||
def __init__(self): self.instructions, self.labels, self.pos = [], {}, 0
|
||||
def label(self, name): self.labels[name] = self.pos
|
||||
def emit(self, inst, target=None):
|
||||
self.instructions.append(inst)
|
||||
inst._target, inst._pos = target, self.pos
|
||||
self.pos += inst.size()
|
||||
return inst
|
||||
def waitcnt(self, lgkm=None, vm=None):
|
||||
# Wait for memory operations. lgkm=N waits until N lgkm ops remain, vm=N waits until N vmem ops remain.
|
||||
vmcnt, lgkmcnt, expcnt = vm if vm is not None else 63, lgkm if lgkm is not None else 63, 7
|
||||
waitcnt = (expcnt & 0x7) | ((lgkmcnt & 0x3f) << 4) | ((vmcnt & 0x3f) << 10)
|
||||
self.emit(s_waitcnt(simm16=waitcnt))
|
||||
def finalize(self, sink:UOp) -> UOp:
|
||||
for inst in self.instructions:
|
||||
if inst._target is None: continue
|
||||
offset_dwords = (self.labels[inst._target] - inst._pos - inst.size()) // 4
|
||||
if not -32768 <= offset_dwords <= 32767: raise ValueError(f"branch to '{inst._target}' offset {offset_dwords} exceeds simm16 range")
|
||||
inst.simm16 = offset_dwords
|
||||
return UOp(Ops.PROGRAM, src=(sink, UOp(Ops.LINEAR, src=tuple([UOp(Ops.INS, arg=x) for x in self.instructions]))))
|
||||
|
||||
CU_COUNT = 32
|
||||
LANES = 64
|
||||
def asm_sum(out:UOp, buf:UOp) -> UOp:
|
||||
V_LANE_ID = 0 # lane_id set on startup
|
||||
S_WORKGROUP_X = 2 # workgroup_id_x
|
||||
S_LOOP_CTR = 3
|
||||
k = Kernel()
|
||||
# mul lane id by 16 for offsets (4 for float, 4 for b128)
|
||||
k.emit(v_mul_lo_u32(v[0], v[V_LANE_ID], 16))
|
||||
k.emit(v_add_nc_u32_e32(v[1], 4096, v[0]))
|
||||
k.emit(v_add_nc_u32_e32(v[2], 4096, v[1]))
|
||||
k.emit(v_add_nc_u32_e32(v[3], 4096, v[2]))
|
||||
# load both addresses
|
||||
k.emit(s_load_b128(sdata=s[4:7], sbase=s[0:1], offset=0x0, soffset=NULL))
|
||||
k.waitcnt(lgkm=0)
|
||||
# offset buffer pointer by workgroup_id_x * chunk_size_bytes
|
||||
k.emit(s_mul_i32(s[S_LOOP_CTR], s[S_WORKGROUP_X], buf.numel()*4//CU_COUNT))
|
||||
k.emit(s_add_u32(s[6], s[6], s[S_LOOP_CTR]))
|
||||
k.emit(s_addc_u32(s[7], s[7], 0))
|
||||
# zero the accumulators
|
||||
k.emit(VOPD(VOPDOp.V_DUAL_MOV_B32, VOPDOp.V_DUAL_MOV_B32, vdstx=v[4], vdsty=v[5], srcx0=0, srcy0=0))
|
||||
k.emit(VOPD(VOPDOp.V_DUAL_MOV_B32, VOPDOp.V_DUAL_MOV_B32, vdstx=v[6], vdsty=v[7], srcx0=0, srcy0=0))
|
||||
|
||||
def emit_loads(base_vreg, reg_len):
|
||||
assert reg_len%4 == 0
|
||||
k.emit(s_clause(simm16=(reg_len//4)-1))
|
||||
for i in range(reg_len//4):
|
||||
offset = i*LANES*16
|
||||
assert offset < 16384
|
||||
k.emit(global_load_b128(vdst=v[base_vreg+i*4:base_vreg+i*4+3], addr=v[offset//4096], saddr=s[6:7], offset=offset%4096))
|
||||
k.emit(s_add_u32(s[6], s[6], reg_len * LANES * 4))
|
||||
k.emit(s_addc_u32(s[7], s[7], 0))
|
||||
|
||||
def tree_reduce_to_4567(base_vreg, reg_len):
|
||||
assert reg_len%4 == 0
|
||||
reg_len //= 4
|
||||
while reg_len > 1:
|
||||
half = reg_len // 2
|
||||
for j in range(half):
|
||||
a, b = base_vreg + j*4, base_vreg + (j+half)*4
|
||||
# v[a+0](bank0) += v[b+2](bank2), v[a+1](bank1) += v[b+3](bank3) — src0 and src1 on different banks
|
||||
k.emit(VOPD(VOPDOp.V_DUAL_ADD_F32, VOPDOp.V_DUAL_ADD_F32, vdstx=v[a], vdsty=v[a+1], srcx0=v[a], vsrcx1=v[b+2], srcy0=v[a+1], vsrcy1=v[b+3]))
|
||||
# v[a+2](bank2) += v[b+0](bank0), v[a+3](bank3) += v[b+1](bank1) — src0 and src1 on different banks
|
||||
k.emit(VOPD(VOPDOp.V_DUAL_ADD_F32, VOPDOp.V_DUAL_ADD_F32, vdstx=v[a+2], vdsty=v[a+3], srcx0=v[a+2], vsrcx1=v[b], srcy0=v[a+3], vsrcy1=v[b+1]))
|
||||
reg_len = half
|
||||
k.emit(VOPD(VOPDOp.V_DUAL_ADD_F32, VOPDOp.V_DUAL_ADD_F32, vdstx=v[4], vdsty=v[5], srcx0=v[4], vsrcx1=v[base_vreg], srcy0=v[5], vsrcy1=v[base_vreg+1]))
|
||||
k.emit(VOPD(VOPDOp.V_DUAL_ADD_F32, VOPDOp.V_DUAL_ADD_F32, vdstx=v[6], vdsty=v[7], srcx0=v[6], vsrcx1=v[base_vreg+2], srcy0=v[7], vsrcy1=v[base_vreg+3]))
|
||||
|
||||
BASE_REG = 8
|
||||
LOAD_UNROLL = 64
|
||||
INNER_UNROLL = 2
|
||||
|
||||
assert buf.numel() % (CU_COUNT*LANES*LOAD_UNROLL*INNER_UNROLL) == 0
|
||||
total_batches = buf.numel()//(CU_COUNT*LANES*LOAD_UNROLL*INNER_UNROLL)
|
||||
k.emit(s_mov_b32(s[S_LOOP_CTR], total_batches-1))
|
||||
|
||||
k.label('LOOP')
|
||||
for _ in range(INNER_UNROLL):
|
||||
emit_loads(BASE_REG, reg_len=LOAD_UNROLL)
|
||||
k.waitcnt(vm=0)
|
||||
tree_reduce_to_4567(BASE_REG, reg_len=LOAD_UNROLL)
|
||||
k.emit(s_sub_u32(s[S_LOOP_CTR], s[S_LOOP_CTR], 1))
|
||||
k.emit(s_cbranch_scc0(), target='LOOP')
|
||||
|
||||
# add into v[4]
|
||||
k.emit(v_add_f32_e32(v[4], v[4], v[5]))
|
||||
k.emit(v_add_f32_e32(v[6], v[6], v[7]))
|
||||
k.emit(v_add_f32_e32(v[4], v[4], v[6]))
|
||||
|
||||
# warp shuffle into v[4] on lane 0 using DPP row_shl within each 16-lane row
|
||||
for shift in [1, 2, 4, 8]:
|
||||
k.emit(v_add_f32_e32(v[4], DPP, v[4], vsrc0=v[4], dpp=0x100 | shift, row_mask=0xf, bank_mask=0xf, bc=1))
|
||||
# combine rows: get lane 16's value to lane 0 via permlanex16
|
||||
k.emit(v_permlanex16_b32(v[5], v[4], 0, 0))
|
||||
k.emit(v_add_f32_e32(v[4], v[4], v[5]))
|
||||
|
||||
# atomic store (only on lane 0)
|
||||
k.emit(s_mov_b32(EXEC_LO, 1))
|
||||
k.emit(v_mov_b32_e32(v[0], 0))
|
||||
k.emit(global_atomic_add_f32(addr=v[0], saddr=s[4:5], data=v[4]))
|
||||
|
||||
k.emit(s_sendmsg(simm16=3)) # DEALLOC_VGPRS
|
||||
k.emit(s_endpgm())
|
||||
return k.finalize(UOp.sink(UOp.special(CU_COUNT, 'gidx0'), UOp.special(LANES, 'lidx0'), out, buf, arg=KernelInfo(name="asm_reduce")))
|
||||
|
||||
out = Tensor.zeros(1,).contiguous().realize()
|
||||
eval_harness("RDNA3 assembly kernel", a, lambda x: out.custom_kernel(x, fxn=asm_sum)[0], check=correct)
|
||||
|
||||
if __name__ == "__main__":
|
||||
examples = [int(x) for x in getenv("EXAMPLES", "1,2,3,4,5").split(",")]
|
||||
|
||||
correct = None
|
||||
# First define a Tensor and realize it. We will focus on a 1GB sum kernel on RDNA3
|
||||
a = (Tensor.randn(SZ) if getenv("RAND") else Tensor.ones(SZ)).contiguous().realize()
|
||||
|
||||
if 1 in examples:
|
||||
# *****
|
||||
# This is the high level tinygrad way.
|
||||
# Note that this is split into multiple kernels for speed.
|
||||
correct = eval_harness("basic kernel", a, lambda x: x.sum())
|
||||
|
||||
if 2 in examples:
|
||||
# *****
|
||||
# You can import kernels from CUDA/HIP/Metal.
|
||||
# ChatGPT is great at writing these Kernel
|
||||
example_2_hip(a, correct)
|
||||
|
||||
if 3 in examples:
|
||||
# *****
|
||||
# Now we get to the lower abstraction layers of tinygrad.
|
||||
# You can write a kernel in UOps, and it's 2.5x faster than normal.
|
||||
example_3_custom_uop(a, correct)
|
||||
|
||||
if 4 in examples:
|
||||
# *****
|
||||
# You can also BEAM search stock tinygrad for a faster kernel.
|
||||
# This does even better than all the kernels to date in this simple case.
|
||||
with Context(BEAM=2):
|
||||
eval_harness("BEAMed kernel", a, lambda x: x.sum(), check=correct)
|
||||
|
||||
if 5 in examples:
|
||||
# *****
|
||||
# If you really want to go crazy with speed, you can code in assembly.
|
||||
# There's not too much to gain here over BEAM, but it's a few percent faster.
|
||||
example_5_custom_assembly(a, correct)
|
||||
@@ -0,0 +1,355 @@
|
||||
/*
|
||||
Inspired by https://spec.draculatheme.com/ specification, they should work
|
||||
decently with both dark and light themes.
|
||||
*/
|
||||
:root {
|
||||
--ansi-red: #ff5555;
|
||||
--ansi-green: #50fa7b;
|
||||
--ansi-blue: #265285;
|
||||
--ansi-yellow: #ffb86c;
|
||||
--ansi-magenta: #bd93f9;
|
||||
--ansi-cyan: #8be9fd;
|
||||
--ansi-black: #282a36;
|
||||
--ansi-white: #f8f8f2;
|
||||
}
|
||||
|
||||
.-Color-Green,
|
||||
.-Color-Faint-Green,
|
||||
.-Color-Bold-Green,
|
||||
.-Color-BrightGreen {
|
||||
color: var(--ansi-green);
|
||||
}
|
||||
|
||||
.-Color-Red,
|
||||
.-Color-Faint-Red,
|
||||
.-Color-Bold-Red,
|
||||
.-Color-BrightRed {
|
||||
color: var(--ansi-red);
|
||||
}
|
||||
|
||||
.-Color-Yellow,
|
||||
.-Color-Faint-Yellow,
|
||||
.-Color-Bold-Yellow,
|
||||
.-Color-BrightYellow {
|
||||
color: var(--ansi-yellow);
|
||||
}
|
||||
|
||||
.-Color-Blue,
|
||||
.-Color-Faint-Blue,
|
||||
.-Color-Bold-Blue,
|
||||
.-Color-BrightBlue {
|
||||
color: var(--ansi-blue);
|
||||
}
|
||||
|
||||
.-Color-Magenta,
|
||||
.-Color-Faint-Magenta,
|
||||
.-Color-Bold-Magenta,
|
||||
.-Color-BrightMagenta {
|
||||
color: var(--ansi-magenta);
|
||||
}
|
||||
|
||||
.-Color-Cyan,
|
||||
.-Color-Faint-Cyan,
|
||||
.-Color-Bold-Cyan,
|
||||
.-Color-BrightCyan {
|
||||
color: var(--ansi-cyan);
|
||||
}
|
||||
|
||||
.-Color-White,
|
||||
.-Color-Faint-White,
|
||||
.-Color-Bold-White,
|
||||
.-Color-BrightWhite {
|
||||
color: var(--ansi-white);
|
||||
}
|
||||
|
||||
.-Color-Black,
|
||||
.-Color-Faint-Black,
|
||||
.-Color-Bold-Black,
|
||||
.-Color-BrightBlack {
|
||||
color: var(--ansi-black);
|
||||
}
|
||||
|
||||
.-Color-Faint {
|
||||
opacity: 0.5;
|
||||
}
|
||||
|
||||
.-Color-Bold {
|
||||
font-weight: bold;
|
||||
}
|
||||
|
||||
.-Color-BGBlack,
|
||||
.-Color-Black-BGBlack,
|
||||
.-Color-Blue-BGBlack,
|
||||
.-Color-Bold-BGBlack,
|
||||
.-Color-BrightBGBlack,
|
||||
.-Color-Bold-Black-BGBlack,
|
||||
.-Color-BrightBlack-BGBlack,
|
||||
.-Color-Bold-Green-BGBlack,
|
||||
.-Color-BrightGreen-BGBlack,
|
||||
.-Color-Bold-Cyan-BGBlack,
|
||||
.-Color-BrightCyan-BGBlack,
|
||||
.-Color-Bold-Blue-BGBlack,
|
||||
.-Color-BrightBlue-BGBlack,
|
||||
.-Color-Bold-Magenta-BGBlack,
|
||||
.-Color-BrightMagenta-BGBlack,
|
||||
.-Color-Bold-Red-BGBlack,
|
||||
.-Color-BrightRed-BGBlack,
|
||||
.-Color-Bold-White-BGBlack,
|
||||
.-Color-BrightWhite-BGBlack,
|
||||
.-Color-Bold-Yellow-BGBlack,
|
||||
.-Color-BrightYellow-BGBlack,
|
||||
.-Color-Cyan-BGBlack,
|
||||
.-Color-Green-BGBlack,
|
||||
.-Color-Magenta-BGBlack,
|
||||
.-Color-Red-BGBlack,
|
||||
.-Color-White-BGBlack,
|
||||
.-Color-Yellow-BGBlack {
|
||||
background-color: var(--ansi-black);
|
||||
}
|
||||
|
||||
.-Color-BGRed,
|
||||
.-Color-Black-BGRed,
|
||||
.-Color-Blue-BGRed,
|
||||
.-Color-Bold-BGRed,
|
||||
.-Color-BrightBGRed,
|
||||
.-Color-Bold-Black-BGRed,
|
||||
.-Color-BrightBlack-BGRed,
|
||||
.-Color-Bold-Green-BGRed,
|
||||
.-Color-BrightGreen-BGRed,
|
||||
.-Color-Bold-Cyan-BGRed,
|
||||
.-Color-BrightCyan-BGRed,
|
||||
.-Color-Bold-Blue-BGRed,
|
||||
.-Color-BrightBlue-BGRed,
|
||||
.-Color-Bold-Magenta-BGRed,
|
||||
.-Color-BrightMagenta-BGRed,
|
||||
.-Color-Bold-Red-BGRed,
|
||||
.-Color-BrightRed-BGRed,
|
||||
.-Color-Bold-White-BGRed,
|
||||
.-Color-BrightWhite-BGRed,
|
||||
.-Color-Bold-Yellow-BGRed,
|
||||
.-Color-BrightYellow-BGRed,
|
||||
.-Color-Cyan-BGRed,
|
||||
.-Color-Green-BGRed,
|
||||
.-Color-Magenta-BGRed,
|
||||
.-Color-Red-BGRed,
|
||||
.-Color-White-BGRed,
|
||||
.-Color-Yellow-BGRed {
|
||||
background-color: var(--ansi-red);
|
||||
}
|
||||
|
||||
.-Color-BGGreen,
|
||||
.-Color-Black-BGGreen,
|
||||
.-Color-Blue-BGGreen,
|
||||
.-Color-Bold-BGGreen,
|
||||
.-Color-BrightBGGreen,
|
||||
.-Color-Bold-Black-BGGreen,
|
||||
.-Color-BrightBlack-BGGreen,
|
||||
.-Color-Bold-Green-BGGreen,
|
||||
.-Color-BrightGreen-BGGreen,
|
||||
.-Color-Bold-Cyan-BGGreen,
|
||||
.-Color-BrightCyan-BGGreen,
|
||||
.-Color-Bold-Blue-BGGreen,
|
||||
.-Color-BrightBlue-BGGreen,
|
||||
.-Color-Bold-Magenta-BGGreen,
|
||||
.-Color-BrightMagenta-BGGreen,
|
||||
.-Color-Bold-Red-BGGreen,
|
||||
.-Color-BrightRed-BGGreen,
|
||||
.-Color-Bold-White-BGGreen,
|
||||
.-Color-BrightWhite-BGGreen,
|
||||
.-Color-Bold-Yellow-BGGreen,
|
||||
.-Color-BrightYellow-BGGreen,
|
||||
.-Color-Cyan-BGGreen,
|
||||
.-Color-Green-BGGreen,
|
||||
.-Color-Magenta-BGGreen,
|
||||
.-Color-Red-BGGreen,
|
||||
.-Color-White-BGGreen,
|
||||
.-Color-Yellow-BGGreen {
|
||||
background-color: var(--ansi-green);
|
||||
}
|
||||
|
||||
.-Color-BGYellow,
|
||||
.-Color-Black-BGYellow,
|
||||
.-Color-Blue-BGYellow,
|
||||
.-Color-Bold-BGYellow,
|
||||
.-Color-BrightBGYellow,
|
||||
.-Color-Bold-Black-BGYellow,
|
||||
.-Color-BrightBlack-BGYellow,
|
||||
.-Color-Bold-Green-BGYellow,
|
||||
.-Color-BrightGreen-BGYellow,
|
||||
.-Color-Bold-Cyan-BGYellow,
|
||||
.-Color-BrightCyan-BGYellow,
|
||||
.-Color-Bold-Blue-BGYellow,
|
||||
.-Color-BrightBlue-BGYellow,
|
||||
.-Color-Bold-Magenta-BGYellow,
|
||||
.-Color-BrightMagenta-BGYellow,
|
||||
.-Color-Bold-Red-BGYellow,
|
||||
.-Color-BrightRed-BGYellow,
|
||||
.-Color-Bold-White-BGYellow,
|
||||
.-Color-BrightWhite-BGYellow,
|
||||
.-Color-Bold-Yellow-BGYellow,
|
||||
.-Color-BrightYellow-BGYellow,
|
||||
.-Color-Cyan-BGYellow,
|
||||
.-Color-Green-BGYellow,
|
||||
.-Color-Magenta-BGYellow,
|
||||
.-Color-Red-BGYellow,
|
||||
.-Color-White-BGYellow,
|
||||
.-Color-Yellow-BGYellow {
|
||||
background-color: var(--ansi-yellow);
|
||||
}
|
||||
|
||||
.-Color-BGBlue,
|
||||
.-Color-Black-BGBlue,
|
||||
.-Color-Blue-BGBlue,
|
||||
.-Color-Bold-BGBlue,
|
||||
.-Color-BrightBGBlue,
|
||||
.-Color-Bold-Black-BGBlue,
|
||||
.-Color-BrightBlack-BGBlue,
|
||||
.-Color-Bold-Green-BGBlue,
|
||||
.-Color-BrightGreen-BGBlue,
|
||||
.-Color-Bold-Cyan-BGBlue,
|
||||
.-Color-BrightCyan-BGBlue,
|
||||
.-Color-Bold-Blue-BGBlue,
|
||||
.-Color-BrightBlue-BGBlue,
|
||||
.-Color-Bold-Magenta-BGBlue,
|
||||
.-Color-BrightMagenta-BGBlue,
|
||||
.-Color-Bold-Red-BGBlue,
|
||||
.-Color-BrightRed-BGBlue,
|
||||
.-Color-Bold-White-BGBlue,
|
||||
.-Color-BrightWhite-BGBlue,
|
||||
.-Color-Bold-Yellow-BGBlue,
|
||||
.-Color-BrightYellow-BGBlue,
|
||||
.-Color-Cyan-BGBlue,
|
||||
.-Color-Green-BGBlue,
|
||||
.-Color-Magenta-BGBlue,
|
||||
.-Color-Red-BGBlue,
|
||||
.-Color-White-BGBlue,
|
||||
.-Color-Yellow-BGBlue {
|
||||
background-color: var(--ansi-blue);
|
||||
}
|
||||
|
||||
.-Color-BGMagenta,
|
||||
.-Color-Black-BGMagenta,
|
||||
.-Color-Blue-BGMagenta,
|
||||
.-Color-Bold-BGMagenta,
|
||||
.-Color-BrightBGMagenta,
|
||||
.-Color-Bold-Black-BGMagenta,
|
||||
.-Color-BrightBlack-BGMagenta,
|
||||
.-Color-Bold-Green-BGMagenta,
|
||||
.-Color-BrightGreen-BGMagenta,
|
||||
.-Color-Bold-Cyan-BGMagenta,
|
||||
.-Color-BrightCyan-BGMagenta,
|
||||
.-Color-Bold-Blue-BGMagenta,
|
||||
.-Color-BrightBlue-BGMagenta,
|
||||
.-Color-Bold-Magenta-BGMagenta,
|
||||
.-Color-BrightMagenta-BGMagenta,
|
||||
.-Color-Bold-Red-BGMagenta,
|
||||
.-Color-BrightRed-BGMagenta,
|
||||
.-Color-Bold-White-BGMagenta,
|
||||
.-Color-BrightWhite-BGMagenta,
|
||||
.-Color-Bold-Yellow-BGMagenta,
|
||||
.-Color-BrightYellow-BGMagenta,
|
||||
.-Color-Cyan-BGMagenta,
|
||||
.-Color-Green-BGMagenta,
|
||||
.-Color-Magenta-BGMagenta,
|
||||
.-Color-Red-BGMagenta,
|
||||
.-Color-White-BGMagenta,
|
||||
.-Color-Yellow-BGMagenta {
|
||||
background-color: var(--ansi-magenta);
|
||||
}
|
||||
|
||||
.-Color-BGCyan,
|
||||
.-Color-Black-BGCyan,
|
||||
.-Color-Blue-BGCyan,
|
||||
.-Color-Bold-BGCyan,
|
||||
.-Color-BrightBGCyan,
|
||||
.-Color-Bold-Black-BGCyan,
|
||||
.-Color-BrightBlack-BGCyan,
|
||||
.-Color-Bold-Green-BGCyan,
|
||||
.-Color-BrightGreen-BGCyan,
|
||||
.-Color-Bold-Cyan-BGCyan,
|
||||
.-Color-BrightCyan-BGCyan,
|
||||
.-Color-Bold-Blue-BGCyan,
|
||||
.-Color-BrightBlue-BGCyan,
|
||||
.-Color-Bold-Magenta-BGCyan,
|
||||
.-Color-BrightMagenta-BGCyan,
|
||||
.-Color-Bold-Red-BGCyan,
|
||||
.-Color-BrightRed-BGCyan,
|
||||
.-Color-Bold-White-BGCyan,
|
||||
.-Color-BrightWhite-BGCyan,
|
||||
.-Color-Bold-Yellow-BGCyan,
|
||||
.-Color-BrightYellow-BGCyan,
|
||||
.-Color-Cyan-BGCyan,
|
||||
.-Color-Green-BGCyan,
|
||||
.-Color-Magenta-BGCyan,
|
||||
.-Color-Red-BGCyan,
|
||||
.-Color-White-BGCyan,
|
||||
.-Color-Yellow-BGCyan {
|
||||
background-color: var(--ansi-cyan);
|
||||
}
|
||||
|
||||
.-Color-BGWhite,
|
||||
.-Color-Black-BGWhite,
|
||||
.-Color-Blue-BGWhite,
|
||||
.-Color-Bold-BGWhite,
|
||||
.-Color-BrightBGWhite,
|
||||
.-Color-Bold-Black-BGWhite,
|
||||
.-Color-BrightBlack-BGWhite,
|
||||
.-Color-Bold-Green-BGWhite,
|
||||
.-Color-BrightGreen-BGWhite,
|
||||
.-Color-Bold-Cyan-BGWhite,
|
||||
.-Color-BrightCyan-BGWhite,
|
||||
.-Color-Bold-Blue-BGWhite,
|
||||
.-Color-BrightBlue-BGWhite,
|
||||
.-Color-Bold-Magenta-BGWhite,
|
||||
.-Color-BrightMagenta-BGWhite,
|
||||
.-Color-Bold-Red-BGWhite,
|
||||
.-Color-BrightRed-BGWhite,
|
||||
.-Color-Bold-White-BGWhite,
|
||||
.-Color-BrightWhite-BGWhite,
|
||||
.-Color-Bold-Yellow-BGWhite,
|
||||
.-Color-BrightYellow-BGWhite,
|
||||
.-Color-Cyan-BGWhite,
|
||||
.-Color-Green-BGWhite,
|
||||
.-Color-Magenta-BGWhite,
|
||||
.-Color-Red-BGWhite,
|
||||
.-Color-White-BGWhite,
|
||||
.-Color-Yellow-BGWhite {
|
||||
background-color: var(--ansi-white);
|
||||
}
|
||||
|
||||
.-Color-Black,
|
||||
.-Color-Bold-Black,
|
||||
.-Color-BrightBlack,
|
||||
.-Color-Black-BGBlack,
|
||||
.-Color-Bold-Black-BGBlack,
|
||||
.-Color-BrightBlack-BGBlack,
|
||||
.-Color-Black-BGGreen,
|
||||
.-Color-Red-BGRed,
|
||||
.-Color-Bold-Red-BGRed,
|
||||
.-Color-BrightRed-BGRed,
|
||||
.-Color-Bold-Blue-BGBlue,
|
||||
.-Color-BrightBlue-BGBlue,
|
||||
.-Color-Blue-BGBlue {
|
||||
text-shadow: 0 0 1px var(--ansi-white);
|
||||
}
|
||||
|
||||
.-Color-Bold-Cyan-BGCyan,
|
||||
.-Color-BrightCyan-BGCyan,
|
||||
.-Color-Bold-Magenta-BGMagenta,
|
||||
.-Color-BrightMagenta-BGMagenta,
|
||||
.-Color-Bold-White,
|
||||
.-Color-BrightWhite,
|
||||
.-Color-Bold-Yellow-BGYellow,
|
||||
.-Color-BrightYellow-BGYellow,
|
||||
.-Color-Bold-Green-BGGreen,
|
||||
.-Color-BrightGreen-BGGreen,
|
||||
.-Color-Cyan-BGCyan,
|
||||
.-Color-Cyan-BGGreen,
|
||||
.-Color-Green-BGCyan,
|
||||
.-Color-Green-BGGreen,
|
||||
.-Color-Magenta-BGMagenta,
|
||||
.-Color-White,
|
||||
.-Color-White-BGWhite,
|
||||
.-Color-Yellow-BGYellow {
|
||||
text-shadow: 0 0 1px var(--ansi-black);
|
||||
}
|
||||
@@ -0,0 +1,53 @@
|
||||
html[data-theme="light"] {
|
||||
@import "https://cdn.jsdelivr.net/npm/highlightjs-themes@1.0.0/tomorrow.css"
|
||||
}
|
||||
|
||||
html[data-theme="dark"] {
|
||||
@import "https://cdn.jsdelivr.net/npm/highlightjs-themes@1.0.0/tomorrow-night-blue.min.css"
|
||||
}
|
||||
|
||||
|
||||
.ace_gutter {
|
||||
z-index: 1;
|
||||
}
|
||||
|
||||
.pyodide-editor {
|
||||
width: 100%;
|
||||
font-size: .85em;
|
||||
}
|
||||
|
||||
.pyodide-editor-bar {
|
||||
color: var(--md-primary-bg-color);
|
||||
background-color: var(--md-primary-fg-color);
|
||||
width: 100%;
|
||||
font: monospace;
|
||||
font-size: 0.75em;
|
||||
padding: 2px 0 2px;
|
||||
}
|
||||
|
||||
.pyodide-bar-item {
|
||||
padding: 0 18px 0;
|
||||
display: inline-block;
|
||||
width: 50%;
|
||||
}
|
||||
|
||||
.pyodide pre {
|
||||
margin: 0;
|
||||
}
|
||||
|
||||
.pyodide-output {
|
||||
width: 100%;
|
||||
margin-bottom: -15px;
|
||||
min-height: 46px;
|
||||
max-height: 400px
|
||||
}
|
||||
|
||||
.pyodide-clickable {
|
||||
cursor: pointer;
|
||||
text-align: right;
|
||||
}
|
||||
|
||||
/* For themes other than Material. */
|
||||
.pyodide .twemoji svg {
|
||||
width: 1rem;
|
||||
}
|
||||
@@ -0,0 +1,131 @@
|
||||
var _sessions = {};
|
||||
|
||||
function getSession(name, pyodide) {
|
||||
if (!(name in _sessions)) {
|
||||
_sessions[name] = pyodide.globals.get("dict")();
|
||||
}
|
||||
return _sessions[name];
|
||||
}
|
||||
|
||||
function writeOutput(element, string) {
|
||||
element.innerHTML += string + '\n';
|
||||
}
|
||||
|
||||
function clearOutput(element) {
|
||||
element.innerHTML = '';
|
||||
}
|
||||
|
||||
async function evaluatePython(pyodide, editor, output, session) {
|
||||
pyodide.setStdout({ batched: (string) => { writeOutput(output, new Option(string).innerHTML); } });
|
||||
let result, code = editor.getValue();
|
||||
clearOutput(output);
|
||||
try {
|
||||
result = await pyodide.runPythonAsync(code, { globals: getSession(session, pyodide) });
|
||||
} catch (error) {
|
||||
writeOutput(output, new Option(error.toString()).innerHTML);
|
||||
}
|
||||
if (result) writeOutput(output, new Option(result).innerHTML);
|
||||
hljs.highlightElement(output);
|
||||
}
|
||||
|
||||
async function initPyodide() {
|
||||
try {
|
||||
let pyodide = await loadPyodide();
|
||||
await pyodide.loadPackage("micropip");
|
||||
return pyodide;
|
||||
} catch(error) {
|
||||
return null;
|
||||
}
|
||||
}
|
||||
|
||||
function getTheme() {
|
||||
return document.body.getAttribute('data-md-color-scheme');
|
||||
}
|
||||
|
||||
function setTheme(editor, currentTheme, light, dark) {
|
||||
// https://gist.github.com/RyanNutt/cb8d60997d97905f0b2aea6c3b5c8ee0
|
||||
if (currentTheme === "default") {
|
||||
editor.setTheme("ace/theme/" + light);
|
||||
document.querySelector(`link[title="light"]`).removeAttribute("disabled");
|
||||
document.querySelector(`link[title="dark"]`).setAttribute("disabled", "disabled");
|
||||
} else if (currentTheme === "slate") {
|
||||
editor.setTheme("ace/theme/" + dark);
|
||||
document.querySelector(`link[title="dark"]`).removeAttribute("disabled");
|
||||
document.querySelector(`link[title="light"]`).setAttribute("disabled", "disabled");
|
||||
}
|
||||
}
|
||||
|
||||
function updateTheme(editor, light, dark) {
|
||||
// Create a new MutationObserver instance
|
||||
const observer = new MutationObserver((mutations) => {
|
||||
// Loop through the mutations that occurred
|
||||
mutations.forEach((mutation) => {
|
||||
// Check if the mutation was a change to the data-md-color-scheme attribute
|
||||
if (mutation.attributeName === 'data-md-color-scheme') {
|
||||
// Get the new value of the attribute
|
||||
const newColorScheme = mutation.target.getAttribute('data-md-color-scheme');
|
||||
// Update the editor theme
|
||||
setTheme(editor, newColorScheme, light, dark);
|
||||
}
|
||||
});
|
||||
});
|
||||
|
||||
// Configure the observer to watch for changes to the data-md-color-scheme attribute
|
||||
observer.observe(document.body, {
|
||||
attributes: true,
|
||||
attributeFilter: ['data-md-color-scheme'],
|
||||
});
|
||||
}
|
||||
|
||||
async function setupPyodide(
|
||||
idPrefix,
|
||||
install = null,
|
||||
themeLight = 'tomorrow',
|
||||
themeDark = 'tomorrow_night',
|
||||
session = null,
|
||||
minLines = 5,
|
||||
maxLines = 30,
|
||||
) {
|
||||
const editor = ace.edit(idPrefix + "editor");
|
||||
const run = document.getElementById(idPrefix + "run");
|
||||
const clear = document.getElementById(idPrefix + "clear");
|
||||
const output = document.getElementById(idPrefix + "output");
|
||||
|
||||
updateTheme(editor, themeLight, themeDark);
|
||||
|
||||
editor.session.setMode("ace/mode/python");
|
||||
setTheme(editor, getTheme(), themeLight, themeDark);
|
||||
|
||||
editor.setOption("minLines", minLines);
|
||||
editor.setOption("maxLines", maxLines);
|
||||
|
||||
// Force editor to resize after setting options
|
||||
editor.resize();
|
||||
|
||||
writeOutput(output, "Initializing...");
|
||||
let pyodide = await pyodidePromise;
|
||||
if (install && install.length) {
|
||||
try {
|
||||
micropip = pyodide.pyimport("micropip");
|
||||
for (const package of install)
|
||||
await micropip.install(package);
|
||||
clearOutput(output);
|
||||
} catch (error) {
|
||||
clearOutput(output);
|
||||
writeOutput(output, `Could not install one or more packages: ${install.join(", ")}\n`);
|
||||
writeOutput(output, new Option(error.toString()).innerHTML);
|
||||
}
|
||||
} else {
|
||||
clearOutput(output);
|
||||
}
|
||||
run.onclick = () => evaluatePython(pyodide, editor, output, session);
|
||||
clear.onclick = () => clearOutput(output);
|
||||
output.parentElement.parentElement.addEventListener("keydown", (event) => {
|
||||
if (event.ctrlKey && event.key.toLowerCase() === 'enter') {
|
||||
event.preventDefault();
|
||||
run.click();
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
var pyodidePromise = initPyodide();
|
||||
@@ -0,0 +1,237 @@
|
||||
|
||||
/* Avoid breaking parameter names, etc. in table cells. */
|
||||
.doc-contents td code {
|
||||
word-break: normal !important;
|
||||
}
|
||||
|
||||
/* No line break before first paragraph of descriptions. */
|
||||
.doc-md-description,
|
||||
.doc-md-description>p:first-child {
|
||||
display: inline;
|
||||
}
|
||||
|
||||
/* No text transformation from Material for MkDocs for H5 headings. */
|
||||
.md-typeset h5 .doc-object-name {
|
||||
text-transform: none;
|
||||
}
|
||||
|
||||
/* Max width for docstring sections tables. */
|
||||
.doc .md-typeset__table,
|
||||
.doc .md-typeset__table table {
|
||||
display: table !important;
|
||||
width: 100%;
|
||||
}
|
||||
|
||||
.doc .md-typeset__table tr {
|
||||
display: table-row;
|
||||
}
|
||||
|
||||
/* Defaults in Spacy table style. */
|
||||
.doc-param-default,
|
||||
.doc-type_param-default {
|
||||
float: right;
|
||||
}
|
||||
|
||||
/* Parameter headings must be inline, not blocks. */
|
||||
.doc-heading-parameter,
|
||||
.doc-heading-type_parameter {
|
||||
display: inline;
|
||||
}
|
||||
|
||||
/* Default font size for parameter headings. */
|
||||
.md-typeset .doc-heading-parameter {
|
||||
font-size: inherit;
|
||||
}
|
||||
|
||||
/* Prefer space on the right, not the left of parameter permalinks. */
|
||||
.doc-heading-parameter .headerlink,
|
||||
.doc-heading-type_parameter .headerlink {
|
||||
margin-left: 0 !important;
|
||||
margin-right: 0.2rem;
|
||||
}
|
||||
|
||||
/* Backward-compatibility: docstring section titles in bold. */
|
||||
.doc-section-title {
|
||||
font-weight: bold;
|
||||
}
|
||||
|
||||
/* Backlinks crumb separator. */
|
||||
.doc-backlink-crumb {
|
||||
display: inline-flex;
|
||||
gap: .2rem;
|
||||
white-space: nowrap;
|
||||
align-items: center;
|
||||
vertical-align: middle;
|
||||
}
|
||||
.doc-backlink-crumb:not(:first-child)::before {
|
||||
background-color: var(--md-default-fg-color--lighter);
|
||||
content: "";
|
||||
display: inline;
|
||||
height: 1rem;
|
||||
--md-path-icon: url('data:image/svg+xml;charset=utf-8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M8.59 16.58 13.17 12 8.59 7.41 10 6l6 6-6 6z"/></svg>');
|
||||
-webkit-mask-image: var(--md-path-icon);
|
||||
mask-image: var(--md-path-icon);
|
||||
width: 1rem;
|
||||
}
|
||||
.doc-backlink-crumb.last {
|
||||
font-weight: bold;
|
||||
}
|
||||
|
||||
/* Symbols in Navigation and ToC. */
|
||||
:root, :host,
|
||||
[data-md-color-scheme="default"] {
|
||||
--doc-symbol-parameter-fg-color: #df50af;
|
||||
--doc-symbol-type_parameter-fg-color: #df50af;
|
||||
--doc-symbol-attribute-fg-color: #953800;
|
||||
--doc-symbol-function-fg-color: #8250df;
|
||||
--doc-symbol-method-fg-color: #8250df;
|
||||
--doc-symbol-class-fg-color: #0550ae;
|
||||
--doc-symbol-type_alias-fg-color: #0550ae;
|
||||
--doc-symbol-module-fg-color: #5cad0f;
|
||||
|
||||
--doc-symbol-parameter-bg-color: #df50af1a;
|
||||
--doc-symbol-type_parameter-bg-color: #df50af1a;
|
||||
--doc-symbol-attribute-bg-color: #9538001a;
|
||||
--doc-symbol-function-bg-color: #8250df1a;
|
||||
--doc-symbol-method-bg-color: #8250df1a;
|
||||
--doc-symbol-class-bg-color: #0550ae1a;
|
||||
--doc-symbol-type_alias-bg-color: #0550ae1a;
|
||||
--doc-symbol-module-bg-color: #5cad0f1a;
|
||||
}
|
||||
|
||||
[data-md-color-scheme="slate"] {
|
||||
--doc-symbol-parameter-fg-color: #ffa8cc;
|
||||
--doc-symbol-type_parameter-fg-color: #ffa8cc;
|
||||
--doc-symbol-attribute-fg-color: #ffa657;
|
||||
--doc-symbol-function-fg-color: #d2a8ff;
|
||||
--doc-symbol-method-fg-color: #d2a8ff;
|
||||
--doc-symbol-class-fg-color: #79c0ff;
|
||||
--doc-symbol-type_alias-fg-color: #79c0ff;
|
||||
--doc-symbol-module-fg-color: #baff79;
|
||||
|
||||
--doc-symbol-parameter-bg-color: #ffa8cc1a;
|
||||
--doc-symbol-type_parameter-bg-color: #ffa8cc1a;
|
||||
--doc-symbol-attribute-bg-color: #ffa6571a;
|
||||
--doc-symbol-function-bg-color: #d2a8ff1a;
|
||||
--doc-symbol-method-bg-color: #d2a8ff1a;
|
||||
--doc-symbol-class-bg-color: #79c0ff1a;
|
||||
--doc-symbol-type_alias-bg-color: #79c0ff1a;
|
||||
--doc-symbol-module-bg-color: #baff791a;
|
||||
}
|
||||
|
||||
code.doc-symbol {
|
||||
border-radius: .1rem;
|
||||
font-size: .85em;
|
||||
padding: 0 .3em;
|
||||
font-weight: bold;
|
||||
}
|
||||
|
||||
code.doc-symbol-parameter,
|
||||
a code.doc-symbol-parameter {
|
||||
color: var(--doc-symbol-parameter-fg-color);
|
||||
background-color: var(--doc-symbol-parameter-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-parameter::after {
|
||||
content: "param";
|
||||
}
|
||||
|
||||
code.doc-symbol-type_parameter,
|
||||
a code.doc-symbol-type_parameter {
|
||||
color: var(--doc-symbol-type_parameter-fg-color);
|
||||
background-color: var(--doc-symbol-type_parameter-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-type_parameter::after {
|
||||
content: "type-param";
|
||||
}
|
||||
|
||||
code.doc-symbol-attribute,
|
||||
a code.doc-symbol-attribute {
|
||||
color: var(--doc-symbol-attribute-fg-color);
|
||||
background-color: var(--doc-symbol-attribute-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-attribute::after {
|
||||
content: "attr";
|
||||
}
|
||||
|
||||
code.doc-symbol-function,
|
||||
a code.doc-symbol-function {
|
||||
color: var(--doc-symbol-function-fg-color);
|
||||
background-color: var(--doc-symbol-function-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-function::after {
|
||||
content: "func";
|
||||
}
|
||||
|
||||
code.doc-symbol-method,
|
||||
a code.doc-symbol-method {
|
||||
color: var(--doc-symbol-method-fg-color);
|
||||
background-color: var(--doc-symbol-method-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-method::after {
|
||||
content: "meth";
|
||||
}
|
||||
|
||||
code.doc-symbol-class,
|
||||
a code.doc-symbol-class {
|
||||
color: var(--doc-symbol-class-fg-color);
|
||||
background-color: var(--doc-symbol-class-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-class::after {
|
||||
content: "class";
|
||||
}
|
||||
|
||||
|
||||
code.doc-symbol-type_alias,
|
||||
a code.doc-symbol-type_alias {
|
||||
color: var(--doc-symbol-type_alias-fg-color);
|
||||
background-color: var(--doc-symbol-type_alias-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-type_alias::after {
|
||||
content: "type";
|
||||
}
|
||||
|
||||
code.doc-symbol-module,
|
||||
a code.doc-symbol-module {
|
||||
color: var(--doc-symbol-module-fg-color);
|
||||
background-color: var(--doc-symbol-module-bg-color);
|
||||
}
|
||||
|
||||
code.doc-symbol-module::after {
|
||||
content: "mod";
|
||||
}
|
||||
|
||||
.doc-signature .autorefs {
|
||||
color: inherit;
|
||||
border-bottom: 1px dotted currentcolor;
|
||||
}
|
||||
|
||||
/* Source code blocks (admonitions). */
|
||||
:root {
|
||||
--md-admonition-icon--mkdocstrings-source: url('data:image/svg+xml;charset=utf-8,<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M15.22 4.97a.75.75 0 0 1 1.06 0l6.5 6.5a.75.75 0 0 1 0 1.06l-6.5 6.5a.749.749 0 0 1-1.275-.326.75.75 0 0 1 .215-.734L21.19 12l-5.97-5.97a.75.75 0 0 1 0-1.06m-6.44 0a.75.75 0 0 1 0 1.06L2.81 12l5.97 5.97a.749.749 0 0 1-.326 1.275.75.75 0 0 1-.734-.215l-6.5-6.5a.75.75 0 0 1 0-1.06l6.5-6.5a.75.75 0 0 1 1.06 0"/></svg>')
|
||||
}
|
||||
.md-typeset .admonition.mkdocstrings-source,
|
||||
.md-typeset details.mkdocstrings-source {
|
||||
border: none;
|
||||
padding: 0;
|
||||
}
|
||||
.md-typeset .admonition.mkdocstrings-source:focus-within,
|
||||
.md-typeset details.mkdocstrings-source:focus-within {
|
||||
box-shadow: none;
|
||||
}
|
||||
.md-typeset .mkdocstrings-source > .admonition-title,
|
||||
.md-typeset .mkdocstrings-source > summary {
|
||||
background-color: inherit;
|
||||
}
|
||||
.md-typeset .mkdocstrings-source > .admonition-title::before,
|
||||
.md-typeset .mkdocstrings-source > summary::before {
|
||||
background-color: var(--md-default-fg-color);
|
||||
-webkit-mask-image: var(--md-admonition-icon--mkdocstrings-source);
|
||||
mask-image: var(--md-admonition-icon--mkdocstrings-source);
|
||||
}
|
||||
Binary file not shown.
|
After Width: | Height: | Size: 1.8 KiB |
+16
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
+1
File diff suppressed because one or more lines are too long
+18
@@ -0,0 +1,18 @@
|
||||
/*!
|
||||
* Lunr languages, `Danish` language
|
||||
* https://github.com/MihaiValentin/lunr-languages
|
||||
*
|
||||
* Copyright 2014, Mihai Valentin
|
||||
* http://www.mozilla.org/MPL/
|
||||
*/
|
||||
/*!
|
||||
* based on
|
||||
* Snowball JavaScript Library v0.3
|
||||
* http://code.google.com/p/urim/
|
||||
* http://snowball.tartarus.org/
|
||||
*
|
||||
* Copyright 2010, Oleg Mazko
|
||||
* http://www.mozilla.org/MPL/
|
||||
*/
|
||||
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.da=function(){this.pipeline.reset(),this.pipeline.add(e.da.trimmer,e.da.stopWordFilter,e.da.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.da.stemmer))},e.da.wordCharacters="A-Za-zªºÀ-ÖØ-öø-ʸˠ-ˤᴀ-ᴥᴬ-ᵜᵢ-ᵥᵫ-ᵷᵹ-ᶾḀ-ỿⁱⁿₐ-ₜKÅℲⅎⅠ-ↈⱠ-ⱿꜢ-ꞇꞋ-ꞭꞰ-ꞷꟷ-ꟿꬰ-ꭚꭜ-ꭤff-stＡ-Ｚａ-ｚ",e.da.trimmer=e.trimmerSupport.generateTrimmer(e.da.wordCharacters),e.Pipeline.registerFunction(e.da.trimmer,"trimmer-da"),e.da.stemmer=function(){var r=e.stemmerSupport.Among,i=e.stemmerSupport.SnowballProgram,n=new function(){function e(){var e,r=f.cursor+3;if(d=f.limit,0<=r&&r<=f.limit){for(a=r;;){if(e=f.cursor,f.in_grouping(w,97,248)){f.cursor=e;break}if(f.cursor=e,e>=f.limit)return;f.cursor++}for(;!f.out_grouping(w,97,248);){if(f.cursor>=f.limit)return;f.cursor++}d=f.cursor,d<a&&(d=a)}}function n(){var e,r;if(f.cursor>=d&&(r=f.limit_backward,f.limit_backward=d,f.ket=f.cursor,e=f.find_among_b(c,32),f.limit_backward=r,e))switch(f.bra=f.cursor,e){case 1:f.slice_del();break;case 2:f.in_grouping_b(p,97,229)&&f.slice_del()}}function t(){var e,r=f.limit-f.cursor;f.cursor>=d&&(e=f.limit_backward,f.limit_backward=d,f.ket=f.cursor,f.find_among_b(l,4)?(f.bra=f.cursor,f.limit_backward=e,f.cursor=f.limit-r,f.cursor>f.limit_backward&&(f.cursor--,f.bra=f.cursor,f.slice_del())):f.limit_backward=e)}function s(){var e,r,i,n=f.limit-f.cursor;if(f.ket=f.cursor,f.eq_s_b(2,"st")&&(f.bra=f.cursor,f.eq_s_b(2,"ig")&&f.slice_del()),f.cursor=f.limit-n,f.cursor>=d&&(r=f.limit_backward,f.limit_backward=d,f.ket=f.cursor,e=f.find_among_b(m,5),f.limit_backward=r,e))switch(f.bra=f.cursor,e){case 1:f.slice_del(),i=f.limit-f.cursor,t(),f.cursor=f.limit-i;break;case 2:f.slice_from("løs")}}function o(){var e;f.cursor>=d&&(e=f.limit_backward,f.limit_backward=d,f.ket=f.cursor,f.out_grouping_b(w,97,248)?(f.bra=f.cursor,u=f.slice_to(u),f.limit_backward=e,f.eq_v_b(u)&&f.slice_del()):f.limit_backward=e)}var a,d,u,c=[new r("hed",-1,1),new r("ethed",0,1),new r("ered",-1,1),new r("e",-1,1),new r("erede",3,1),new r("ende",3,1),new r("erende",5,1),new r("ene",3,1),new r("erne",3,1),new r("ere",3,1),new r("en",-1,1),new r("heden",10,1),new r("eren",10,1),new r("er",-1,1),new r("heder",13,1),new r("erer",13,1),new r("s",-1,2),new r("heds",16,1),new r("es",16,1),new r("endes",18,1),new r("erendes",19,1),new r("enes",18,1),new r("ernes",18,1),new r("eres",18,1),new r("ens",16,1),new r("hedens",24,1),new r("erens",24,1),new r("ers",16,1),new r("ets",16,1),new r("erets",28,1),new r("et",-1,1),new r("eret",30,1)],l=[new r("gd",-1,-1),new r("dt",-1,-1),new r("gt",-1,-1),new r("kt",-1,-1)],m=[new r("ig",-1,1),new r("lig",0,1),new r("elig",1,1),new r("els",-1,1),new r("løst",-1,2)],w=[17,65,16,1,0,0,0,0,0,0,0,0,0,0,0,0,48,0,128],p=[239,254,42,3,0,0,0,0,0,0,0,0,0,0,0,0,16],f=new i;this.setCurrent=function(e){f.setCurrent(e)},this.getCurrent=function(){return f.getCurrent()},this.stem=function(){var r=f.cursor;return e(),f.limit_backward=r,f.cursor=f.limit,n(),f.cursor=f.limit,t(),f.cursor=f.limit,s(),f.cursor=f.limit,o(),!0}};return function(e){return"function"==typeof e.update?e.update(function(e){return n.setCurrent(e),n.stem(),n.getCurrent()}):(n.setCurrent(e),n.stem(),n.getCurrent())}}(),e.Pipeline.registerFunction(e.da.stemmer,"stemmer-da"),e.da.stopWordFilter=e.generateStopWordFilter("ad af alle alt anden at blev blive bliver da de dem den denne der deres det dette dig din disse dog du efter eller en end er et for fra ham han hans har havde have hende hendes her hos hun hvad hvis hvor i ikke ind jeg jer jo kunne man mange med meget men mig min mine mit mod ned noget nogle nu når og også om op os over på selv sig sin sine sit skal skulle som sådan thi til ud under var vi vil ville vor være været".split(" ")),e.Pipeline.registerFunction(e.da.stopWordFilter,"stopWordFilter-da")}});
|
||||
+18
File diff suppressed because one or more lines are too long
+18
File diff suppressed because one or more lines are too long
+1
File diff suppressed because one or more lines are too long
+18
File diff suppressed because one or more lines are too long
+18
File diff suppressed because one or more lines are too long
+18
File diff suppressed because one or more lines are too long
+1
File diff suppressed because one or more lines are too long
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.hi=function(){this.pipeline.reset(),this.pipeline.add(e.hi.trimmer,e.hi.stopWordFilter,e.hi.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.hi.stemmer))},e.hi.wordCharacters="ऀ-ःऄ-एऐ-टठ-यर-िी-ॏॐ-य़ॠ-९॰-ॿa-zA-Zａ-ｚＡ-Ｚ0-9０-９",e.hi.trimmer=e.trimmerSupport.generateTrimmer(e.hi.wordCharacters),e.Pipeline.registerFunction(e.hi.trimmer,"trimmer-hi"),e.hi.stopWordFilter=e.generateStopWordFilter("अत अपना अपनी अपने अभी अंदर आदि आप इत्यादि इन इनका इन्हीं इन्हें इन्हों इस इसका इसकी इसके इसमें इसी इसे उन उनका उनकी उनके उनको उन्हीं उन्हें उन्हों उस उसके उसी उसे एक एवं एस ऐसे और कई कर करता करते करना करने करें कहते कहा का काफ़ी कि कितना किन्हें किन्हों किया किर किस किसी किसे की कुछ कुल के को कोई कौन कौनसा गया घर जब जहाँ जा जितना जिन जिन्हें जिन्हों जिस जिसे जीधर जैसा जैसे जो तक तब तरह तिन तिन्हें तिन्हों तिस तिसे तो था थी थे दबारा दिया दुसरा दूसरे दो द्वारा न नके नहीं ना निहायत नीचे ने पर पहले पूरा पे फिर बनी बही बहुत बाद बाला बिलकुल भी भीतर मगर मानो मे में यदि यह यहाँ यही या यिह ये रखें रहा रहे ऱ्वासा लिए लिये लेकिन व वग़ैरह वर्ग वह वहाँ वहीं वाले वुह वे वो सकता सकते सबसे सभी साथ साबुत साभ सारा से सो संग ही हुआ हुई हुए है हैं हो होता होती होते होना होने".split(" ")),e.hi.stemmer=function(){return function(e){return"function"==typeof e.update?e.update(function(e){return e}):e}}();var r=e.wordcut;r.init(),e.hi.tokenizer=function(i){if(!arguments.length||null==i||void 0==i)return[];if(Array.isArray(i))return i.map(function(r){return isLunr2?new e.Token(r.toLowerCase()):r.toLowerCase()});var t=i.toString().toLowerCase().replace(/^\s+/,"");return r.cut(t).split("|")},e.Pipeline.registerFunction(e.hi.stemmer,"stemmer-hi"),e.Pipeline.registerFunction(e.hi.stopWordFilter,"stopWordFilter-hi")}});
|
||||
+18
File diff suppressed because one or more lines are too long
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.hy=function(){this.pipeline.reset(),this.pipeline.add(e.hy.trimmer,e.hy.stopWordFilter)},e.hy.wordCharacters="[A-Za-z-֏ff-ﭏ]",e.hy.trimmer=e.trimmerSupport.generateTrimmer(e.hy.wordCharacters),e.Pipeline.registerFunction(e.hy.trimmer,"trimmer-hy"),e.hy.stopWordFilter=e.generateStopWordFilter("դու և եք էիր էիք հետո նաև նրանք որը վրա է որ պիտի են այս մեջ ն իր ու ի այդ որոնք այն կամ էր մի ես համար այլ իսկ էին ենք հետ ին թ էինք մենք նրա նա դուք եմ էի ըստ որպես ում".split(" ")),e.Pipeline.registerFunction(e.hy.stopWordFilter,"stopWordFilter-hy"),e.hy.stemmer=function(){return function(e){return"function"==typeof e.update?e.update(function(e){return e}):e}}(),e.Pipeline.registerFunction(e.hy.stemmer,"stemmer-hy")}});
|
||||
+18
File diff suppressed because one or more lines are too long
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");var r="2"==e.version[0];e.ja=function(){this.pipeline.reset(),this.pipeline.add(e.ja.trimmer,e.ja.stopWordFilter,e.ja.stemmer),r?this.tokenizer=e.ja.tokenizer:(e.tokenizer&&(e.tokenizer=e.ja.tokenizer),this.tokenizerFn&&(this.tokenizerFn=e.ja.tokenizer))};var t=new e.TinySegmenter;e.ja.tokenizer=function(i){var n,o,s,p,a,u,m,l,c,f;if(!arguments.length||null==i||void 0==i)return[];if(Array.isArray(i))return i.map(function(t){return r?new e.Token(t.toLowerCase()):t.toLowerCase()});for(o=i.toString().toLowerCase().replace(/^\s+/,""),n=o.length-1;n>=0;n--)if(/\S/.test(o.charAt(n))){o=o.substring(0,n+1);break}for(a=[],s=o.length,c=0,l=0;c<=s;c++)if(u=o.charAt(c),m=c-l,u.match(/\s/)||c==s){if(m>0)for(p=t.segment(o.slice(l,c)).filter(function(e){return!!e}),f=l,n=0;n<p.length;n++)r?a.push(new e.Token(p[n],{position:[f,p[n].length],index:a.length})):a.push(p[n]),f+=p[n].length;l=c+1}return a},e.ja.stemmer=function(){return function(e){return e}}(),e.Pipeline.registerFunction(e.ja.stemmer,"stemmer-ja"),e.ja.wordCharacters="一二三四五六七八九十百千万億兆一-龠々〆ヵヶぁ-んァ-ヴーｱ-ﾝﾞa-zA-Zａ-ｚＡ-Ｚ0-9０-９",e.ja.trimmer=e.trimmerSupport.generateTrimmer(e.ja.wordCharacters),e.Pipeline.registerFunction(e.ja.trimmer,"trimmer-ja"),e.ja.stopWordFilter=e.generateStopWordFilter("これ それ あれ この その あの ここ そこ あそこ こちら どこ だれ なに なん 何 私 貴方 貴方方 我々 私達 あの人 あのかた 彼女 彼 です あります おります います は が の に を で え から まで より も どの と し それで しかし".split(" ")),e.Pipeline.registerFunction(e.ja.stopWordFilter,"stopWordFilter-ja"),e.jp=e.ja,e.Pipeline.registerFunction(e.jp.stemmer,"stemmer-jp"),e.Pipeline.registerFunction(e.jp.trimmer,"trimmer-jp"),e.Pipeline.registerFunction(e.jp.stopWordFilter,"stopWordFilter-jp")}});
|
||||
+1
@@ -0,0 +1 @@
|
||||
module.exports=require("./lunr.ja");
|
||||
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.kn=function(){this.pipeline.reset(),this.pipeline.add(e.kn.trimmer,e.kn.stopWordFilter,e.kn.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.kn.stemmer))},e.kn.wordCharacters="ಀ-಄ಅ-ಔಕ-ಹಾ-ೌ಼-ಽೕ-ೖೝ-ೞೠ-ೡೢ-ೣ೦-೯ೱ-ೳ",e.kn.trimmer=e.trimmerSupport.generateTrimmer(e.kn.wordCharacters),e.Pipeline.registerFunction(e.kn.trimmer,"trimmer-kn"),e.kn.stopWordFilter=e.generateStopWordFilter("ಮತ್ತು ಈ ಒಂದು ರಲ್ಲಿ ಹಾಗೂ ಎಂದು ಅಥವಾ ಇದು ರ ಅವರು ಎಂಬ ಮೇಲೆ ಅವರ ತನ್ನ ಆದರೆ ತಮ್ಮ ನಂತರ ಮೂಲಕ ಹೆಚ್ಚು ನ ಆ ಕೆಲವು ಅನೇಕ ಎರಡು ಹಾಗು ಪ್ರಮುಖ ಇದನ್ನು ಇದರ ಸುಮಾರು ಅದರ ಅದು ಮೊದಲ ಬಗ್ಗೆ ನಲ್ಲಿ ರಂದು ಇತರ ಅತ್ಯಂತ ಹೆಚ್ಚಿನ ಸಹ ಸಾಮಾನ್ಯವಾಗಿ ನೇ ಹಲವಾರು ಹೊಸ ದಿ ಕಡಿಮೆ ಯಾವುದೇ ಹೊಂದಿದೆ ದೊಡ್ಡ ಅನ್ನು ಇವರು ಪ್ರಕಾರ ಇದೆ ಮಾತ್ರ ಕೂಡ ಇಲ್ಲಿ ಎಲ್ಲಾ ವಿವಿಧ ಅದನ್ನು ಹಲವು ರಿಂದ ಕೇವಲ ದ ದಕ್ಷಿಣ ಗೆ ಅವನ ಅತಿ ನೆಯ ಬಹಳ ಕೆಲಸ ಎಲ್ಲ ಪ್ರತಿ ಇತ್ಯಾದಿ ಇವು ಬೇರೆ ಹೀಗೆ ನಡುವೆ ಇದಕ್ಕೆ ಎಸ್ ಇವರ ಮೊದಲು ಶ್ರೀ ಮಾಡುವ ಇದರಲ್ಲಿ ರೀತಿಯ ಮಾಡಿದ ಕಾಲ ಅಲ್ಲಿ ಮಾಡಲು ಅದೇ ಈಗ ಅವು ಗಳು ಎ ಎಂಬುದು ಅವನು ಅಂದರೆ ಅವರಿಗೆ ಇರುವ ವಿಶೇಷ ಮುಂದೆ ಅವುಗಳ ಮುಂತಾದ ಮೂಲ ಬಿ ಮೀ ಒಂದೇ ಇನ್ನೂ ಹೆಚ್ಚಾಗಿ ಮಾಡಿ ಅವರನ್ನು ಇದೇ ಯ ರೀತಿಯಲ್ಲಿ ಜೊತೆ ಅದರಲ್ಲಿ ಮಾಡಿದರು ನಡೆದ ಆಗ ಮತ್ತೆ ಪೂರ್ವ ಆತ ಬಂದ ಯಾವ ಒಟ್ಟು ಇತರೆ ಹಿಂದೆ ಪ್ರಮಾಣದ ಗಳನ್ನು ಕುರಿತು ಯು ಆದ್ದರಿಂದ ಅಲ್ಲದೆ ನಗರದ ಮೇಲಿನ ಏಕೆಂದರೆ ರಷ್ಟು ಎಂಬುದನ್ನು ಬಾರಿ ಎಂದರೆ ಹಿಂದಿನ ಆದರೂ ಆದ ಸಂಬಂಧಿಸಿದ ಮತ್ತೊಂದು ಸಿ ಆತನ ".split(" ")),e.kn.stemmer=function(){return function(e){return"function"==typeof e.update?e.update(function(e){return e}):e}}();var r=e.wordcut;r.init(),e.kn.tokenizer=function(t){if(!arguments.length||null==t||void 0==t)return[];if(Array.isArray(t))return t.map(function(r){return isLunr2?new e.Token(r.toLowerCase()):r.toLowerCase()});var n=t.toString().toLowerCase().replace(/^\s+/,"");return r.cut(n).split("|")},e.Pipeline.registerFunction(e.kn.stemmer,"stemmer-kn"),e.Pipeline.registerFunction(e.kn.stopWordFilter,"stopWordFilter-kn")}});
|
||||
+1
File diff suppressed because one or more lines are too long
@@ -0,0 +1 @@
|
||||
!function(e,t){"function"==typeof define&&define.amd?define(t):"object"==typeof exports?module.exports=t():t()(e.lunr)}(this,function(){return function(e){e.multiLanguage=function(){for(var t=Array.prototype.slice.call(arguments),i=t.join("-"),r="",n=[],s=[],p=0;p<t.length;++p)"en"==t[p]?(r+="\\w",n.unshift(e.stopWordFilter),n.push(e.stemmer),s.push(e.stemmer)):(r+=e[t[p]].wordCharacters,e[t[p]].stopWordFilter&&n.unshift(e[t[p]].stopWordFilter),e[t[p]].stemmer&&(n.push(e[t[p]].stemmer),s.push(e[t[p]].stemmer)));var o=e.trimmerSupport.generateTrimmer(r);return e.Pipeline.registerFunction(o,"lunr-multi-trimmer-"+i),n.unshift(o),function(){this.pipeline.reset(),this.pipeline.add.apply(this.pipeline,n),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add.apply(this.searchPipeline,s))}}}});
|
||||
+18
File diff suppressed because one or more lines are too long
+18
@@ -0,0 +1,18 @@
|
||||
/*!
|
||||
* Lunr languages, `Norwegian` language
|
||||
* https://github.com/MihaiValentin/lunr-languages
|
||||
*
|
||||
* Copyright 2014, Mihai Valentin
|
||||
* http://www.mozilla.org/MPL/
|
||||
*/
|
||||
/*!
|
||||
* based on
|
||||
* Snowball JavaScript Library v0.3
|
||||
* http://code.google.com/p/urim/
|
||||
* http://snowball.tartarus.org/
|
||||
*
|
||||
* Copyright 2010, Oleg Mazko
|
||||
* http://www.mozilla.org/MPL/
|
||||
*/
|
||||
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.no=function(){this.pipeline.reset(),this.pipeline.add(e.no.trimmer,e.no.stopWordFilter,e.no.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.no.stemmer))},e.no.wordCharacters="A-Za-zªºÀ-ÖØ-öø-ʸˠ-ˤᴀ-ᴥᴬ-ᵜᵢ-ᵥᵫ-ᵷᵹ-ᶾḀ-ỿⁱⁿₐ-ₜKÅℲⅎⅠ-ↈⱠ-ⱿꜢ-ꞇꞋ-ꞭꞰ-ꞷꟷ-ꟿꬰ-ꭚꭜ-ꭤff-stＡ-Ｚａ-ｚ",e.no.trimmer=e.trimmerSupport.generateTrimmer(e.no.wordCharacters),e.Pipeline.registerFunction(e.no.trimmer,"trimmer-no"),e.no.stemmer=function(){var r=e.stemmerSupport.Among,n=e.stemmerSupport.SnowballProgram,i=new function(){function e(){var e,r=w.cursor+3;if(a=w.limit,0<=r||r<=w.limit){for(s=r;;){if(e=w.cursor,w.in_grouping(d,97,248)){w.cursor=e;break}if(e>=w.limit)return;w.cursor=e+1}for(;!w.out_grouping(d,97,248);){if(w.cursor>=w.limit)return;w.cursor++}a=w.cursor,a<s&&(a=s)}}function i(){var e,r,n;if(w.cursor>=a&&(r=w.limit_backward,w.limit_backward=a,w.ket=w.cursor,e=w.find_among_b(m,29),w.limit_backward=r,e))switch(w.bra=w.cursor,e){case 1:w.slice_del();break;case 2:n=w.limit-w.cursor,w.in_grouping_b(c,98,122)?w.slice_del():(w.cursor=w.limit-n,w.eq_s_b(1,"k")&&w.out_grouping_b(d,97,248)&&w.slice_del());break;case 3:w.slice_from("er")}}function t(){var e,r=w.limit-w.cursor;w.cursor>=a&&(e=w.limit_backward,w.limit_backward=a,w.ket=w.cursor,w.find_among_b(u,2)?(w.bra=w.cursor,w.limit_backward=e,w.cursor=w.limit-r,w.cursor>w.limit_backward&&(w.cursor--,w.bra=w.cursor,w.slice_del())):w.limit_backward=e)}function o(){var e,r;w.cursor>=a&&(r=w.limit_backward,w.limit_backward=a,w.ket=w.cursor,e=w.find_among_b(l,11),e?(w.bra=w.cursor,w.limit_backward=r,1==e&&w.slice_del()):w.limit_backward=r)}var s,a,m=[new r("a",-1,1),new r("e",-1,1),new r("ede",1,1),new r("ande",1,1),new r("ende",1,1),new r("ane",1,1),new r("ene",1,1),new r("hetene",6,1),new r("erte",1,3),new r("en",-1,1),new r("heten",9,1),new r("ar",-1,1),new r("er",-1,1),new r("heter",12,1),new r("s",-1,2),new r("as",14,1),new r("es",14,1),new r("edes",16,1),new r("endes",16,1),new r("enes",16,1),new r("hetenes",19,1),new r("ens",14,1),new r("hetens",21,1),new r("ers",14,1),new r("ets",14,1),new r("et",-1,1),new r("het",25,1),new r("ert",-1,3),new r("ast",-1,1)],u=[new r("dt",-1,-1),new r("vt",-1,-1)],l=[new r("leg",-1,1),new r("eleg",0,1),new r("ig",-1,1),new r("eig",2,1),new r("lig",2,1),new r("elig",4,1),new r("els",-1,1),new r("lov",-1,1),new r("elov",7,1),new r("slov",7,1),new r("hetslov",9,1)],d=[17,65,16,1,0,0,0,0,0,0,0,0,0,0,0,0,48,0,128],c=[119,125,149,1],w=new n;this.setCurrent=function(e){w.setCurrent(e)},this.getCurrent=function(){return w.getCurrent()},this.stem=function(){var r=w.cursor;return e(),w.limit_backward=r,w.cursor=w.limit,i(),w.cursor=w.limit,t(),w.cursor=w.limit,o(),!0}};return function(e){return"function"==typeof e.update?e.update(function(e){return i.setCurrent(e),i.stem(),i.getCurrent()}):(i.setCurrent(e),i.stem(),i.getCurrent())}}(),e.Pipeline.registerFunction(e.no.stemmer,"stemmer-no"),e.no.stopWordFilter=e.generateStopWordFilter("alle at av bare begge ble blei bli blir blitt både båe da de deg dei deim deira deires dem den denne der dere deres det dette di din disse ditt du dykk dykkar då eg ein eit eitt eller elles en enn er et ett etter for fordi fra før ha hadde han hans har hennar henne hennes her hjå ho hoe honom hoss hossen hun hva hvem hver hvilke hvilken hvis hvor hvordan hvorfor i ikke ikkje ikkje ingen ingi inkje inn inni ja jeg kan kom korleis korso kun kunne kva kvar kvarhelst kven kvi kvifor man mange me med medan meg meget mellom men mi min mine mitt mot mykje ned no noe noen noka noko nokon nokor nokre nå når og også om opp oss over på samme seg selv si si sia sidan siden sin sine sitt sjøl skal skulle slik so som som somme somt så sånn til um upp ut uten var vart varte ved vere verte vi vil ville vore vors vort vår være være vært å".split(" ")),e.Pipeline.registerFunction(e.no.stopWordFilter,"stopWordFilter-no")}});
|
||||
+18
File diff suppressed because one or more lines are too long
+18
File diff suppressed because one or more lines are too long
+18
File diff suppressed because one or more lines are too long
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.sa=function(){this.pipeline.reset(),this.pipeline.add(e.sa.trimmer,e.sa.stopWordFilter,e.sa.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.sa.stemmer))},e.sa.wordCharacters="ऀ-ःऄ-एऐ-टठ-यर-िी-ॏॐ-य़ॠ-९॰-ॿ꣠-꣱ꣲ-ꣷ꣸-ꣻ꣼-ꣽꣾ-ꣿᆰ0-ᆰ9",e.sa.trimmer=e.trimmerSupport.generateTrimmer(e.sa.wordCharacters),e.Pipeline.registerFunction(e.sa.trimmer,"trimmer-sa"),e.sa.stopWordFilter=e.generateStopWordFilter('तथा अयम् एकम् इत्यस्मिन् तथा तत् वा अयम् इत्यस्य ते आहूत उपरि तेषाम् किन्तु तेषाम् तदा इत्यनेन अधिकः इत्यस्य तत् केचन बहवः द्वि तथा महत्वपूर्णः अयम् अस्य विषये अयं अस्ति तत् प्रथमः विषये इत्युपरि इत्युपरि इतर अधिकतमः अधिकः अपि सामान्यतया ठ इतरेतर नूतनम् द न्यूनम् कश्चित् वा विशालः द सः अस्ति तदनुसारम् तत्र अस्ति केवलम् अपि अत्र सर्वे विविधाः तत् बहवः यतः इदानीम् द दक्षिण इत्यस्मै तस्य उपरि नथ अतीव कार्यम् सर्वे एकैकम् इत्यादि। एते सन्ति उत इत्थम् मध्ये एतदर्थं . स कस्य प्रथमः श्री. करोति अस्मिन् प्रकारः निर्मिता कालः तत्र कर्तुं समान अधुना ते सन्ति स एकः अस्ति सः अर्थात् तेषां कृते . स्थितम् विशेषः अग्रिम तेषाम् समान स्रोतः ख म समान इदानीमपि अधिकतया करोतु ते समान इत्यस्य वीथी सह यस्मिन् कृतवान् धृतः तदा पुनः पूर्वं सः आगतः किम् कुल इतर पुरा मात्रा स विषये उ अतएव अपि नगरस्य उपरि यतः प्रतिशतं कतरः कालः साधनानि भूत तथापि जात सम्बन्धि अन्यत् ग अतः अस्माकं स्वकीयाः अस्माकं इदानीं अन्तः इत्यादयः भवन्तः इत्यादयः एते एताः तस्य अस्य इदम् एते तेषां तेषां तेषां तान् तेषां तेषां तेषां समानः सः एकः च तादृशाः बहवः अन्ये च वदन्ति यत् कियत् कस्मै कस्मै यस्मै यस्मै यस्मै यस्मै न अतिनीचः किन्तु प्रथमं सम्पूर्णतया ततः चिरकालानन्तरं पुस्तकं सम्पूर्णतया अन्तः किन्तु अत्र वा इह इव श्रद्धाय अवशिष्यते परन्तु अन्ये वर्गाः सन्ति ते सन्ति शक्नुवन्ति सर्वे मिलित्वा सर्वे एकत्र"'.split(" ")),e.sa.stemmer=function(){return function(e){return"function"==typeof e.update?e.update(function(e){return e}):e}}();var r=e.wordcut;r.init(),e.sa.tokenizer=function(t){if(!arguments.length||null==t||void 0==t)return[];if(Array.isArray(t))return t.map(function(r){return isLunr2?new e.Token(r.toLowerCase()):r.toLowerCase()});var i=t.toString().toLowerCase().replace(/^\s+/,"");return r.cut(i).split("|")},e.Pipeline.registerFunction(e.sa.stemmer,"stemmer-sa"),e.Pipeline.registerFunction(e.sa.stopWordFilter,"stopWordFilter-sa")}});
|
||||
@@ -0,0 +1 @@
|
||||
!function(r,t){"function"==typeof define&&define.amd?define(t):"object"==typeof exports?module.exports=t():t()(r.lunr)}(this,function(){return function(r){r.stemmerSupport={Among:function(r,t,i,s){if(this.toCharArray=function(r){for(var t=r.length,i=new Array(t),s=0;s<t;s++)i[s]=r.charCodeAt(s);return i},!r&&""!=r||!t&&0!=t||!i)throw"Bad Among initialisation: s:"+r+", substring_i: "+t+", result: "+i;this.s_size=r.length,this.s=this.toCharArray(r),this.substring_i=t,this.result=i,this.method=s},SnowballProgram:function(){var r;return{bra:0,ket:0,limit:0,cursor:0,limit_backward:0,setCurrent:function(t){r=t,this.cursor=0,this.limit=t.length,this.limit_backward=0,this.bra=this.cursor,this.ket=this.limit},getCurrent:function(){var t=r;return r=null,t},in_grouping:function(t,i,s){if(this.cursor<this.limit){var e=r.charCodeAt(this.cursor);if(e<=s&&e>=i&&(e-=i,t[e>>3]&1<<(7&e)))return this.cursor++,!0}return!1},in_grouping_b:function(t,i,s){if(this.cursor>this.limit_backward){var e=r.charCodeAt(this.cursor-1);if(e<=s&&e>=i&&(e-=i,t[e>>3]&1<<(7&e)))return this.cursor--,!0}return!1},out_grouping:function(t,i,s){if(this.cursor<this.limit){var e=r.charCodeAt(this.cursor);if(e>s||e<i)return this.cursor++,!0;if(e-=i,!(t[e>>3]&1<<(7&e)))return this.cursor++,!0}return!1},out_grouping_b:function(t,i,s){if(this.cursor>this.limit_backward){var e=r.charCodeAt(this.cursor-1);if(e>s||e<i)return this.cursor--,!0;if(e-=i,!(t[e>>3]&1<<(7&e)))return this.cursor--,!0}return!1},eq_s:function(t,i){if(this.limit-this.cursor<t)return!1;for(var s=0;s<t;s++)if(r.charCodeAt(this.cursor+s)!=i.charCodeAt(s))return!1;return this.cursor+=t,!0},eq_s_b:function(t,i){if(this.cursor-this.limit_backward<t)return!1;for(var s=0;s<t;s++)if(r.charCodeAt(this.cursor-t+s)!=i.charCodeAt(s))return!1;return this.cursor-=t,!0},find_among:function(t,i){for(var s=0,e=i,n=this.cursor,u=this.limit,o=0,h=0,c=!1;;){for(var a=s+(e-s>>1),f=0,l=o<h?o:h,_=t[a],m=l;m<_.s_size;m++){if(n+l==u){f=-1;break}if(f=r.charCodeAt(n+l)-_.s[m])break;l++}if(f<0?(e=a,h=l):(s=a,o=l),e-s<=1){if(s>0||e==s||c)break;c=!0}}for(;;){var _=t[s];if(o>=_.s_size){if(this.cursor=n+_.s_size,!_.method)return _.result;var b=_.method();if(this.cursor=n+_.s_size,b)return _.result}if((s=_.substring_i)<0)return 0}},find_among_b:function(t,i){for(var s=0,e=i,n=this.cursor,u=this.limit_backward,o=0,h=0,c=!1;;){for(var a=s+(e-s>>1),f=0,l=o<h?o:h,_=t[a],m=_.s_size-1-l;m>=0;m--){if(n-l==u){f=-1;break}if(f=r.charCodeAt(n-1-l)-_.s[m])break;l++}if(f<0?(e=a,h=l):(s=a,o=l),e-s<=1){if(s>0||e==s||c)break;c=!0}}for(;;){var _=t[s];if(o>=_.s_size){if(this.cursor=n-_.s_size,!_.method)return _.result;var b=_.method();if(this.cursor=n-_.s_size,b)return _.result}if((s=_.substring_i)<0)return 0}},replace_s:function(t,i,s){var e=s.length-(i-t),n=r.substring(0,t),u=r.substring(i);return r=n+s+u,this.limit+=e,this.cursor>=i?this.cursor+=e:this.cursor>t&&(this.cursor=t),e},slice_check:function(){if(this.bra<0||this.bra>this.ket||this.ket>this.limit||this.limit>r.length)throw"faulty slice operation"},slice_from:function(r){this.slice_check(),this.replace_s(this.bra,this.ket,r)},slice_del:function(){this.slice_from("")},insert:function(r,t,i){var s=this.replace_s(r,t,i);r<=this.bra&&(this.bra+=s),r<=this.ket&&(this.ket+=s)},slice_to:function(){return this.slice_check(),r.substring(this.bra,this.ket)},eq_v_b:function(r){return this.eq_s_b(r.length,r)}}}},r.trimmerSupport={generateTrimmer:function(r){var t=new RegExp("^[^"+r+"]+"),i=new RegExp("[^"+r+"]+$");return function(r){return"function"==typeof r.update?r.update(function(r){return r.replace(t,"").replace(i,"")}):r.replace(t,"").replace(i,"")}}}}});
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
/*!
|
||||
* Lunr languages, `Swedish` language
|
||||
* https://github.com/MihaiValentin/lunr-languages
|
||||
*
|
||||
* Copyright 2014, Mihai Valentin
|
||||
* http://www.mozilla.org/MPL/
|
||||
*/
|
||||
/*!
|
||||
* based on
|
||||
* Snowball JavaScript Library v0.3
|
||||
* http://code.google.com/p/urim/
|
||||
* http://snowball.tartarus.org/
|
||||
*
|
||||
* Copyright 2010, Oleg Mazko
|
||||
* http://www.mozilla.org/MPL/
|
||||
*/
|
||||
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.sv=function(){this.pipeline.reset(),this.pipeline.add(e.sv.trimmer,e.sv.stopWordFilter,e.sv.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.sv.stemmer))},e.sv.wordCharacters="A-Za-zªºÀ-ÖØ-öø-ʸˠ-ˤᴀ-ᴥᴬ-ᵜᵢ-ᵥᵫ-ᵷᵹ-ᶾḀ-ỿⁱⁿₐ-ₜKÅℲⅎⅠ-ↈⱠ-ⱿꜢ-ꞇꞋ-ꞭꞰ-ꞷꟷ-ꟿꬰ-ꭚꭜ-ꭤff-stＡ-Ｚａ-ｚ",e.sv.trimmer=e.trimmerSupport.generateTrimmer(e.sv.wordCharacters),e.Pipeline.registerFunction(e.sv.trimmer,"trimmer-sv"),e.sv.stemmer=function(){var r=e.stemmerSupport.Among,n=e.stemmerSupport.SnowballProgram,t=new function(){function e(){var e,r=w.cursor+3;if(o=w.limit,0<=r||r<=w.limit){for(a=r;;){if(e=w.cursor,w.in_grouping(l,97,246)){w.cursor=e;break}if(w.cursor=e,w.cursor>=w.limit)return;w.cursor++}for(;!w.out_grouping(l,97,246);){if(w.cursor>=w.limit)return;w.cursor++}o=w.cursor,o<a&&(o=a)}}function t(){var e,r=w.limit_backward;if(w.cursor>=o&&(w.limit_backward=o,w.cursor=w.limit,w.ket=w.cursor,e=w.find_among_b(u,37),w.limit_backward=r,e))switch(w.bra=w.cursor,e){case 1:w.slice_del();break;case 2:w.in_grouping_b(d,98,121)&&w.slice_del()}}function i(){var e=w.limit_backward;w.cursor>=o&&(w.limit_backward=o,w.cursor=w.limit,w.find_among_b(c,7)&&(w.cursor=w.limit,w.ket=w.cursor,w.cursor>w.limit_backward&&(w.bra=--w.cursor,w.slice_del())),w.limit_backward=e)}function s(){var e,r;if(w.cursor>=o){if(r=w.limit_backward,w.limit_backward=o,w.cursor=w.limit,w.ket=w.cursor,e=w.find_among_b(m,5))switch(w.bra=w.cursor,e){case 1:w.slice_del();break;case 2:w.slice_from("lös");break;case 3:w.slice_from("full")}w.limit_backward=r}}var a,o,u=[new r("a",-1,1),new r("arna",0,1),new r("erna",0,1),new r("heterna",2,1),new r("orna",0,1),new r("ad",-1,1),new r("e",-1,1),new r("ade",6,1),new r("ande",6,1),new r("arne",6,1),new r("are",6,1),new r("aste",6,1),new r("en",-1,1),new r("anden",12,1),new r("aren",12,1),new r("heten",12,1),new r("ern",-1,1),new r("ar",-1,1),new r("er",-1,1),new r("heter",18,1),new r("or",-1,1),new r("s",-1,2),new r("as",21,1),new r("arnas",22,1),new r("ernas",22,1),new r("ornas",22,1),new r("es",21,1),new r("ades",26,1),new r("andes",26,1),new r("ens",21,1),new r("arens",29,1),new r("hetens",29,1),new r("erns",21,1),new r("at",-1,1),new r("andet",-1,1),new r("het",-1,1),new r("ast",-1,1)],c=[new r("dd",-1,-1),new r("gd",-1,-1),new r("nn",-1,-1),new r("dt",-1,-1),new r("gt",-1,-1),new r("kt",-1,-1),new r("tt",-1,-1)],m=[new r("ig",-1,1),new r("lig",0,1),new r("els",-1,1),new r("fullt",-1,3),new r("löst",-1,2)],l=[17,65,16,1,0,0,0,0,0,0,0,0,0,0,0,0,24,0,32],d=[119,127,149],w=new n;this.setCurrent=function(e){w.setCurrent(e)},this.getCurrent=function(){return w.getCurrent()},this.stem=function(){var r=w.cursor;return e(),w.limit_backward=r,w.cursor=w.limit,t(),w.cursor=w.limit,i(),w.cursor=w.limit,s(),!0}};return function(e){return"function"==typeof e.update?e.update(function(e){return t.setCurrent(e),t.stem(),t.getCurrent()}):(t.setCurrent(e),t.stem(),t.getCurrent())}}(),e.Pipeline.registerFunction(e.sv.stemmer,"stemmer-sv"),e.sv.stopWordFilter=e.generateStopWordFilter("alla allt att av blev bli blir blivit de dem den denna deras dess dessa det detta dig din dina ditt du där då efter ej eller en er era ert ett från för ha hade han hans har henne hennes hon honom hur här i icke ingen inom inte jag ju kan kunde man med mellan men mig min mina mitt mot mycket ni nu när någon något några och om oss på samma sedan sig sin sina sitta själv skulle som så sådan sådana sådant till under upp ut utan vad var vara varför varit varje vars vart vem vi vid vilka vilkas vilken vilket vår våra vårt än är åt över".split(" ")),e.Pipeline.registerFunction(e.sv.stopWordFilter,"stopWordFilter-sv")}});
|
||||
+1
@@ -0,0 +1 @@
|
||||
!function(e,t){"function"==typeof define&&define.amd?define(t):"object"==typeof exports?module.exports=t():t()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.ta=function(){this.pipeline.reset(),this.pipeline.add(e.ta.trimmer,e.ta.stopWordFilter,e.ta.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.ta.stemmer))},e.ta.wordCharacters="-உஊ-ஏஐ-ஙச-ட-னப-யர-ஹ-ிீ-ொ-ௐ---௩௪-௯௰-௹௺-a-zA-Zａ-ｚＡ-Ｚ0-9０-９",e.ta.trimmer=e.trimmerSupport.generateTrimmer(e.ta.wordCharacters),e.Pipeline.registerFunction(e.ta.trimmer,"trimmer-ta"),e.ta.stopWordFilter=e.generateStopWordFilter("அங்கு அங்கே அது அதை அந்த அவர் அவர்கள் அவள் அவன் அவை ஆக ஆகவே ஆகையால் ஆதலால் ஆதலினால் ஆனாலும் ஆனால் இங்கு இங்கே இது இதை இந்த இப்படி இவர் இவர்கள் இவள் இவன் இவை இவ்வளவு உனக்கு உனது உன் உன்னால் எங்கு எங்கே எது எதை எந்த எப்படி எவர் எவர்கள் எவள் எவன் எவை எவ்வளவு எனக்கு எனது எனவே என் என்ன என்னால் ஏது ஏன் தனது தன்னால் தானே தான் நாங்கள் நாம் நான் நீ நீங்கள்".split(" ")),e.ta.stemmer=function(){return function(e){return"function"==typeof e.update?e.update(function(e){return e}):e}}();var t=e.wordcut;t.init(),e.ta.tokenizer=function(r){if(!arguments.length||null==r||void 0==r)return[];if(Array.isArray(r))return r.map(function(t){return isLunr2?new e.Token(t.toLowerCase()):t.toLowerCase()});var i=r.toString().toLowerCase().replace(/^\s+/,"");return t.cut(i).split("|")},e.Pipeline.registerFunction(e.ta.stemmer,"stemmer-ta"),e.Pipeline.registerFunction(e.ta.stopWordFilter,"stopWordFilter-ta")}});
|
||||
+1
@@ -0,0 +1 @@
|
||||
!function(e,t){"function"==typeof define&&define.amd?define(t):"object"==typeof exports?module.exports=t():t()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.te=function(){this.pipeline.reset(),this.pipeline.add(e.te.trimmer,e.te.stopWordFilter,e.te.stemmer),this.searchPipeline&&(this.searchPipeline.reset(),this.searchPipeline.add(e.te.stemmer))},e.te.wordCharacters="ఀ-ఄఅ-ఔక-హా-ౌౕ-ౖౘ-ౚౠ-ౡౢ-ౣ౦-౯౸-౿఼ఽ్ౝ౷",e.te.trimmer=e.trimmerSupport.generateTrimmer(e.te.wordCharacters),e.Pipeline.registerFunction(e.te.trimmer,"trimmer-te"),e.te.stopWordFilter=e.generateStopWordFilter("అందరూ అందుబాటులో అడగండి అడగడం అడ్డంగా అనుగుణంగా అనుమతించు అనుమతిస్తుంది అయితే ఇప్పటికే ఉన్నారు ఎక్కడైనా ఎప్పుడు ఎవరైనా ఎవరో ఏ ఏదైనా ఏమైనప్పటికి ఒక ఒకరు కనిపిస్తాయి కాదు కూడా గా గురించి చుట్టూ చేయగలిగింది తగిన తర్వాత దాదాపు దూరంగా నిజంగా పై ప్రకారం ప్రక్కన మధ్య మరియు మరొక మళ్ళీ మాత్రమే మెచ్చుకో వద్ద వెంట వేరుగా వ్యతిరేకంగా సంబంధం".split(" ")),e.te.stemmer=function(){return function(e){return"function"==typeof e.update?e.update(function(e){return e}):e}}();var t=e.wordcut;t.init(),e.te.tokenizer=function(r){if(!arguments.length||null==r||void 0==r)return[];if(Array.isArray(r))return r.map(function(t){return isLunr2?new e.Token(t.toLowerCase()):t.toLowerCase()});var i=r.toString().toLowerCase().replace(/^\s+/,"");return t.cut(i).split("|")},e.Pipeline.registerFunction(e.te.stemmer,"stemmer-te"),e.Pipeline.registerFunction(e.te.stopWordFilter,"stopWordFilter-te")}});
|
||||
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");var r="2"==e.version[0];e.th=function(){this.pipeline.reset(),this.pipeline.add(e.th.trimmer),r?this.tokenizer=e.th.tokenizer:(e.tokenizer&&(e.tokenizer=e.th.tokenizer),this.tokenizerFn&&(this.tokenizerFn=e.th.tokenizer))},e.th.wordCharacters="[-]",e.th.trimmer=e.trimmerSupport.generateTrimmer(e.th.wordCharacters),e.Pipeline.registerFunction(e.th.trimmer,"trimmer-th");var t=e.wordcut;t.init(),e.th.tokenizer=function(i){if(!arguments.length||null==i||void 0==i)return[];if(Array.isArray(i))return i.map(function(t){return r?new e.Token(t):t});var n=i.toString().replace(/^\s+/,"");return t.cut(n).split("|")}}});
|
||||
+18
File diff suppressed because one or more lines are too long
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r():r()(e.lunr)}(this,function(){return function(e){if(void 0===e)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===e.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");e.vi=function(){this.pipeline.reset(),this.pipeline.add(e.vi.stopWordFilter,e.vi.trimmer)},e.vi.wordCharacters="[A-Za-ẓ̀͐́͑̉̃̓ÂâÊêÔôĂ-ăĐ-đƠ-ơƯ-ư]",e.vi.trimmer=e.trimmerSupport.generateTrimmer(e.vi.wordCharacters),e.Pipeline.registerFunction(e.vi.trimmer,"trimmer-vi"),e.vi.stopWordFilter=e.generateStopWordFilter("là cái nhưng mà".split(" "))}});
|
||||
+1
@@ -0,0 +1 @@
|
||||
!function(e,r){"function"==typeof define&&define.amd?define(r):"object"==typeof exports?module.exports=r(require("@node-rs/jieba")):r()(e.lunr)}(this,function(e){return function(r,t){if(void 0===r)throw new Error("Lunr is not present. Please include / require Lunr before this script.");if(void 0===r.stemmerSupport)throw new Error("Lunr stemmer support is not present. Please include / require Lunr stemmer support before this script.");var i="2"==r.version[0];r.zh=function(){this.pipeline.reset(),this.pipeline.add(r.zh.trimmer,r.zh.stopWordFilter,r.zh.stemmer),i?this.tokenizer=r.zh.tokenizer:(r.tokenizer&&(r.tokenizer=r.zh.tokenizer),this.tokenizerFn&&(this.tokenizerFn=r.zh.tokenizer))},r.zh.tokenizer=function(n){if(!arguments.length||null==n||void 0==n)return[];if(Array.isArray(n))return n.map(function(e){return i?new r.Token(e.toLowerCase()):e.toLowerCase()});t&&e.load(t);var o=n.toString().trim().toLowerCase(),s=[];e.cut(o,!0).forEach(function(e){s=s.concat(e.split(" "))}),s=s.filter(function(e){return!!e});var u=0;return s.map(function(e,t){if(i){var n=o.indexOf(e,u),s={};return s.position=[n,e.length],s.index=t,u=n,new r.Token(e,s)}return e})},r.zh.wordCharacters="\\w一-龥",r.zh.trimmer=r.trimmerSupport.generateTrimmer(r.zh.wordCharacters),r.Pipeline.registerFunction(r.zh.trimmer,"trimmer-zh"),r.zh.stemmer=function(){return function(e){return e}}(),r.Pipeline.registerFunction(r.zh.stemmer,"stemmer-zh"),r.zh.stopWordFilter=r.generateStopWordFilter("的 一 不 在 人 有 是 为 為 以 于 於 上 他 而 后 後 之 来 來 及 了 因 下 可 到 由 这 這 与 與 也 此 但 并 並 个 個 其 已 无 無 小 我 们 們 起 最 再 今 去 好 只 又 或 很 亦 某 把 那 你 乃 它 吧 被 比 别 趁 当 當 从 從 得 打 凡 儿 兒 尔 爾 该 該 各 给 給 跟 和 何 还 還 即 几 幾 既 看 据 據 距 靠 啦 另 么 麽 每 嘛 拿 哪 您 凭 憑 且 却 卻 让 讓 仍 啥 如 若 使 谁 誰 虽 雖 随 隨 同 所 她 哇 嗡 往 些 向 沿 哟 喲 用 咱 则 則 怎 曾 至 致 着 著 诸 諸 自".split(" ")),r.Pipeline.registerFunction(r.zh.stopWordFilter,"stopWordFilter-zh")}});
|
||||
@@ -0,0 +1,206 @@
|
||||
/**
|
||||
* export the module via AMD, CommonJS or as a browser global
|
||||
* Export code from https://github.com/umdjs/umd/blob/master/returnExports.js
|
||||
*/
|
||||
;(function (root, factory) {
|
||||
if (typeof define === 'function' && define.amd) {
|
||||
// AMD. Register as an anonymous module.
|
||||
define(factory)
|
||||
} else if (typeof exports === 'object') {
|
||||
/**
|
||||
* Node. Does not work with strict CommonJS, but
|
||||
* only CommonJS-like environments that support module.exports,
|
||||
* like Node.
|
||||
*/
|
||||
module.exports = factory()
|
||||
} else {
|
||||
// Browser globals (root is window)
|
||||
factory()(root.lunr);
|
||||
}
|
||||
}(this, function () {
|
||||
/**
|
||||
* Just return a value to define the module export.
|
||||
* This example returns an object, but the module
|
||||
* can return a function as the exported value.
|
||||
*/
|
||||
|
||||
return function(lunr) {
|
||||
// TinySegmenter 0.1 -- Super compact Japanese tokenizer in Javascript
|
||||
// (c) 2008 Taku Kudo <[email protected]>
|
||||
// TinySegmenter is freely distributable under the terms of a new BSD licence.
|
||||
// For details, see http://chasen.org/~taku/software/TinySegmenter/LICENCE.txt
|
||||
|
||||
function TinySegmenter() {
|
||||
var patterns = {
|
||||
"[一二三四五六七八九十百千万億兆]":"M",
|
||||
"[一-龠々〆ヵヶ]":"H",
|
||||
"[ぁ-ん]":"I",
|
||||
"[ァ-ヴーｱ-ﾝﾞｰ]":"K",
|
||||
"[a-zA-Zａ-ｚＡ-Ｚ]":"A",
|
||||
"[0-9０-９]":"N"
|
||||
}
|
||||
this.chartype_ = [];
|
||||
for (var i in patterns) {
|
||||
var regexp = new RegExp(i);
|
||||
this.chartype_.push([regexp, patterns[i]]);
|
||||
}
|
||||
|
||||
this.BIAS__ = -332
|
||||
this.BC1__ = {"HH":6,"II":2461,"KH":406,"OH":-1378};
|
||||
this.BC2__ = {"AA":-3267,"AI":2744,"AN":-878,"HH":-4070,"HM":-1711,"HN":4012,"HO":3761,"IA":1327,"IH":-1184,"II":-1332,"IK":1721,"IO":5492,"KI":3831,"KK":-8741,"MH":-3132,"MK":3334,"OO":-2920};
|
||||
this.BC3__ = {"HH":996,"HI":626,"HK":-721,"HN":-1307,"HO":-836,"IH":-301,"KK":2762,"MK":1079,"MM":4034,"OA":-1652,"OH":266};
|
||||
this.BP1__ = {"BB":295,"OB":304,"OO":-125,"UB":352};
|
||||
this.BP2__ = {"BO":60,"OO":-1762};
|
||||
this.BQ1__ = {"BHH":1150,"BHM":1521,"BII":-1158,"BIM":886,"BMH":1208,"BNH":449,"BOH":-91,"BOO":-2597,"OHI":451,"OIH":-296,"OKA":1851,"OKH":-1020,"OKK":904,"OOO":2965};
|
||||
this.BQ2__ = {"BHH":118,"BHI":-1159,"BHM":466,"BIH":-919,"BKK":-1720,"BKO":864,"OHH":-1139,"OHM":-181,"OIH":153,"UHI":-1146};
|
||||
this.BQ3__ = {"BHH":-792,"BHI":2664,"BII":-299,"BKI":419,"BMH":937,"BMM":8335,"BNN":998,"BOH":775,"OHH":2174,"OHM":439,"OII":280,"OKH":1798,"OKI":-793,"OKO":-2242,"OMH":-2402,"OOO":11699};
|
||||
this.BQ4__ = {"BHH":-3895,"BIH":3761,"BII":-4654,"BIK":1348,"BKK":-1806,"BMI":-3385,"BOO":-12396,"OAH":926,"OHH":266,"OHK":-2036,"ONN":-973};
|
||||
this.BW1__ = {",と":660,",同":727,"B1あ":1404,"B1同":542,"、と":660,"、同":727,"」と":1682,"あっ":1505,"いう":1743,"いっ":-2055,"いる":672,"うし":-4817,"うん":665,"から":3472,"がら":600,"こう":-790,"こと":2083,"こん":-1262,"さら":-4143,"さん":4573,"した":2641,"して":1104,"すで":-3399,"そこ":1977,"それ":-871,"たち":1122,"ため":601,"った":3463,"つい":-802,"てい":805,"てき":1249,"でき":1127,"です":3445,"では":844,"とい":-4915,"とみ":1922,"どこ":3887,"ない":5713,"なっ":3015,"など":7379,"なん":-1113,"にし":2468,"には":1498,"にも":1671,"に対":-912,"の一":-501,"の中":741,"ませ":2448,"まで":1711,"まま":2600,"まる":-2155,"やむ":-1947,"よっ":-2565,"れた":2369,"れで":-913,"をし":1860,"を見":731,"亡く":-1886,"京都":2558,"取り":-2784,"大き":-2604,"大阪":1497,"平方":-2314,"引き":-1336,"日本":-195,"本当":-2423,"毎日":-2113,"目指":-724,"Ｂ１あ":1404,"Ｂ１同":542,"｣と":1682};
|
||||
this.BW2__ = {"..":-11822,"11":-669,"――":-5730,"−−":-13175,"いう":-1609,"うか":2490,"かし":-1350,"かも":-602,"から":-7194,"かれ":4612,"がい":853,"がら":-3198,"きた":1941,"くな":-1597,"こと":-8392,"この":-4193,"させ":4533,"され":13168,"さん":-3977,"しい":-1819,"しか":-545,"した":5078,"して":972,"しな":939,"その":-3744,"たい":-1253,"たた":-662,"ただ":-3857,"たち":-786,"たと":1224,"たは":-939,"った":4589,"って":1647,"っと":-2094,"てい":6144,"てき":3640,"てく":2551,"ては":-3110,"ても":-3065,"でい":2666,"でき":-1528,"でし":-3828,"です":-4761,"でも":-4203,"とい":1890,"とこ":-1746,"とと":-2279,"との":720,"とみ":5168,"とも":-3941,"ない":-2488,"なが":-1313,"など":-6509,"なの":2614,"なん":3099,"にお":-1615,"にし":2748,"にな":2454,"によ":-7236,"に対":-14943,"に従":-4688,"に関":-11388,"のか":2093,"ので":-7059,"のに":-6041,"のの":-6125,"はい":1073,"はが":-1033,"はず":-2532,"ばれ":1813,"まし":-1316,"まで":-6621,"まれ":5409,"めて":-3153,"もい":2230,"もの":-10713,"らか":-944,"らし":-1611,"らに":-1897,"りし":651,"りま":1620,"れた":4270,"れて":849,"れば":4114,"ろう":6067,"われ":7901,"を通":-11877,"んだ":728,"んな":-4115,"一人":602,"一方":-1375,"一日":970,"一部":-1051,"上が":-4479,"会社":-1116,"出て":2163,"分の":-7758,"同党":970,"同日":-913,"大阪":-2471,"委員":-1250,"少な":-1050,"年度":-8669,"年間":-1626,"府県":-2363,"手権":-1982,"新聞":-4066,"日新":-722,"日本":-7068,"日米":3372,"曜日":-601,"朝鮮":-2355,"本人":-2697,"東京":-1543,"然と":-1384,"社会":-1276,"立て":-990,"第に":-1612,"米国":-4268,"１１":-669};
|
||||
this.BW3__ = {"あた":-2194,"あり":719,"ある":3846,"い.":-1185,"い。":-1185,"いい":5308,"いえ":2079,"いく":3029,"いた":2056,"いっ":1883,"いる":5600,"いわ":1527,"うち":1117,"うと":4798,"えと":1454,"か.":2857,"か。":2857,"かけ":-743,"かっ":-4098,"かに":-669,"から":6520,"かり":-2670,"が,":1816,"が、":1816,"がき":-4855,"がけ":-1127,"がっ":-913,"がら":-4977,"がり":-2064,"きた":1645,"けど":1374,"こと":7397,"この":1542,"ころ":-2757,"さい":-714,"さを":976,"し,":1557,"し、":1557,"しい":-3714,"した":3562,"して":1449,"しな":2608,"しま":1200,"す.":-1310,"す。":-1310,"する":6521,"ず,":3426,"ず、":3426,"ずに":841,"そう":428,"た.":8875,"た。":8875,"たい":-594,"たの":812,"たり":-1183,"たる":-853,"だ.":4098,"だ。":4098,"だっ":1004,"った":-4748,"って":300,"てい":6240,"てお":855,"ても":302,"です":1437,"でに":-1482,"では":2295,"とう":-1387,"とし":2266,"との":541,"とも":-3543,"どう":4664,"ない":1796,"なく":-903,"など":2135,"に,":-1021,"に、":-1021,"にし":1771,"にな":1906,"には":2644,"の,":-724,"の、":-724,"の子":-1000,"は,":1337,"は、":1337,"べき":2181,"まし":1113,"ます":6943,"まっ":-1549,"まで":6154,"まれ":-793,"らし":1479,"られ":6820,"るる":3818,"れ,":854,"れ、":854,"れた":1850,"れて":1375,"れば":-3246,"れる":1091,"われ":-605,"んだ":606,"んで":798,"カ月":990,"会議":860,"入り":1232,"大会":2217,"始め":1681,"市":965,"新聞":-5055,"日,":974,"日、":974,"社会":2024,"ｶ月":990};
|
||||
this.TC1__ = {"AAA":1093,"HHH":1029,"HHM":580,"HII":998,"HOH":-390,"HOM":-331,"IHI":1169,"IOH":-142,"IOI":-1015,"IOM":467,"MMH":187,"OOI":-1832};
|
||||
this.TC2__ = {"HHO":2088,"HII":-1023,"HMM":-1154,"IHI":-1965,"KKH":703,"OII":-2649};
|
||||
this.TC3__ = {"AAA":-294,"HHH":346,"HHI":-341,"HII":-1088,"HIK":731,"HOH":-1486,"IHH":128,"IHI":-3041,"IHO":-1935,"IIH":-825,"IIM":-1035,"IOI":-542,"KHH":-1216,"KKA":491,"KKH":-1217,"KOK":-1009,"MHH":-2694,"MHM":-457,"MHO":123,"MMH":-471,"NNH":-1689,"NNO":662,"OHO":-3393};
|
||||
this.TC4__ = {"HHH":-203,"HHI":1344,"HHK":365,"HHM":-122,"HHN":182,"HHO":669,"HIH":804,"HII":679,"HOH":446,"IHH":695,"IHO":-2324,"IIH":321,"III":1497,"IIO":656,"IOO":54,"KAK":4845,"KKA":3386,"KKK":3065,"MHH":-405,"MHI":201,"MMH":-241,"MMM":661,"MOM":841};
|
||||
this.TQ1__ = {"BHHH":-227,"BHHI":316,"BHIH":-132,"BIHH":60,"BIII":1595,"BNHH":-744,"BOHH":225,"BOOO":-908,"OAKK":482,"OHHH":281,"OHIH":249,"OIHI":200,"OIIH":-68};
|
||||
this.TQ2__ = {"BIHH":-1401,"BIII":-1033,"BKAK":-543,"BOOO":-5591};
|
||||
this.TQ3__ = {"BHHH":478,"BHHM":-1073,"BHIH":222,"BHII":-504,"BIIH":-116,"BIII":-105,"BMHI":-863,"BMHM":-464,"BOMH":620,"OHHH":346,"OHHI":1729,"OHII":997,"OHMH":481,"OIHH":623,"OIIH":1344,"OKAK":2792,"OKHH":587,"OKKA":679,"OOHH":110,"OOII":-685};
|
||||
this.TQ4__ = {"BHHH":-721,"BHHM":-3604,"BHII":-966,"BIIH":-607,"BIII":-2181,"OAAA":-2763,"OAKK":180,"OHHH":-294,"OHHI":2446,"OHHO":480,"OHIH":-1573,"OIHH":1935,"OIHI":-493,"OIIH":626,"OIII":-4007,"OKAK":-8156};
|
||||
this.TW1__ = {"につい":-4681,"東京都":2026};
|
||||
this.TW2__ = {"ある程":-2049,"いった":-1256,"ころが":-2434,"しょう":3873,"その後":-4430,"だって":-1049,"ていた":1833,"として":-4657,"ともに":-4517,"もので":1882,"一気に":-792,"初めて":-1512,"同時に":-8097,"大きな":-1255,"対して":-2721,"社会党":-3216};
|
||||
this.TW3__ = {"いただ":-1734,"してい":1314,"として":-4314,"につい":-5483,"にとっ":-5989,"に当た":-6247,"ので,":-727,"ので、":-727,"のもの":-600,"れから":-3752,"十二月":-2287};
|
||||
this.TW4__ = {"いう.":8576,"いう。":8576,"からな":-2348,"してい":2958,"たが,":1516,"たが、":1516,"ている":1538,"という":1349,"ました":5543,"ません":1097,"ようと":-4258,"よると":5865};
|
||||
this.UC1__ = {"A":484,"K":93,"M":645,"O":-505};
|
||||
this.UC2__ = {"A":819,"H":1059,"I":409,"M":3987,"N":5775,"O":646};
|
||||
this.UC3__ = {"A":-1370,"I":2311};
|
||||
this.UC4__ = {"A":-2643,"H":1809,"I":-1032,"K":-3450,"M":3565,"N":3876,"O":6646};
|
||||
this.UC5__ = {"H":313,"I":-1238,"K":-799,"M":539,"O":-831};
|
||||
this.UC6__ = {"H":-506,"I":-253,"K":87,"M":247,"O":-387};
|
||||
this.UP1__ = {"O":-214};
|
||||
this.UP2__ = {"B":69,"O":935};
|
||||
this.UP3__ = {"B":189};
|
||||
this.UQ1__ = {"BH":21,"BI":-12,"BK":-99,"BN":142,"BO":-56,"OH":-95,"OI":477,"OK":410,"OO":-2422};
|
||||
this.UQ2__ = {"BH":216,"BI":113,"OK":1759};
|
||||
this.UQ3__ = {"BA":-479,"BH":42,"BI":1913,"BK":-7198,"BM":3160,"BN":6427,"BO":14761,"OI":-827,"ON":-3212};
|
||||
this.UW1__ = {",":156,"、":156,"「":-463,"あ":-941,"う":-127,"が":-553,"き":121,"こ":505,"で":-201,"と":-547,"ど":-123,"に":-789,"の":-185,"は":-847,"も":-466,"や":-470,"よ":182,"ら":-292,"り":208,"れ":169,"を":-446,"ん":-137,"・":-135,"主":-402,"京":-268,"区":-912,"午":871,"国":-460,"大":561,"委":729,"市":-411,"日":-141,"理":361,"生":-408,"県":-386,"都":-718,"｢":-463,"･":-135};
|
||||
this.UW2__ = {",":-829,"、":-829,"〇":892,"「":-645,"」":3145,"あ":-538,"い":505,"う":134,"お":-502,"か":1454,"が":-856,"く":-412,"こ":1141,"さ":878,"ざ":540,"し":1529,"す":-675,"せ":300,"そ":-1011,"た":188,"だ":1837,"つ":-949,"て":-291,"で":-268,"と":-981,"ど":1273,"な":1063,"に":-1764,"の":130,"は":-409,"ひ":-1273,"べ":1261,"ま":600,"も":-1263,"や":-402,"よ":1639,"り":-579,"る":-694,"れ":571,"を":-2516,"ん":2095,"ア":-587,"カ":306,"キ":568,"ッ":831,"三":-758,"不":-2150,"世":-302,"中":-968,"主":-861,"事":492,"人":-123,"会":978,"保":362,"入":548,"初":-3025,"副":-1566,"北":-3414,"区":-422,"大":-1769,"天":-865,"太":-483,"子":-1519,"学":760,"実":1023,"小":-2009,"市":-813,"年":-1060,"強":1067,"手":-1519,"揺":-1033,"政":1522,"文":-1355,"新":-1682,"日":-1815,"明":-1462,"最":-630,"朝":-1843,"本":-1650,"東":-931,"果":-665,"次":-2378,"民":-180,"気":-1740,"理":752,"発":529,"目":-1584,"相":-242,"県":-1165,"立":-763,"第":810,"米":509,"自":-1353,"行":838,"西":-744,"見":-3874,"調":1010,"議":1198,"込":3041,"開":1758,"間":-1257,"｢":-645,"｣":3145,"ｯ":831,"ｱ":-587,"ｶ":306,"ｷ":568};
|
||||
this.UW3__ = {",":4889,"1":-800,"−":-1723,"、":4889,"々":-2311,"〇":5827,"」":2670,"〓":-3573,"あ":-2696,"い":1006,"う":2342,"え":1983,"お":-4864,"か":-1163,"が":3271,"く":1004,"け":388,"げ":401,"こ":-3552,"ご":-3116,"さ":-1058,"し":-395,"す":584,"せ":3685,"そ":-5228,"た":842,"ち":-521,"っ":-1444,"つ":-1081,"て":6167,"で":2318,"と":1691,"ど":-899,"な":-2788,"に":2745,"の":4056,"は":4555,"ひ":-2171,"ふ":-1798,"へ":1199,"ほ":-5516,"ま":-4384,"み":-120,"め":1205,"も":2323,"や":-788,"よ":-202,"ら":727,"り":649,"る":5905,"れ":2773,"わ":-1207,"を":6620,"ん":-518,"ア":551,"グ":1319,"ス":874,"ッ":-1350,"ト":521,"ム":1109,"ル":1591,"ロ":2201,"ン":278,"・":-3794,"一":-1619,"下":-1759,"世":-2087,"両":3815,"中":653,"主":-758,"予":-1193,"二":974,"人":2742,"今":792,"他":1889,"以":-1368,"低":811,"何":4265,"作":-361,"保":-2439,"元":4858,"党":3593,"全":1574,"公":-3030,"六":755,"共":-1880,"円":5807,"再":3095,"分":457,"初":2475,"別":1129,"前":2286,"副":4437,"力":365,"動":-949,"務":-1872,"化":1327,"北":-1038,"区":4646,"千":-2309,"午":-783,"協":-1006,"口":483,"右":1233,"各":3588,"合":-241,"同":3906,"和":-837,"員":4513,"国":642,"型":1389,"場":1219,"外":-241,"妻":2016,"学":-1356,"安":-423,"実":-1008,"家":1078,"小":-513,"少":-3102,"州":1155,"市":3197,"平":-1804,"年":2416,"広":-1030,"府":1605,"度":1452,"建":-2352,"当":-3885,"得":1905,"思":-1291,"性":1822,"戸":-488,"指":-3973,"政":-2013,"教":-1479,"数":3222,"文":-1489,"新":1764,"日":2099,"旧":5792,"昨":-661,"時":-1248,"曜":-951,"最":-937,"月":4125,"期":360,"李":3094,"村":364,"東":-805,"核":5156,"森":2438,"業":484,"氏":2613,"民":-1694,"決":-1073,"法":1868,"海":-495,"無":979,"物":461,"特":-3850,"生":-273,"用":914,"町":1215,"的":7313,"直":-1835,"省":792,"県":6293,"知":-1528,"私":4231,"税":401,"立":-960,"第":1201,"米":7767,"系":3066,"約":3663,"級":1384,"統":-4229,"総":1163,"線":1255,"者":6457,"能":725,"自":-2869,"英":785,"見":1044,"調":-562,"財":-733,"費":1777,"車":1835,"軍":1375,"込":-1504,"通":-1136,"選":-681,"郎":1026,"郡":4404,"部":1200,"金":2163,"長":421,"開":-1432,"間":1302,"関":-1282,"雨":2009,"電":-1045,"非":2066,"駅":1620,"１":-800,"｣":2670,"･":-3794,"ｯ":-1350,"ｱ":551,"ｸﾞ":1319,"ｽ":874,"ﾄ":521,"ﾑ":1109,"ﾙ":1591,"ﾛ":2201,"ﾝ":278};
|
||||
this.UW4__ = {",":3930,".":3508,"―":-4841,"、":3930,"。":3508,"〇":4999,"「":1895,"」":3798,"〓":-5156,"あ":4752,"い":-3435,"う":-640,"え":-2514,"お":2405,"か":530,"が":6006,"き":-4482,"ぎ":-3821,"く":-3788,"け":-4376,"げ":-4734,"こ":2255,"ご":1979,"さ":2864,"し":-843,"じ":-2506,"す":-731,"ず":1251,"せ":181,"そ":4091,"た":5034,"だ":5408,"ち":-3654,"っ":-5882,"つ":-1659,"て":3994,"で":7410,"と":4547,"な":5433,"に":6499,"ぬ":1853,"ね":1413,"の":7396,"は":8578,"ば":1940,"ひ":4249,"び":-4134,"ふ":1345,"へ":6665,"べ":-744,"ほ":1464,"ま":1051,"み":-2082,"む":-882,"め":-5046,"も":4169,"ゃ":-2666,"や":2795,"ょ":-1544,"よ":3351,"ら":-2922,"り":-9726,"る":-14896,"れ":-2613,"ろ":-4570,"わ":-1783,"を":13150,"ん":-2352,"カ":2145,"コ":1789,"セ":1287,"ッ":-724,"ト":-403,"メ":-1635,"ラ":-881,"リ":-541,"ル":-856,"ン":-3637,"・":-4371,"ー":-11870,"一":-2069,"中":2210,"予":782,"事":-190,"井":-1768,"人":1036,"以":544,"会":950,"体":-1286,"作":530,"側":4292,"先":601,"党":-2006,"共":-1212,"内":584,"円":788,"初":1347,"前":1623,"副":3879,"力":-302,"動":-740,"務":-2715,"化":776,"区":4517,"協":1013,"参":1555,"合":-1834,"和":-681,"員":-910,"器":-851,"回":1500,"国":-619,"園":-1200,"地":866,"場":-1410,"塁":-2094,"士":-1413,"多":1067,"大":571,"子":-4802,"学":-1397,"定":-1057,"寺":-809,"小":1910,"屋":-1328,"山":-1500,"島":-2056,"川":-2667,"市":2771,"年":374,"庁":-4556,"後":456,"性":553,"感":916,"所":-1566,"支":856,"改":787,"政":2182,"教":704,"文":522,"方":-856,"日":1798,"時":1829,"最":845,"月":-9066,"木":-485,"来":-442,"校":-360,"業":-1043,"氏":5388,"民":-2716,"気":-910,"沢":-939,"済":-543,"物":-735,"率":672,"球":-1267,"生":-1286,"産":-1101,"田":-2900,"町":1826,"的":2586,"目":922,"省":-3485,"県":2997,"空":-867,"立":-2112,"第":788,"米":2937,"系":786,"約":2171,"経":1146,"統":-1169,"総":940,"線":-994,"署":749,"者":2145,"能":-730,"般":-852,"行":-792,"規":792,"警":-1184,"議":-244,"谷":-1000,"賞":730,"車":-1481,"軍":1158,"輪":-1433,"込":-3370,"近":929,"道":-1291,"選":2596,"郎":-4866,"都":1192,"野":-1100,"銀":-2213,"長":357,"間":-2344,"院":-2297,"際":-2604,"電":-878,"領":-1659,"題":-792,"館":-1984,"首":1749,"高":2120,"｢":1895,"｣":3798,"･":-4371,"ｯ":-724,"ｰ":-11870,"ｶ":2145,"ｺ":1789,"ｾ":1287,"ﾄ":-403,"ﾒ":-1635,"ﾗ":-881,"ﾘ":-541,"ﾙ":-856,"ﾝ":-3637};
|
||||
this.UW5__ = {",":465,".":-299,"1":-514,"E2":-32768,"]":-2762,"、":465,"。":-299,"「":363,"あ":1655,"い":331,"う":-503,"え":1199,"お":527,"か":647,"が":-421,"き":1624,"ぎ":1971,"く":312,"げ":-983,"さ":-1537,"し":-1371,"す":-852,"だ":-1186,"ち":1093,"っ":52,"つ":921,"て":-18,"で":-850,"と":-127,"ど":1682,"な":-787,"に":-1224,"の":-635,"は":-578,"べ":1001,"み":502,"め":865,"ゃ":3350,"ょ":854,"り":-208,"る":429,"れ":504,"わ":419,"を":-1264,"ん":327,"イ":241,"ル":451,"ン":-343,"中":-871,"京":722,"会":-1153,"党":-654,"務":3519,"区":-901,"告":848,"員":2104,"大":-1296,"学":-548,"定":1785,"嵐":-1304,"市":-2991,"席":921,"年":1763,"思":872,"所":-814,"挙":1618,"新":-1682,"日":218,"月":-4353,"査":932,"格":1356,"機":-1508,"氏":-1347,"田":240,"町":-3912,"的":-3149,"相":1319,"省":-1052,"県":-4003,"研":-997,"社":-278,"空":-813,"統":1955,"者":-2233,"表":663,"語":-1073,"議":1219,"選":-1018,"郎":-368,"長":786,"間":1191,"題":2368,"館":-689,"１":-514,"Ｅ２":-32768,"｢":363,"ｲ":241,"ﾙ":451,"ﾝ":-343};
|
||||
this.UW6__ = {",":227,".":808,"1":-270,"E1":306,"、":227,"。":808,"あ":-307,"う":189,"か":241,"が":-73,"く":-121,"こ":-200,"じ":1782,"す":383,"た":-428,"っ":573,"て":-1014,"で":101,"と":-105,"な":-253,"に":-149,"の":-417,"は":-236,"も":-206,"り":187,"る":-135,"を":195,"ル":-673,"ン":-496,"一":-277,"中":201,"件":-800,"会":624,"前":302,"区":1792,"員":-1212,"委":798,"学":-960,"市":887,"広":-695,"後":535,"業":-697,"相":753,"社":-507,"福":974,"空":-822,"者":1811,"連":463,"郎":1082,"１":-270,"Ｅ１":306,"ﾙ":-673,"ﾝ":-496};
|
||||
|
||||
return this;
|
||||
}
|
||||
TinySegmenter.prototype.ctype_ = function(str) {
|
||||
for (var i in this.chartype_) {
|
||||
if (str.match(this.chartype_[i][0])) {
|
||||
return this.chartype_[i][1];
|
||||
}
|
||||
}
|
||||
return "O";
|
||||
}
|
||||
|
||||
TinySegmenter.prototype.ts_ = function(v) {
|
||||
if (v) { return v; }
|
||||
return 0;
|
||||
}
|
||||
|
||||
TinySegmenter.prototype.segment = function(input) {
|
||||
if (input == null || input == undefined || input == "") {
|
||||
return [];
|
||||
}
|
||||
var result = [];
|
||||
var seg = ["B3","B2","B1"];
|
||||
var ctype = ["O","O","O"];
|
||||
var o = input.split("");
|
||||
for (i = 0; i < o.length; ++i) {
|
||||
seg.push(o[i]);
|
||||
ctype.push(this.ctype_(o[i]))
|
||||
}
|
||||
seg.push("E1");
|
||||
seg.push("E2");
|
||||
seg.push("E3");
|
||||
ctype.push("O");
|
||||
ctype.push("O");
|
||||
ctype.push("O");
|
||||
var word = seg[3];
|
||||
var p1 = "U";
|
||||
var p2 = "U";
|
||||
var p3 = "U";
|
||||
for (var i = 4; i < seg.length - 3; ++i) {
|
||||
var score = this.BIAS__;
|
||||
var w1 = seg[i-3];
|
||||
var w2 = seg[i-2];
|
||||
var w3 = seg[i-1];
|
||||
var w4 = seg[i];
|
||||
var w5 = seg[i+1];
|
||||
var w6 = seg[i+2];
|
||||
var c1 = ctype[i-3];
|
||||
var c2 = ctype[i-2];
|
||||
var c3 = ctype[i-1];
|
||||
var c4 = ctype[i];
|
||||
var c5 = ctype[i+1];
|
||||
var c6 = ctype[i+2];
|
||||
score += this.ts_(this.UP1__[p1]);
|
||||
score += this.ts_(this.UP2__[p2]);
|
||||
score += this.ts_(this.UP3__[p3]);
|
||||
score += this.ts_(this.BP1__[p1 + p2]);
|
||||
score += this.ts_(this.BP2__[p2 + p3]);
|
||||
score += this.ts_(this.UW1__[w1]);
|
||||
score += this.ts_(this.UW2__[w2]);
|
||||
score += this.ts_(this.UW3__[w3]);
|
||||
score += this.ts_(this.UW4__[w4]);
|
||||
score += this.ts_(this.UW5__[w5]);
|
||||
score += this.ts_(this.UW6__[w6]);
|
||||
score += this.ts_(this.BW1__[w2 + w3]);
|
||||
score += this.ts_(this.BW2__[w3 + w4]);
|
||||
score += this.ts_(this.BW3__[w4 + w5]);
|
||||
score += this.ts_(this.TW1__[w1 + w2 + w3]);
|
||||
score += this.ts_(this.TW2__[w2 + w3 + w4]);
|
||||
score += this.ts_(this.TW3__[w3 + w4 + w5]);
|
||||
score += this.ts_(this.TW4__[w4 + w5 + w6]);
|
||||
score += this.ts_(this.UC1__[c1]);
|
||||
score += this.ts_(this.UC2__[c2]);
|
||||
score += this.ts_(this.UC3__[c3]);
|
||||
score += this.ts_(this.UC4__[c4]);
|
||||
score += this.ts_(this.UC5__[c5]);
|
||||
score += this.ts_(this.UC6__[c6]);
|
||||
score += this.ts_(this.BC1__[c2 + c3]);
|
||||
score += this.ts_(this.BC2__[c3 + c4]);
|
||||
score += this.ts_(this.BC3__[c4 + c5]);
|
||||
score += this.ts_(this.TC1__[c1 + c2 + c3]);
|
||||
score += this.ts_(this.TC2__[c2 + c3 + c4]);
|
||||
score += this.ts_(this.TC3__[c3 + c4 + c5]);
|
||||
score += this.ts_(this.TC4__[c4 + c5 + c6]);
|
||||
// score += this.ts_(this.TC5__[c4 + c5 + c6]);
|
||||
score += this.ts_(this.UQ1__[p1 + c1]);
|
||||
score += this.ts_(this.UQ2__[p2 + c2]);
|
||||
score += this.ts_(this.UQ3__[p3 + c3]);
|
||||
score += this.ts_(this.BQ1__[p2 + c2 + c3]);
|
||||
score += this.ts_(this.BQ2__[p2 + c3 + c4]);
|
||||
score += this.ts_(this.BQ3__[p3 + c2 + c3]);
|
||||
score += this.ts_(this.BQ4__[p3 + c3 + c4]);
|
||||
score += this.ts_(this.TQ1__[p2 + c1 + c2 + c3]);
|
||||
score += this.ts_(this.TQ2__[p2 + c2 + c3 + c4]);
|
||||
score += this.ts_(this.TQ3__[p3 + c1 + c2 + c3]);
|
||||
score += this.ts_(this.TQ4__[p3 + c2 + c3 + c4]);
|
||||
var p = "O";
|
||||
if (score > 0) {
|
||||
result.push(word);
|
||||
word = "";
|
||||
p = "B";
|
||||
}
|
||||
p1 = p2;
|
||||
p2 = p3;
|
||||
p3 = p;
|
||||
word += seg[i];
|
||||
}
|
||||
result.push(word);
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
lunr.TinySegmenter = TinySegmenter;
|
||||
};
|
||||
|
||||
}));
|
||||
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
+1
File diff suppressed because one or more lines are too long
File diff suppressed because one or more lines are too long
+1
File diff suppressed because one or more lines are too long
@@ -0,0 +1 @@
|
||||
{"version":3,"sources":["src/templates/assets/stylesheets/palette/_scheme.scss","../../../../src/templates/assets/stylesheets/palette.scss","src/templates/assets/stylesheets/palette/_accent.scss","src/templates/assets/stylesheets/palette/_primary.scss","src/templates/assets/stylesheets/utilities/_break.scss"],"names":[],"mappings":"AA2BA,cAGE,6BAME,sDAAA,CACA,6DAAA,CACA,+DAAA,CACA,gEAAA,CACA,mDAAA,CACA,6DAAA,CACA,+DAAA,CACA,gEAAA,CAGA,mDAAA,CACA,gDAAA,CACA,yDAAA,CACA,4DAAA,CAGA,0BAAA,CACA,mCAAA,CAGA,iCAAA,CACA,kCAAA,CACA,mCAAA,CACA,mCAAA,CACA,kCAAA,CACA,iCAAA,CACA,+CAAA,CACA,6DAAA,CACA,gEAAA,CACA,4DAAA,CACA,4DAAA,CACA,6DAAA,CAGA,6CAAA,CAGA,+CAAA,CAGA,uDAAA,CACA,6DAAA,CACA,2DAAA,CAGA,iCAAA,CAGA,yDAAA,CACA,iEAAA,CAGA,mDAAA,CACA,mDAAA,CAGA,qDAAA,CACA,uDAAA,CAGA,8DAAA,CAKA,8DAAA,CAKA,0DAAA,CAzEA,iBCiBF,CD6DE,kHAEE,YC3DJ,CDkFE,yDACE,4BChFJ,CD+EE,2DACE,4BC7EJ,CD4EE,gEACE,4BC1EJ,CDyEE,2DACE,4BCvEJ,CDsEE,yDACE,4BCpEJ,CDmEE,0DACE,4BCjEJ,CDgEE,gEACE,4BC9DJ,CD6DE,0DACE,4BC3DJ,CD0DE,2OACE,4BC/CJ,CDsDA,+FAGE,iCCpDF,CACF,CCjDE,2BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCD6CN,CCvDE,4BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDoDN,CC9DE,8BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCD2DN,CCrEE,mCACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDkEN,CC5EE,8BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDyEN,CCnFE,4BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDgFN,CC1FE,kCACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDuFN,CCjGE,4BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCD8FN,CCxGE,4BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDqGN,CC/GE,6BACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCD4GN,CCtHE,mCACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDmHN,CC7HE,4BACE,4BAAA,CACA,2CAAA,CAIE,8BAAA,CACA,qCD6HN,CCpIE,8BACE,4BAAA,CACA,2CAAA,CAIE,8BAAA,CACA,qCDoIN,CC3IE,6BACE,yBAAA,CACA,2CAAA,CAIE,8BAAA,CACA,qCD2IN,CClJE,8BACE,4BAAA,CACA,2CAAA,CAIE,8BAAA,CACA,qCDkJN,CCzJE,mCACE,4BAAA,CACA,2CAAA,CAOE,yBAAA,CACA,qCDsJN,CE3JE,4BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwJN,CEnKE,6BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFgKN,CE3KE,+BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwKN,CEnLE,oCACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFgLN,CE3LE,+BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwLN,CEnME,6BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFgMN,CE3ME,mCACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwMN,CEnNE,6BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFgNN,CE3NE,6BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwNN,CEnOE,8BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFgON,CE3OE,oCACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwON,CEnPE,6BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAIE,+BAAA,CACA,sCFmPN,CE3PE,+BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAIE,+BAAA,CACA,sCF2PN,CEnQE,8BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAIE,+BAAA,CACA,sCFmQN,CE3QE,+BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAIE,+BAAA,CACA,sCF2QN,CEnRE,oCACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFgRN,CE3RE,8BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCFwRN,CEnSE,6BACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCAAA,CAKA,4BF4RN,CE5SE,kCACE,6BAAA,CACA,oCAAA,CACA,mCAAA,CAOE,0BAAA,CACA,sCAAA,CAKA,4BFqSN,CEtRE,sEACE,4BFyRJ,CE1RE,+DACE,4BF6RJ,CE9RE,iEACE,4BFiSJ,CElSE,gEACE,4BFqSJ,CEtSE,iEACE,4BFySJ,CEhSA,8BACE,mDAAA,CACA,4DAAA,CACA,0DAAA,CACA,oDAAA,CACA,2DAAA,CAGA,4BFiSF,CE9RE,yCACE,+BFgSJ,CE7RI,kDAEE,0CAAA,CACA,sCAAA,CAFA,mCFiSN,CG7MI,mCD1EA,+CACE,8CF0RJ,CEvRI,qDACE,8CFyRN,CEpRE,iEACE,mCFsRJ,CACF,CGxNI,sCDvDA,uCACE,oCFkRJ,CACF,CEzQA,8BACE,kDAAA,CACA,4DAAA,CACA,wDAAA,CACA,oDAAA,CACA,6DAAA,CAGA,4BF0QF,CEvQE,yCACE,+BFyQJ,CEtQI,kDAEE,0CAAA,CACA,sCAAA,CAFA,mCF0QN,CEnQE,yCACE,6CFqQJ,CG9NI,0CDhCA,8CACE,gDFiQJ,CACF,CGnOI,0CDvBA,iFACE,6CF6PJ,CACF,CG3PI,sCDKA,uCACE,6CFyPJ,CACF","file":"palette.css"}
|
||||
@@ -1,266 +0,0 @@
|
||||
#!/bin/bash -e
|
||||
|
||||
# setup instructions for clang2py
|
||||
if [[ ! $(clang2py -V) ]]; then
|
||||
pushd .
|
||||
cd /tmp
|
||||
sudo apt-get install -y --no-install-recommends clang
|
||||
pip install --upgrade pip setuptools
|
||||
pip install clang==14.0.6
|
||||
git clone https://github.com/nimlgen/ctypeslib.git
|
||||
cd ctypeslib
|
||||
pip install --user .
|
||||
clang2py -V
|
||||
popd
|
||||
fi
|
||||
|
||||
BASE=tinygrad/runtime/autogen/
|
||||
|
||||
fixup() {
|
||||
sed -i '1s/^/# mypy: ignore-errors\n/' $1
|
||||
sed -i 's/ *$//' $1
|
||||
grep FIXME_STUB $1 || true
|
||||
}
|
||||
|
||||
patch_dlopen() {
|
||||
path=$1; shift
|
||||
name=$1; shift
|
||||
cat <<EOF | sed -i "/import ctypes.*/r /dev/stdin" $path
|
||||
PATHS_TO_TRY = [
|
||||
$(for p in "$@"; do echo " $p,"; done)
|
||||
]
|
||||
def _try_dlopen_$name():
|
||||
library = ctypes.util.find_library("$name")
|
||||
if library: return ctypes.CDLL(library)
|
||||
for candidate in PATHS_TO_TRY:
|
||||
try: return ctypes.CDLL(candidate)
|
||||
except OSError: pass
|
||||
raise RuntimeError("library $name not found")
|
||||
EOF
|
||||
}
|
||||
|
||||
generate_opencl() {
|
||||
clang2py /usr/include/CL/cl.h -o $BASE/opencl.py -l /usr/lib/x86_64-linux-gnu/libOpenCL.so.1 -k cdefstum
|
||||
fixup $BASE/opencl.py
|
||||
# hot patches
|
||||
sed -i "s\import ctypes\import ctypes, ctypes.util\g" $BASE/opencl.py
|
||||
sed -i "s\ctypes.CDLL('/usr/lib/x86_64-linux-gnu/libOpenCL.so.1')\ctypes.CDLL(ctypes.util.find_library('OpenCL'))\g" $BASE/opencl.py
|
||||
python3 -c "import tinygrad.runtime.autogen.opencl"
|
||||
}
|
||||
|
||||
generate_hip() {
|
||||
clang2py /opt/rocm/include/hip/hip_ext.h /opt/rocm/include/hip/hiprtc.h \
|
||||
/opt/rocm/include/hip/hip_runtime_api.h /opt/rocm/include/hip/driver_types.h \
|
||||
--clang-args="-D__HIP_PLATFORM_AMD__ -I/opt/rocm/include -x c++" -o $BASE/hip.py -l /opt/rocm/lib/libamdhip64.so
|
||||
echo "hipDeviceProp_t = hipDeviceProp_tR0600" >> $BASE/hip.py
|
||||
echo "hipGetDeviceProperties = hipGetDevicePropertiesR0600" >> $BASE/hip.py
|
||||
fixup $BASE/hip.py
|
||||
# we can trust HIP is always at /opt/rocm/lib
|
||||
#sed -i "s\import ctypes\import ctypes, ctypes.util\g" $BASE/hip.py
|
||||
#sed -i "s\ctypes.CDLL('/opt/rocm/lib/libhiprtc.so')\ctypes.CDLL(ctypes.util.find_library('hiprtc'))\g" $BASE/hip.py
|
||||
#sed -i "s\ctypes.CDLL('/opt/rocm/lib/libamdhip64.so')\ctypes.CDLL(ctypes.util.find_library('amdhip64'))\g" $BASE/hip.py
|
||||
sed -i "s\import ctypes\import ctypes, os\g" $BASE/hip.py
|
||||
sed -i "s\'/opt/rocm/\os.getenv('ROCM_PATH', '/opt/rocm/')+'/\g" $BASE/hip.py
|
||||
python3 -c "import tinygrad.runtime.autogen.hip"
|
||||
}
|
||||
|
||||
generate_comgr() {
|
||||
clang2py /opt/rocm/include/amd_comgr/amd_comgr.h \
|
||||
--clang-args="-D__HIP_PLATFORM_AMD__ -I/opt/rocm/include -x c++" -o $BASE/comgr.py -l /opt/rocm/lib/libamd_comgr.so
|
||||
fixup $BASE/comgr.py
|
||||
sed -i "s\import ctypes\import ctypes, ctypes.util, os\g" $BASE/comgr.py
|
||||
patch_dlopen $BASE/comgr.py amd_comgr "'/opt/rocm/lib/libamd_comgr.so'" "os.getenv('ROCM_PATH', '')+'/lib/libamd_comgr.so'"
|
||||
sed -i "s\ctypes.CDLL('/opt/rocm/lib/libamd_comgr.so')\_try_dlopen_amd_comgr()\g" $BASE/comgr.py
|
||||
python3 -c "import tinygrad.runtime.autogen.comgr"
|
||||
}
|
||||
|
||||
generate_kfd() {
|
||||
clang2py /usr/include/linux/kfd_ioctl.h -o $BASE/kfd.py -k cdefstum
|
||||
|
||||
fixup $BASE/kfd.py
|
||||
sed -i "s\import ctypes\import ctypes, os\g" $BASE/kfd.py
|
||||
python3 -c "import tinygrad.runtime.autogen.kfd"
|
||||
}
|
||||
|
||||
generate_cuda() {
|
||||
clang2py /usr/include/cuda.h -o $BASE/cuda.py -l /usr/lib/x86_64-linux-gnu/libcuda.so
|
||||
sed -i "s\import ctypes\import ctypes, ctypes.util\g" $BASE/cuda.py
|
||||
sed -i "s\ctypes.CDLL('/usr/lib/x86_64-linux-gnu/libcuda.so')\ctypes.CDLL(ctypes.util.find_library('cuda'))\g" $BASE/cuda.py
|
||||
fixup $BASE/cuda.py
|
||||
python3 -c "import tinygrad.runtime.autogen.cuda"
|
||||
}
|
||||
|
||||
generate_nvrtc() {
|
||||
clang2py /usr/local/cuda/include/nvrtc.h /usr/local/cuda/include/nvJitLink.h -o $BASE/nvrtc.py -l /usr/local/cuda/lib64/libnvrtc.so -l /usr/local/cuda/lib64/libnvJitLink.so
|
||||
sed -i "s\import ctypes\import ctypes, ctypes.util\g" $BASE/nvrtc.py
|
||||
sed -i "s\ctypes.CDLL('/usr/local/cuda/lib64/libnvrtc.so')\ctypes.CDLL(ctypes.util.find_library('nvrtc'))\g" $BASE/nvrtc.py
|
||||
sed -i "s\ctypes.CDLL('/usr/local/cuda/lib64/libnvJitLink.so')\ctypes.CDLL(ctypes.util.find_library('nvJitLink'))\g" $BASE/nvrtc.py
|
||||
fixup $BASE/nvrtc.py
|
||||
python3 -c "import tinygrad.runtime.autogen.nvrtc"
|
||||
}
|
||||
|
||||
generate_nv() {
|
||||
NVKERN_COMMIT_HASH=d6b75a34094b0f56c2ccadf14e5d0bd515ed1ab6
|
||||
NVKERN_SRC=/tmp/open-gpu-kernel-modules-$NVKERN_COMMIT_HASH
|
||||
if [ ! -d "$NVKERN_SRC" ]; then
|
||||
git clone https://github.com/tinygrad/open-gpu-kernel-modules $NVKERN_SRC
|
||||
pushd .
|
||||
cd $NVKERN_SRC
|
||||
git reset --hard $NVKERN_COMMIT_HASH
|
||||
popd
|
||||
fi
|
||||
|
||||
clang2py -k cdefstum \
|
||||
extra/nv_gpu_driver/clc6c0qmd.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/cl0080.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/cl2080_notification.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/clc56f.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/clc56f.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/clc56f.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/cl83de.h \
|
||||
$NVKERN_SRC/src/nvidia/generated/g_allclasses.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/class/clc6c0.h \
|
||||
$NVKERN_SRC/kernel-open/nvidia-uvm/clc6b5.h \
|
||||
$NVKERN_SRC/kernel-open/nvidia-uvm/uvm_ioctl.h \
|
||||
$NVKERN_SRC/kernel-open/nvidia-uvm/uvm_linux_ioctl.h \
|
||||
$NVKERN_SRC/kernel-open/nvidia-uvm/hwref/ampere/ga100/dev_fault.h \
|
||||
$NVKERN_SRC/src/nvidia/arch/nvalloc/unix/include/nv_escape.h \
|
||||
$NVKERN_SRC/src/nvidia/arch/nvalloc/unix/include/nv-ioctl.h \
|
||||
$NVKERN_SRC/src/nvidia/arch/nvalloc/unix/include/nv-ioctl-numbers.h \
|
||||
$NVKERN_SRC/src/nvidia/arch/nvalloc/unix/include/nv-ioctl-numa.h \
|
||||
$NVKERN_SRC/src/nvidia/arch/nvalloc/unix/include/nv-unix-nvos-params-wrappers.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/alloc/alloc_channel.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/nvos.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrl0000/*.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrl0080/*.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrl2080/*.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrl83de/*.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrlc36f.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrlcb33.h \
|
||||
$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl/ctrla06c.h \
|
||||
--clang-args="-include $NVKERN_SRC/src/common/sdk/nvidia/inc/nvtypes.h -I$NVKERN_SRC/src/common/inc -I$NVKERN_SRC/kernel-open/nvidia-uvm -I$NVKERN_SRC/kernel-open/common/inc -I$NVKERN_SRC/src/common/sdk/nvidia/inc -I$NVKERN_SRC/src/nvidia/arch/nvalloc/unix/include -I$NVKERN_SRC/src/common/sdk/nvidia/inc/ctrl" \
|
||||
-o $BASE/nv_gpu.py
|
||||
fixup $BASE/nv_gpu.py
|
||||
sed -i "s\(0000000001)\1\g" $BASE/nv_gpu.py
|
||||
sed -i "s\import ctypes\import ctypes, os\g" $BASE/nv_gpu.py
|
||||
sed -i 's/#\?\s\([A-Za-z0-9_]\+\) = MW ( \([0-9]\+\) : \([0-9]\+\) )/\1 = (\2 , \3)/' $BASE/nv_gpu.py # NVC6C0_QMDV03_00 processing
|
||||
sed -i 's/#\sdef NVC6C0_QMD\([A-Za-z0-9_()]\+\):/def NVC6C0_QMD\1:/' $BASE/nv_gpu.py
|
||||
sed -i 's/#\s*return MW(\([0-9i()*+]\+\):\([0-9i()*+]\+\))/ return (\1 , \2)/' $BASE/nv_gpu.py
|
||||
sed -i 's/#\?\s*\(.*\)\s*=\s*\(NV\)\?BIT\(32\)\?\s*(\s*\([0-9]\+\)\s*)/\1 = (1 << \4)/' $BASE/nv_gpu.py # name = BIT(x) -> name = (1 << x)
|
||||
sed -i "s/UVM_\([A-Za-z0-9_]\+\) = \['i', '(', '\([0-9]\+\)', ')'\]/UVM_\1 = \2/" $BASE/nv_gpu.py # UVM_name = ['i', '(', '<num>', ')'] -> UVM_name = <num>
|
||||
|
||||
# Parse status codes
|
||||
sed -n '1i\
|
||||
nv_status_codes = {}
|
||||
/^NV_STATUS_CODE/ { s/^NV_STATUS_CODE(\([^,]*\), *\([^,]*\), *"\([^"]*\)") *.*$/\1 = \2\nnv_status_codes[\1] = "\3"/; p }' $NVKERN_SRC/src/common/sdk/nvidia/inc/nvstatuscodes.h >> $BASE/nv_gpu.py
|
||||
|
||||
python3 -c "import tinygrad.runtime.autogen.nv_gpu"
|
||||
}
|
||||
|
||||
generate_amd() {
|
||||
# clang2py broken when pass -x c++ to prev headers
|
||||
clang2py -k cdefstum \
|
||||
extra/hip_gpu_driver/sdma_registers.h \
|
||||
extra/hip_gpu_driver/nvd.h \
|
||||
extra/hip_gpu_driver/kfd_pm4_headers_ai.h \
|
||||
extra/hip_gpu_driver/soc21_enum.h \
|
||||
extra/hip_gpu_driver/sdma_v6_0_0_pkt_open.h \
|
||||
extra/hip_gpu_driver/gc_11_0_0_offset.h \
|
||||
extra/hip_gpu_driver/gc_10_3_0_offset.h \
|
||||
--clang-args="-I/opt/rocm/include -x c++" \
|
||||
-o $BASE/amd_gpu.py
|
||||
|
||||
fixup $BASE/amd_gpu.py
|
||||
sed -i "s\import ctypes\import ctypes, os\g" $BASE/amd_gpu.py
|
||||
python3 -c "import tinygrad.runtime.autogen.amd_gpu"
|
||||
}
|
||||
|
||||
generate_hsa() {
|
||||
clang2py \
|
||||
/opt/rocm/include/hsa/hsa.h \
|
||||
/opt/rocm/include/hsa/hsa_ext_amd.h \
|
||||
/opt/rocm/include/hsa/amd_hsa_signal.h \
|
||||
/opt/rocm/include/hsa/amd_hsa_queue.h \
|
||||
/opt/rocm/include/hsa/amd_hsa_kernel_code.h \
|
||||
/opt/rocm/include/hsa/hsa_ext_finalize.h /opt/rocm/include/hsa/hsa_ext_image.h \
|
||||
/opt/rocm/include/hsa/hsa_ven_amd_aqlprofile.h \
|
||||
--clang-args="-I/opt/rocm/include" \
|
||||
-o $BASE/hsa.py -l /opt/rocm/lib/libhsa-runtime64.so
|
||||
|
||||
fixup $BASE/hsa.py
|
||||
sed -i "s\import ctypes\import ctypes, ctypes.util, os\g" $BASE/hsa.py
|
||||
sed -i "s\ctypes.CDLL('/opt/rocm/lib/libhsa-runtime64.so')\ctypes.CDLL(os.getenv('ROCM_PATH')+'/lib/libhsa-runtime64.so' if os.getenv('ROCM_PATH') else ctypes.util.find_library('hsa-runtime64'))\g" $BASE/hsa.py
|
||||
python3 -c "import tinygrad.runtime.autogen.hsa"
|
||||
}
|
||||
|
||||
generate_io_uring() {
|
||||
clang2py -k cdefstum \
|
||||
/usr/include/liburing.h \
|
||||
/usr/include/linux/io_uring.h \
|
||||
-o $BASE/io_uring.py
|
||||
|
||||
sed -r '/^#define __NR_io_uring/ s/^#define __(NR_io_uring[^ ]+) (.*)$/\1 = \2/; t; d' /usr/include/asm-generic/unistd.h >> $BASE/io_uring.py # io_uring syscalls numbers
|
||||
fixup $BASE/io_uring.py
|
||||
}
|
||||
|
||||
generate_libc() {
|
||||
clang2py -k cdefstum \
|
||||
$(dpkg -L libc6-dev | grep sys/mman.h) \
|
||||
$(dpkg -L libc6-dev | grep sys/syscall.h) \
|
||||
/usr/include/elf.h \
|
||||
/usr/include/unistd.h \
|
||||
-o $BASE/libc.py
|
||||
|
||||
sed -i "s\import ctypes\import ctypes, ctypes.util, os\g" $BASE/libc.py
|
||||
sed -i "s\FIXME_STUB\libc\g" $BASE/libc.py
|
||||
sed -i "s\FunctionFactoryStub()\None if (libc_path := ctypes.util.find_library('c')) is None else ctypes.CDLL(libc_path)\g" $BASE/libc.py
|
||||
|
||||
fixup $BASE/libc.py
|
||||
}
|
||||
|
||||
generate_kgsl() {
|
||||
clang2py extra/qcom_gpu_driver/msm_kgsl.h -o $BASE/kgsl.py -k cdefstum
|
||||
fixup $BASE/kgsl.py
|
||||
sed -i "s\import ctypes\import ctypes, os\g" $BASE/kgsl.py
|
||||
sed -nE 's/#define ([A-Za-z0-9_]+)_SHIFT\s*[^\S\r\n]*[0-9]*$/def \1(val): return (val << \1_SHIFT) \& \1_MASK/p' extra/qcom_gpu_driver/msm_kgsl.h >> $BASE/kgsl.py
|
||||
python3 -c "import tinygrad.runtime.autogen.kgsl"
|
||||
}
|
||||
|
||||
generate_adreno() {
|
||||
clang2py extra/qcom_gpu_driver/a6xx.xml.h -o $BASE/adreno.py -k cestum
|
||||
sed -nE 's/#define ([A-Za-z0-9_]+)__SHIFT\s*[^\S\r\n]*[0-9]*$/def \1(val): return (val << \1__SHIFT) \& \1__MASK/p' extra/qcom_gpu_driver/a6xx.xml.h >> $BASE/adreno.py
|
||||
fixup $BASE/adreno.py
|
||||
sed -i "s\import ctypes\import ctypes, os\g" $BASE/adreno.py
|
||||
python3 -c "import tinygrad.runtime.autogen.adreno"
|
||||
}
|
||||
|
||||
generate_qcom() {
|
||||
clang2py -k cdefstum \
|
||||
extra/dsp/include/ion.h \
|
||||
extra/dsp/include/msm_ion.h \
|
||||
extra/dsp/include/adsprpc_shared.h \
|
||||
extra/dsp/include/remote_default.h \
|
||||
extra/dsp/include/apps_std.h \
|
||||
-o $BASE/qcom_dsp.py
|
||||
|
||||
fixup $BASE/qcom_dsp.py
|
||||
python3 -c "import tinygrad.runtime.autogen.qcom_dsp"
|
||||
}
|
||||
|
||||
if [ "$1" == "opencl" ]; then generate_opencl
|
||||
elif [ "$1" == "hip" ]; then generate_hip
|
||||
elif [ "$1" == "comgr" ]; then generate_comgr
|
||||
elif [ "$1" == "cuda" ]; then generate_cuda
|
||||
elif [ "$1" == "nvrtc" ]; then generate_nvrtc
|
||||
elif [ "$1" == "hsa" ]; then generate_hsa
|
||||
elif [ "$1" == "kfd" ]; then generate_kfd
|
||||
elif [ "$1" == "nv" ]; then generate_nv
|
||||
elif [ "$1" == "amd" ]; then generate_amd
|
||||
elif [ "$1" == "qcom" ]; then generate_qcom
|
||||
elif [ "$1" == "io_uring" ]; then generate_io_uring
|
||||
elif [ "$1" == "libc" ]; then generate_libc
|
||||
elif [ "$1" == "kgsl" ]; then generate_kgsl
|
||||
elif [ "$1" == "adreno" ]; then generate_adreno
|
||||
elif [ "$1" == "all" ]; then generate_opencl; generate_hip; generate_comgr; generate_cuda; generate_nvrtc; generate_hsa; generate_kfd; generate_nv; generate_amd; generate_io_uring; generate_libc
|
||||
else echo "usage: $0 <type>"
|
||||
fi
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -1,118 +0,0 @@
|
||||
# tinygrad is a tensor library, and as a tensor library it has multiple parts
|
||||
# 1. a "runtime". this allows buffer management, compilation, and running programs
|
||||
# 2. a "Device" that uses the runtime but specifies compute in an abstract way for all
|
||||
# 3. a "LazyBuffer" that fuses the compute into kernels, using memory only when needed
|
||||
# 4. a "Tensor" that provides an easy to use frontend with autograd ".backward()"
|
||||
|
||||
|
||||
print("******** first, the runtime ***********")
|
||||
|
||||
from tinygrad.runtime.ops_clang import ClangProgram, ClangCompiler, MallocAllocator
|
||||
|
||||
# allocate some buffers
|
||||
out = MallocAllocator.alloc(4)
|
||||
a = MallocAllocator.alloc(4)
|
||||
b = MallocAllocator.alloc(4)
|
||||
|
||||
# load in some values (little endian)
|
||||
MallocAllocator._copyin(a, memoryview(bytearray([2,0,0,0])))
|
||||
MallocAllocator._copyin(b, memoryview(bytearray([3,0,0,0])))
|
||||
|
||||
# compile a program to a binary
|
||||
lib = ClangCompiler().compile("void add(int *out, int *a, int *b) { out[0] = a[0] + b[0]; }")
|
||||
|
||||
# create a runtime for the program (ctypes.CDLL)
|
||||
fxn = ClangProgram("add", lib)
|
||||
|
||||
# run the program
|
||||
fxn(out, a, b)
|
||||
|
||||
# check the data out
|
||||
print(val := MallocAllocator._as_buffer(out).cast("I").tolist()[0])
|
||||
assert val == 5
|
||||
|
||||
|
||||
print("******** second, the Device ***********")
|
||||
|
||||
DEVICE = "CLANG" # NOTE: you can change this!
|
||||
|
||||
import struct
|
||||
from tinygrad.dtype import dtypes
|
||||
from tinygrad.device import Buffer, Device
|
||||
from tinygrad.ops import UOp, Ops
|
||||
from tinygrad.shape.shapetracker import ShapeTracker
|
||||
|
||||
# allocate some buffers + load in values
|
||||
out = Buffer(DEVICE, 1, dtypes.int32).allocate()
|
||||
a = Buffer(DEVICE, 1, dtypes.int32).allocate().copyin(memoryview(bytearray(struct.pack("I", 2))))
|
||||
b = Buffer(DEVICE, 1, dtypes.int32).allocate().copyin(memoryview(bytearray(struct.pack("I", 3))))
|
||||
# NOTE: a._buf is the same as the return from MallocAllocator.alloc
|
||||
|
||||
# describe the computation
|
||||
buf_1 = UOp(Ops.DEFINE_GLOBAL, dtypes.int32.ptr(), (), 1)
|
||||
buf_2 = UOp(Ops.DEFINE_GLOBAL, dtypes.int32.ptr(), (), 2)
|
||||
ld_1 = UOp(Ops.LOAD, dtypes.int32, (buf_1, ShapeTracker.from_shape((1,)).to_uop()))
|
||||
ld_2 = UOp(Ops.LOAD, dtypes.int32, (buf_2, ShapeTracker.from_shape((1,)).to_uop()))
|
||||
alu = ld_1 + ld_2
|
||||
output_buf = UOp(Ops.DEFINE_GLOBAL, dtypes.int32.ptr(), (), 0)
|
||||
st_0 = UOp(Ops.STORE, dtypes.void, (output_buf, ShapeTracker.from_shape((1,)).to_uop(), alu))
|
||||
s = UOp(Ops.SINK, dtypes.void, (st_0,))
|
||||
|
||||
# convert the computation to a "linearized" format (print the format)
|
||||
from tinygrad.engine.realize import get_kernel, CompiledRunner
|
||||
kernel = get_kernel(Device[DEVICE].renderer, s).linearize()
|
||||
|
||||
# compile a program (and print the source)
|
||||
fxn = CompiledRunner(kernel.to_program())
|
||||
print(fxn.p.src)
|
||||
# NOTE: fxn.clprg is the ClangProgram
|
||||
|
||||
# run the program
|
||||
fxn.exec([out, a, b])
|
||||
|
||||
# check the data out
|
||||
assert out.as_buffer().cast('I')[0] == 5
|
||||
|
||||
|
||||
print("******** third, the LazyBuffer ***********")
|
||||
|
||||
from tinygrad.engine.realize import run_schedule
|
||||
from tinygrad.engine.schedule import create_schedule
|
||||
|
||||
# allocate some values + load in values
|
||||
a = UOp.metaop(Ops.EMPTY, (1,), dtypes.int32, DEVICE)
|
||||
b = UOp.metaop(Ops.EMPTY, (1,), dtypes.int32, DEVICE)
|
||||
a.buffer.allocate().copyin(memoryview(bytearray(struct.pack("I", 2))))
|
||||
b.buffer.allocate().copyin(memoryview(bytearray(struct.pack("I", 3))))
|
||||
del a.srcs
|
||||
del b.srcs
|
||||
|
||||
# describe the computation
|
||||
out = a.alu(Ops.ADD, b)
|
||||
|
||||
# schedule the computation as a list of kernels
|
||||
sched = create_schedule([out])
|
||||
for si in sched: print(si.ast.op) # NOTE: the first two convert it to CLANG
|
||||
|
||||
# DEBUGGING: print the compute ast
|
||||
print(sched[-1].ast)
|
||||
# NOTE: sched[-1].ast is the same as st_0 above
|
||||
|
||||
# run that schedule
|
||||
run_schedule(sched)
|
||||
|
||||
# check the data out
|
||||
assert out.realized is not None and out.realized.as_buffer().cast('I')[0] == 5
|
||||
|
||||
|
||||
print("******** fourth, the Tensor ***********")
|
||||
|
||||
from tinygrad import Tensor
|
||||
|
||||
a = Tensor([2], dtype=dtypes.int32, device=DEVICE)
|
||||
b = Tensor([3], dtype=dtypes.int32, device=DEVICE)
|
||||
out = a + b
|
||||
|
||||
# check the data out
|
||||
print(val:=out.item())
|
||||
assert val == 5
|
||||
@@ -1,61 +0,0 @@
|
||||
# abstractions2 goes from back to front, here we will go from front to back
|
||||
from typing import List
|
||||
from tinygrad.helpers import tqdm
|
||||
|
||||
# *****
|
||||
# 0. Load mnist on the device
|
||||
|
||||
from tinygrad.nn.datasets import mnist
|
||||
X_train, Y_train, _, _ = mnist()
|
||||
X_train = X_train.float()
|
||||
X_train -= X_train.mean()
|
||||
|
||||
# *****
|
||||
# 1. Define an MNIST model.
|
||||
|
||||
from tinygrad import Tensor
|
||||
|
||||
l1 = Tensor.kaiming_uniform(128, 784)
|
||||
l2 = Tensor.kaiming_uniform(10, 128)
|
||||
def model(x): return x.flatten(1).dot(l1.T).relu().dot(l2.T)
|
||||
l1n, l2n = l1.numpy(), l2.numpy()
|
||||
|
||||
# *****
|
||||
# 2. Choose a batch for training and do the backward pass.
|
||||
|
||||
from tinygrad.nn.optim import SGD
|
||||
optim = SGD([l1, l2])
|
||||
|
||||
X, Y = X_train[(samples:=Tensor.randint(128, high=X_train.shape[0]))], Y_train[samples]
|
||||
optim.zero_grad()
|
||||
model(X).sparse_categorical_crossentropy(Y).backward()
|
||||
optim._step() # this will step the optimizer without running realize
|
||||
|
||||
# *****
|
||||
# 3. Create a schedule.
|
||||
|
||||
# The weight Tensors have been assigned to, but not yet realized. Everything is still lazy at this point
|
||||
# l1.lazydata and l2.lazydata define a computation graph
|
||||
|
||||
from tinygrad.engine.schedule import ScheduleItem
|
||||
schedule: List[ScheduleItem] = Tensor.schedule(l1, l2)
|
||||
|
||||
print(f"The schedule contains {len(schedule)} items.")
|
||||
for si in schedule: print(str(si)[:80])
|
||||
|
||||
# *****
|
||||
# 4. Lower a schedule.
|
||||
|
||||
from tinygrad.engine.realize import lower_schedule_item, ExecItem
|
||||
lowered: List[ExecItem] = [ExecItem(lower_schedule_item(si).prg, list(si.bufs)) for si in tqdm(schedule)]
|
||||
|
||||
# *****
|
||||
# 5. Run the schedule
|
||||
|
||||
for ei in tqdm(lowered): ei.run()
|
||||
|
||||
# *****
|
||||
# 6. Print the weight change
|
||||
|
||||
print("first weight change\n", l1.numpy()-l1n)
|
||||
print("second weight change\n", l2.numpy()-l2n)
|
||||
@@ -1,52 +0,0 @@
|
||||
The tinygrad framework has four pieces
|
||||
|
||||
* a PyTorch like <b>frontend</b>.
|
||||
* a <b>scheduler</b> which breaks the compute into kernels.
|
||||
* a <b>lowering</b> engine which converts ASTs into code that can run on the accelerator.
|
||||
* an <b>execution</b> engine which can run that code.
|
||||
|
||||
There is a good [bunch of tutorials](https://mesozoic-egg.github.io/tinygrad-notes/) by Di Zhu that go over tinygrad internals.
|
||||
|
||||
## Frontend
|
||||
|
||||
Everything in [Tensor](../tensor/index.md) is syntactic sugar around [function.py](function.md), where the forwards and backwards passes are implemented for the different functions. There's about 25 of them, implemented using about 20 basic ops. Those basic ops go on to construct a graph of [UOps](../developer/uop.md).
|
||||
|
||||
The `UOp` graph specifies the compute in terms of low level tinygrad ops. Not all UOps will actually become realized. There's two types of UOps, base and view. base contains compute into a contiguous buffer, and view is a view (specified by a ShapeTracker). Inputs to a base can be either base or view, inputs to a view can only be a single base.
|
||||
|
||||
## Scheduling
|
||||
|
||||
The [scheduler](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/engine/schedule.py) converts the graph of UOps into a list of `ScheduleItem`. One `ScheduleItem` is one kernel on the GPU, and the scheduler is responsible for breaking the large compute graph into subgraphs that can fit in a kernel. `ast` specifies what compute to run, and `bufs` specifies what buffers to run it on.
|
||||
|
||||
::: tinygrad.engine.schedule.ScheduleItem
|
||||
|
||||
## Lowering
|
||||
|
||||
The code in [realize](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/engine/realize.py) lowers `ScheduleItem` to `ExecItem` with
|
||||
|
||||
::: tinygrad.engine.realize.lower_schedule
|
||||
|
||||
There's a ton of complexity hidden behind this, see the `codegen/` directory.
|
||||
|
||||
First we lower the AST to UOps, which is a linear list of the compute to be run. This is where the BEAM search happens.
|
||||
|
||||
Then we render the UOps into code with a `Renderer`, then we compile the code to binary with a `Compiler`.
|
||||
|
||||
## Execution
|
||||
|
||||
Creating `ExecItem`, which has a run method
|
||||
|
||||
::: tinygrad.engine.realize.ExecItem
|
||||
options:
|
||||
members: true
|
||||
|
||||
Lists of `ExecItem` can be condensed into a single ExecItem with the Graph API (rename to Queue?)
|
||||
|
||||
## Runtime
|
||||
|
||||
Runtimes are responsible for device-specific interactions. They handle tasks such as initializing devices, allocating memory, loading/launching programs, and more. You can find more information about the runtimes API on the [runtime overview page](runtime.md).
|
||||
|
||||
All runtime implementations can be found in the [runtime directory](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime).
|
||||
|
||||
### HCQ Compatible Runtimes
|
||||
|
||||
HCQ API is a lower-level API for defining runtimes. Interaction with HCQ-compatible devices occurs at a lower level, with commands issued directly to hardware queues. Some examples of such backends are [NV](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_nv.py) and [AMD](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_amd.py), which are userspace drivers for NVIDIA and AMD devices respectively. You can find more information about the API on [HCQ overview page](hcq.md)
|
||||
@@ -1,33 +0,0 @@
|
||||
::: tinygrad.function
|
||||
options:
|
||||
members: [
|
||||
"Contiguous",
|
||||
"ContiguousBackward",
|
||||
"Cast",
|
||||
"Neg",
|
||||
"Reciprocal",
|
||||
"Sin",
|
||||
"Relu",
|
||||
"Log",
|
||||
"Exp",
|
||||
"Sqrt",
|
||||
"Sigmoid",
|
||||
"Sign",
|
||||
"Less",
|
||||
"Eq",
|
||||
"Xor",
|
||||
"Add",
|
||||
"Sub",
|
||||
"Mul",
|
||||
"Div",
|
||||
"Where",
|
||||
"Sum",
|
||||
"Max",
|
||||
"Expand",
|
||||
"Reshape",
|
||||
"Permute",
|
||||
"Pad",
|
||||
"Shrink",
|
||||
"Flip",
|
||||
]
|
||||
show_source: false
|
||||
@@ -1,129 +0,0 @@
|
||||
# HCQ Compatible Runtime
|
||||
|
||||
## Overview
|
||||
|
||||
The main aspect of HCQ-compatible runtimes is how they interact with devices. In HCQ, all interactions with devices occur in a hardware-friendly manner using [command queues](#command-queues). This approach allows commands to be issued directly to devices, bypassing runtime overhead such as HIP or CUDA. Additionally, by using the HCQ API, these runtimes can benefit from various optimizations and features, including [HCQGraph](#hcqgraph) and built-in profiling capabilities.
|
||||
|
||||
### Command Queues
|
||||
|
||||
To interact with devices you create a `HWQueue`. Some methods are required, like timestamp and synchronization methods like [signal](#tinygrad.runtime.support.hcq.HWQueue.signal) and [wait](#tinygrad.runtime.support.hcq.HWQueue.wait), while others are dependent on it being a compute or copy queue.
|
||||
|
||||
For example, the following Python code enqueues a wait, execute, and signal command on the HCQ-compatible device:
|
||||
```python
|
||||
HWQueue().wait(signal_to_wait, value_to_wait) \
|
||||
.exec(program, args_state, global_dims, local_dims) \
|
||||
.signal(signal_to_fire, value_to_fire) \
|
||||
.submit(your_device)
|
||||
```
|
||||
|
||||
Each runtime should implement the required functions that are defined in the `HWQueue` classes.
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HWQueue
|
||||
options:
|
||||
members: [
|
||||
"signal",
|
||||
"wait",
|
||||
"timestamp",
|
||||
"bind",
|
||||
"submit",
|
||||
"memory_barrier",
|
||||
"exec",
|
||||
"copy",
|
||||
]
|
||||
show_source: false
|
||||
|
||||
### HCQ Compatible Device
|
||||
|
||||
The `HCQCompiled` class defines the API for HCQ-compatible devices. This class serves as an abstract base class that device-specific implementations should inherit from and implement.
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HCQCompiled
|
||||
options:
|
||||
show_source: false
|
||||
|
||||
#### Signals
|
||||
|
||||
Signals are device-dependent structures used for synchronization and timing in HCQ-compatible devices. They should be designed to record both a `value` and a `timestamp` within the same signal. HCQ-compatible backend implementations should use `HCQSignal` as a base class.
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HCQSignal
|
||||
options:
|
||||
members: [value, timestamp, wait]
|
||||
show_source: false
|
||||
|
||||
The following Python code demonstrates the usage of signals:
|
||||
|
||||
```python
|
||||
signal = your_device.signal_t()
|
||||
|
||||
HWQueue().timestamp(signal) \
|
||||
.signal(signal, value_to_fire) \
|
||||
.submit(your_device)
|
||||
|
||||
signal.wait(value_to_fire)
|
||||
signaled_value = signal.value # should be the same as `value_to_fire`
|
||||
timestamp = signal.timestamp
|
||||
```
|
||||
|
||||
##### Synchronization signals
|
||||
|
||||
Each HCQ-compatible device must allocate two signals for global synchronization purposes. These signals are passed to the `HCQCompiled` base class during initialization: an active timeline signal `self.timeline_signal` and a shadow timeline signal `self._shadow_timeline_signal` which helps to handle signal value overflow issues. You can find more about synchronization in the [synchronization section](#synchronization)
|
||||
|
||||
### HCQ Compatible Allocator
|
||||
|
||||
The `HCQAllocator` base class simplifies allocator logic by leveraging [command queues](#command-queues) abstractions. This class efficiently handles copy and transfer operations, leaving only the alloc and free functions to be implemented by individual backends.
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HCQAllocator
|
||||
options:
|
||||
members: [
|
||||
"_alloc",
|
||||
"_free",
|
||||
]
|
||||
show_source: false
|
||||
|
||||
#### HCQ Allocator Result Protocol
|
||||
|
||||
Backends must adhere to the `HCQBuffer` protocol when returning allocation results.
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HCQBuffer
|
||||
options:
|
||||
members: true
|
||||
show_source: false
|
||||
|
||||
### HCQ Compatible Program
|
||||
|
||||
`HCQProgram` is a base class for defining programs compatible with HCQ-enabled devices. It provides a flexible framework for handling different argument layouts (see `HCQArgsState`).
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HCQProgram
|
||||
options:
|
||||
members: true
|
||||
show_source: false
|
||||
|
||||
#### Arguments State
|
||||
|
||||
`HCQArgsState` is a base class for managing the argument state for HCQ programs. Backend implementations should create a subclass of `HCQArgsState` to manage arguments for the given program.
|
||||
|
||||
::: tinygrad.runtime.support.hcq.HCQArgsState
|
||||
options:
|
||||
members: true
|
||||
show_source: false
|
||||
|
||||
**Lifetime**: The `HCQArgsState` is passed to `HWQueue.exec` and is guaranteed not to be freed until `HWQueue.submit` for the same queue is called.
|
||||
|
||||
### Synchronization
|
||||
|
||||
HCQ-compatible devices use a global timeline signal for synchronizing all operations. This mechanism ensures proper ordering and completion of tasks across the device. By convention, `self.timeline_value` points to the next value to signal. So, to wait for all previous operations on the device to complete, wait for `self.timeline_value - 1` value. The following Python code demonstrates the typical usage of signals to synchronize execution to other operations on the device:
|
||||
|
||||
```python
|
||||
HWQueue().wait(your_device.timeline_signal, your_device.timeline_value - 1) \
|
||||
.exec(...)
|
||||
.signal(your_device.timeline_signal, your_device.timeline_value) \
|
||||
.submit(your_device)
|
||||
your_device.timeline_value += 1
|
||||
|
||||
# Optionally wait for execution
|
||||
your_device.timeline_signal.wait(your_device.timeline_value - 1)
|
||||
```
|
||||
|
||||
## HCQGraph
|
||||
|
||||
[HCQGraph](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/graph/hcq.py) is a core feature that implements `GraphRunner` for HCQ-compatible devices. `HCQGraph` builds static `HWQueue` for all operations per device. To optimize enqueue time, only the necessary parts of the queues are updated for each run using the symbolic variables, avoiding a complete rebuild.
|
||||
Optionally, queues can implement a `bind` API, which allows further optimization by eliminating the need to copy the queues into the device ring.
|
||||
@@ -1,51 +0,0 @@
|
||||
# Runtime Overview
|
||||
|
||||
## Overview
|
||||
|
||||
A typical runtime consists of the following parts:
|
||||
|
||||
- [Compiled](#compiled)
|
||||
- [Allocator](#allocator)
|
||||
- [Program](#program)
|
||||
- [Compiler](#compiler)
|
||||
|
||||
### Compiled
|
||||
|
||||
The `Compiled` class is responsible for initializing and managing a device.
|
||||
|
||||
::: tinygrad.device.Compiled
|
||||
options:
|
||||
members: [
|
||||
"synchronize"
|
||||
]
|
||||
show_source: false
|
||||
|
||||
### Allocator
|
||||
|
||||
The `Allocator` class is responsible for managing memory on the device. There is also a version called the `LRUAllocator`, which caches allocated buffers to optimize performance.
|
||||
|
||||
::: tinygrad.device.Allocator
|
||||
options:
|
||||
members: true
|
||||
show_source: false
|
||||
|
||||
::: tinygrad.device.LRUAllocator
|
||||
options:
|
||||
members: true
|
||||
show_source: false
|
||||
|
||||
### Program
|
||||
|
||||
The `Program` class is created for each loaded program. It is responsible for compiling and executing the program on the device. As an example, here is a `ClangProgram` implementation which loads program and runs it.
|
||||
|
||||
::: tinygrad.runtime.ops_clang.ClangProgram
|
||||
options:
|
||||
members: true
|
||||
|
||||
### Compiler
|
||||
|
||||
The `Compiler` class compiles the output from the `Renderer` and produces it in a device-specific format.
|
||||
|
||||
::: tinygrad.device.Compiler
|
||||
options:
|
||||
members: true
|
||||
@@ -1,11 +0,0 @@
|
||||
::: tinygrad.ops.UOp
|
||||
options:
|
||||
members: false
|
||||
members_order: source
|
||||
show_labels: false
|
||||
|
||||
::: tinygrad.ops.Ops
|
||||
options:
|
||||
members: true
|
||||
members_order: source
|
||||
show_labels: false
|
||||
@@ -1,9 +0,0 @@
|
||||
::: tinygrad.dtype.DType
|
||||
|
||||
::: tinygrad.dtype.dtypes
|
||||
options:
|
||||
members: true
|
||||
members_order: source
|
||||
show_labels: false
|
||||
|
||||
::: tinygrad.dtype.ConstType
|
||||
@@ -1,50 +0,0 @@
|
||||
# List of environment variables that control tinygrad behavior.
|
||||
|
||||
This is a list of environment variable that control the runtime behavior of tinygrad and its examples.
|
||||
Most of these are self-explanatory, and are usually used to set an option at runtime.
|
||||
|
||||
Example: `GPU=1 DEBUG=4 python3 -m pytest`
|
||||
|
||||
However you can also decorate a function to set a value only inside that function.
|
||||
|
||||
```python
|
||||
# in tensor.py (probably only useful if you are a tinygrad developer)
|
||||
@Context(DEBUG=4)
|
||||
def numpy(self) -> ...
|
||||
```
|
||||
|
||||
Or use contextmanager to temporarily set a value inside some scope:
|
||||
|
||||
```python
|
||||
with Context(DEBUG=0):
|
||||
a = Tensor.ones(10, 10)
|
||||
a *= 2
|
||||
```
|
||||
|
||||
## Global Variables
|
||||
The columns of this list are are: Variable, Possible Value(s) and Description.
|
||||
|
||||
- A `#` means that the variable can take any integer value.
|
||||
|
||||
These control the behavior of core tinygrad even when used as a library.
|
||||
|
||||
Variable | Possible Value(s) | Description
|
||||
---|---|---
|
||||
DEBUG | [1-6] | enable debugging output, with 4 you get operations, timings, speed, generated code and more
|
||||
GPU | [1] | enable the GPU backend
|
||||
CUDA | [1] | enable CUDA backend
|
||||
AMD | [1] | enable AMD backend
|
||||
NV | [1] | enable NV backend
|
||||
METAL | [1] | enable Metal backend (for Mac M1 and after)
|
||||
METAL_XCODE | [1] | enable Metal using macOS Xcode SDK
|
||||
CLANG | [1] | enable Clang backend
|
||||
LLVM | [1] | enable LLVM backend
|
||||
BEAM | [#] | number of beams in kernel beam search
|
||||
DEFAULT_FLOAT | [HALF, ...]| specify the default float dtype (FLOAT32, HALF, BFLOAT16, FLOAT64, ...), default to FLOAT32
|
||||
IMAGE | [1-2] | enable 2d specific optimizations
|
||||
FLOAT16 | [1] | use float16 for images instead of float32
|
||||
PTX | [1] | enable the specialized [PTX](https://docs.nvidia.com/cuda/parallel-thread-execution/) assembler for Nvidia GPUs. If not set, defaults to generic CUDA codegen backend.
|
||||
PROFILE | [1] | enable output of [perfetto](https://ui.perfetto.dev/) compatible profile. This feature is supported in NV and AMD backends.
|
||||
VISIBLE_DEVICES | [list[int]]| restricts the NV/AMD devices that are available. The format is a comma-separated list of identifiers (indexing starts with 0).
|
||||
JIT | [0-2] | 0=disabled, 1=[jit enabled](quickstart.md#jit) (default), 2=jit enabled, but graphs are disabled
|
||||
VIZ | [1] | 0=disabled, 1=[viz enabled](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/viz)
|
||||
@@ -1,53 +0,0 @@
|
||||
# tinygrad documentation
|
||||
|
||||
Welcome to the docs for tinygrad. This page is for users of the tinygrad library. tinygrad is not 1.0 yet, but it will be soon. The API has been pretty stable for a while.
|
||||
|
||||
While you can `pip install tinygrad`, we encourage you to install from source:
|
||||
|
||||
```bash
|
||||
git clone https://github.com/tinygrad/tinygrad.git
|
||||
cd tinygrad
|
||||
python3 -m pip install -e .
|
||||
```
|
||||
|
||||
After you have installed tinygrad, try the [MNIST tutorial](mnist.md).
|
||||
|
||||
If you are new to tensor libraries, learn how to use them by solving puzzles from [tinygrad-tensor-puzzles](https://github.com/obadakhalili/tinygrad-tensor-puzzles).
|
||||
|
||||
We also have [developer docs](developer/developer.md), and Di Zhu has created a [bunch of tutorials](https://mesozoic-egg.github.io/tinygrad-notes/) to help understand how tinygrad works.
|
||||
|
||||
## tinygrad Usage
|
||||
|
||||
The main class you will interact with is [Tensor](tensor/index.md). It functions very similarly to PyTorch, but has a bit more of a functional style. tinygrad supports [many datatypes](dtypes.md). All operations in tinygrad are lazy, meaning they won't do anything until you realize.
|
||||
|
||||
* tinygrad has a built in [neural network library](nn.md) with some classes, optimizers, and load/save state management.
|
||||
* tinygrad has a JIT to make things fast. Decorate your pure function with `TinyJit`
|
||||
* tinygrad has amazing support for multiple GPUs, allowing you to shard your Tensors with `Tensor.shard`
|
||||
|
||||
To understand what training looks like in tinygrad, you should read `beautiful_mnist.py`
|
||||
|
||||
We have a [quickstart guide](quickstart.md) and a [showcase](showcase.md)
|
||||
|
||||
## tinygrad Stack
|
||||
|
||||
<img src="./tinygrad_vs_others.png" alt="Tinygrad vs others" style="max-width: 1000px; height: auto;" />
|
||||
|
||||
## Differences from PyTorch
|
||||
|
||||
If you are migrating from PyTorch, welcome. Most of the API is the same. We hope you will find tinygrad both familiar and somehow more "correct feeling"
|
||||
|
||||
### tinygrad doesn't have nn.Module
|
||||
|
||||
There's nothing special about a "Module" class in tinygrad, it's just a normal class. [`nn.state.get_parameters`](nn.md/#tinygrad.nn.state.get_parameters) can be used to recursively search normal classes for valid tensors. Instead of the `forward` method in PyTorch, tinygrad just uses `__call__`
|
||||
|
||||
### tinygrad is functional
|
||||
|
||||
In tinygrad, you can do [`x.conv2d(w, b)`](tensor/ops.md/#tinygrad.Tensor.conv2d) or [`x.sparse_categorical_cross_entropy(y)`](tensor/ops.md/#tinygrad.Tensor.sparse_categorical_crossentropy). We do also have a [`Conv2D`](nn.md/#tinygrad.nn.Conv2d) class like PyTorch if you want a place to keep the state, but all stateless operations don't have classes.
|
||||
|
||||
### tinygrad is lazy
|
||||
|
||||
When you do `a+b` in tinygrad, nothing happens. It's not until you [`realize`](tensor/properties.md#tinygrad.Tensor.realize) the Tensor that the computation actually runs.
|
||||
|
||||
### tinygrad requires @TinyJit to be fast
|
||||
|
||||
PyTorch spends a lot of development effort to make dispatch very fast. tinygrad doesn't. We have a simple decorator that will replay the kernels used in the decorated function.
|
||||
-185
@@ -1,185 +0,0 @@
|
||||
# MNIST Tutorial
|
||||
|
||||
After you have installed tinygrad, this is a great first tutorial.
|
||||
|
||||
Start up a notebook locally, or use [colab](https://colab.research.google.com/). tinygrad is very lightweight, so it's easy to install anywhere and doesn't need a special colab image, but for speed we recommend a T4 GPU image.
|
||||
|
||||
### One-liner to install tinygrad in colab
|
||||
|
||||
```python
|
||||
!pip install git+https://github.com/tinygrad/tinygrad.git
|
||||
```
|
||||
|
||||
### What's the default device?
|
||||
|
||||
```python
|
||||
from tinygrad import Device
|
||||
print(Device.DEFAULT)
|
||||
```
|
||||
|
||||
You will see `CUDA` here on a GPU instance, or `CLANG` here on a CPU instance.
|
||||
|
||||
## A simple model
|
||||
|
||||
We'll use the model from [the Keras tutorial](https://keras.io/examples/vision/mnist_convnet/).
|
||||
|
||||
```python
|
||||
from tinygrad import Tensor, nn
|
||||
|
||||
class Model:
|
||||
def __init__(self):
|
||||
self.l1 = nn.Conv2d(1, 32, kernel_size=(3,3))
|
||||
self.l2 = nn.Conv2d(32, 64, kernel_size=(3,3))
|
||||
self.l3 = nn.Linear(1600, 10)
|
||||
|
||||
def __call__(self, x:Tensor) -> Tensor:
|
||||
x = self.l1(x).relu().max_pool2d((2,2))
|
||||
x = self.l2(x).relu().max_pool2d((2,2))
|
||||
return self.l3(x.flatten(1).dropout(0.5))
|
||||
```
|
||||
|
||||
Two key differences from PyTorch:
|
||||
|
||||
* Only the stateful layers are declared in `__init__`
|
||||
* There's no `nn.Module` class or `forward` function, just a normal class and `__call__`
|
||||
|
||||
### Getting the dataset
|
||||
|
||||
```python
|
||||
from tinygrad.nn.datasets import mnist
|
||||
X_train, Y_train, X_test, Y_test = mnist()
|
||||
print(X_train.shape, X_train.dtype, Y_train.shape, Y_train.dtype)
|
||||
# (60000, 1, 28, 28) dtypes.uchar (60000,) dtypes.uchar
|
||||
```
|
||||
|
||||
tinygrad includes MNIST, it only adds four lines. Feel free to read the [function](https://github.com/tinygrad/tinygrad/blob/master/tinygrad/nn/datasets.py).
|
||||
|
||||
## Using the model
|
||||
|
||||
MNIST is small enough that the `mnist()` function copies the dataset to the default device.
|
||||
|
||||
So creating the model and evaluating it is a matter of:
|
||||
|
||||
```python
|
||||
model = Model()
|
||||
acc = (model(X_test).argmax(axis=1) == Y_test).mean()
|
||||
# NOTE: tinygrad is lazy, and hasn't actually run anything by this point
|
||||
print(acc.item()) # ~10% accuracy, as expected from a random model
|
||||
```
|
||||
|
||||
### Training the model
|
||||
|
||||
We'll use the Adam optimizer. The `nn.state.get_parameters` will walk the model class and pull out the parameters for the optimizer. Also, in tinygrad, it's typical to write a function to do the training step so it can be jitted.
|
||||
|
||||
```python
|
||||
optim = nn.optim.Adam(nn.state.get_parameters(model))
|
||||
batch_size = 128
|
||||
def step():
|
||||
Tensor.training = True # makes dropout work
|
||||
samples = Tensor.randint(batch_size, high=X_train.shape[0])
|
||||
X, Y = X_train[samples], Y_train[samples]
|
||||
optim.zero_grad()
|
||||
loss = model(X).sparse_categorical_crossentropy(Y).backward()
|
||||
optim.step()
|
||||
return loss
|
||||
```
|
||||
|
||||
You can time a step with:
|
||||
|
||||
```python
|
||||
import timeit
|
||||
timeit.repeat(step, repeat=5, number=1)
|
||||
#[0.08268719699981375,
|
||||
# 0.07478952900009972,
|
||||
# 0.07714716600003158,
|
||||
# 0.07785399599970333,
|
||||
# 0.07605237000007037]
|
||||
```
|
||||
|
||||
So around 75 ms on T4 colab.
|
||||
|
||||
If you want to see a breakdown of the time by kernel:
|
||||
|
||||
```python
|
||||
from tinygrad import GlobalCounters, Context
|
||||
GlobalCounters.reset()
|
||||
with Context(DEBUG=2): step()
|
||||
```
|
||||
|
||||
### Why so slow?
|
||||
|
||||
Unlike PyTorch, tinygrad isn't designed to be fast like that. While 75 ms for one step is plenty fast for debugging, it's not great for training. Here, we introduce the first quintessentially tinygrad concept, the `TinyJit`.
|
||||
|
||||
```python
|
||||
from tinygrad import TinyJit
|
||||
jit_step = TinyJit(step)
|
||||
```
|
||||
|
||||
NOTE: It can also be used as a decorator `@TinyJit`
|
||||
|
||||
Now when we time it:
|
||||
|
||||
```python
|
||||
import timeit
|
||||
timeit.repeat(jit_step, repeat=5, number=1)
|
||||
# [0.2596786549997887,
|
||||
# 0.08989566299987928,
|
||||
# 0.0012115650001760514,
|
||||
# 0.001010227999813651,
|
||||
# 0.0012164899999334011]
|
||||
```
|
||||
|
||||
1.0 ms is 75x faster! Note that we aren't syncing the GPU, so GPU time may be slower.
|
||||
|
||||
The slowness the first two times is the JIT capturing the kernels. And this JIT will not run any Python in the function, it will just replay the tinygrad kernels that were run, so be aware that non tinygrad Python operations won't work. Randomness functions work as expected.
|
||||
|
||||
Unlike other JITs, we JIT everything, including the optimizer. Think of it as a dumb replay on different data.
|
||||
|
||||
## Putting it together
|
||||
|
||||
Since we are just randomly sampling from the dataset, there's no real concept of an epoch. We have a batch size of 128, so the Keras example is taking about 7000 steps.
|
||||
|
||||
```python
|
||||
for step in range(7000):
|
||||
loss = jit_step()
|
||||
if step%100 == 0:
|
||||
Tensor.training = False
|
||||
acc = (model(X_test).argmax(axis=1) == Y_test).mean().item()
|
||||
print(f"step {step:4d}, loss {loss.item():.2f}, acc {acc*100.:.2f}%")
|
||||
```
|
||||
|
||||
It doesn't take long to reach 98%, and it usually reaches 99%.
|
||||
|
||||
```
|
||||
step 0, loss 4.03, acc 71.43%
|
||||
step 100, loss 0.34, acc 93.86%
|
||||
step 200, loss 0.23, acc 95.97%
|
||||
step 300, loss 0.18, acc 96.32%
|
||||
step 400, loss 0.18, acc 96.76%
|
||||
step 500, loss 0.13, acc 97.46%
|
||||
step 600, loss 0.14, acc 97.45%
|
||||
step 700, loss 0.10, acc 97.27%
|
||||
step 800, loss 0.23, acc 97.49%
|
||||
step 900, loss 0.13, acc 97.51%
|
||||
step 1000, loss 0.13, acc 97.88%
|
||||
step 1100, loss 0.11, acc 97.72%
|
||||
step 1200, loss 0.14, acc 97.65%
|
||||
step 1300, loss 0.12, acc 98.04%
|
||||
step 1400, loss 0.25, acc 98.17%
|
||||
step 1500, loss 0.11, acc 97.86%
|
||||
step 1600, loss 0.21, acc 98.21%
|
||||
step 1700, loss 0.14, acc 98.34%
|
||||
...
|
||||
```
|
||||
|
||||
## From here?
|
||||
|
||||
tinygrad is yours to play with now. It's pure Python and short, so unlike PyTorch, fixing library bugs is well within your abilities.
|
||||
|
||||
- It's two lines to add multiGPU support to this example (can you find them?). You have to `.shard` the model to all GPUs, and `.shard` the dataset by batch.
|
||||
- `with Context(DEBUG=2)` shows the running kernels, `DEBUG=4` shows the code. All `Context` variables can also be environment variables.
|
||||
- `with Context(BEAM=2)` will do a BEAM search on the kernels, searching many possible implementations for what runs the fastest on your hardware. After this search, tinygrad is usually speed competitive with PyTorch, and the results are cached so you won't have to search next time.
|
||||
|
||||
[Join our Discord](https://discord.gg/ZjZadyC7PK) for help, and if you want to be a tinygrad developer. Please read the Discord rules when you get there.
|
||||
|
||||
[Follow us on Twitter](https://twitter.com/__tinygrad__) to keep up with the project.
|
||||
-32
@@ -1,32 +0,0 @@
|
||||
## Neural Network classes
|
||||
|
||||
::: tinygrad.nn.BatchNorm
|
||||
::: tinygrad.nn.Conv1d
|
||||
::: tinygrad.nn.Conv2d
|
||||
::: tinygrad.nn.ConvTranspose1d
|
||||
::: tinygrad.nn.ConvTranspose2d
|
||||
::: tinygrad.nn.Linear
|
||||
::: tinygrad.nn.GroupNorm
|
||||
::: tinygrad.nn.InstanceNorm
|
||||
::: tinygrad.nn.LayerNorm
|
||||
::: tinygrad.nn.LayerNorm2d
|
||||
::: tinygrad.nn.RMSNorm
|
||||
::: tinygrad.nn.Embedding
|
||||
::: tinygrad.nn.LSTMCell
|
||||
|
||||
## Optimizers
|
||||
|
||||
::: tinygrad.nn.optim.SGD
|
||||
::: tinygrad.nn.optim.LARS
|
||||
::: tinygrad.nn.optim.AdamW
|
||||
::: tinygrad.nn.optim.Adam
|
||||
::: tinygrad.nn.optim.LAMB
|
||||
|
||||
## Load/Save
|
||||
|
||||
::: tinygrad.nn.state.safe_load
|
||||
::: tinygrad.nn.state.safe_save
|
||||
::: tinygrad.nn.state.get_state_dict
|
||||
::: tinygrad.nn.state.get_parameters
|
||||
::: tinygrad.nn.state.load_state_dict
|
||||
::: tinygrad.nn.state.torch_load
|
||||
@@ -1,305 +0,0 @@
|
||||
# Quick Start Guide
|
||||
|
||||
This guide assumes no prior knowledge of pytorch or any other deep learning framework, but does assume some basic knowledge of neural networks.
|
||||
It is intended to be a very quick overview of the high level API that tinygrad provides.
|
||||
|
||||
This guide is also structured as a tutorial which at the end of it you will have a working model that can classify handwritten digits.
|
||||
|
||||
We need some imports to get started:
|
||||
|
||||
```python
|
||||
import numpy as np
|
||||
from tinygrad.helpers import Timing
|
||||
```
|
||||
|
||||
## Tensors
|
||||
|
||||
Tensors are the base data structure in tinygrad. They can be thought of as a multidimensional array of a specific data type.
|
||||
All high level operations in tinygrad operate on these tensors.
|
||||
|
||||
The tensor class can be imported like so:
|
||||
|
||||
```python
|
||||
from tinygrad import Tensor
|
||||
```
|
||||
|
||||
Tensors can be created from an existing data structure like a python list or numpy ndarray:
|
||||
|
||||
```python
|
||||
t1 = Tensor([1, 2, 3, 4, 5])
|
||||
na = np.array([1, 2, 3, 4, 5])
|
||||
t2 = Tensor(na)
|
||||
```
|
||||
|
||||
Tensors can also be created using one of the many factory methods:
|
||||
|
||||
```python
|
||||
full = Tensor.full(shape=(2, 3), fill_value=5) # create a tensor of shape (2, 3) filled with 5
|
||||
zeros = Tensor.zeros(2, 3) # create a tensor of shape (2, 3) filled with 0
|
||||
ones = Tensor.ones(2, 3) # create a tensor of shape (2, 3) filled with 1
|
||||
|
||||
full_like = Tensor.full_like(full, fill_value=2) # create a tensor of the same shape as `full` filled with 2
|
||||
zeros_like = Tensor.zeros_like(full) # create a tensor of the same shape as `full` filled with 0
|
||||
ones_like = Tensor.ones_like(full) # create a tensor of the same shape as `full` filled with 1
|
||||
|
||||
eye = Tensor.eye(3) # create a 3x3 identity matrix
|
||||
arange = Tensor.arange(start=0, stop=10, step=1) # create a tensor of shape (10,) filled with values from 0 to 9
|
||||
|
||||
rand = Tensor.rand(2, 3) # create a tensor of shape (2, 3) filled with random values from a uniform distribution
|
||||
randn = Tensor.randn(2, 3) # create a tensor of shape (2, 3) filled with random values from a standard normal distribution
|
||||
uniform = Tensor.uniform(2, 3, low=0, high=10) # create a tensor of shape (2, 3) filled with random values from a uniform distribution between 0 and 10
|
||||
```
|
||||
|
||||
There are even more of these factory methods, you can find them in the [Tensor Creation](tensor/creation.md) file.
|
||||
|
||||
All the tensors creation methods can take a `dtype` argument to specify the data type of the tensor, find the supported `dtype` in [dtypes](dtypes.md).
|
||||
|
||||
```python
|
||||
from tinygrad import dtypes
|
||||
|
||||
t3 = Tensor([1, 2, 3, 4, 5], dtype=dtypes.int32)
|
||||
```
|
||||
|
||||
Tensors allow you to perform operations on them like so:
|
||||
|
||||
```python
|
||||
t4 = Tensor([1, 2, 3, 4, 5])
|
||||
t5 = (t4 + 1) * 2
|
||||
t6 = (t5 * t4).relu().log_softmax()
|
||||
```
|
||||
|
||||
All of these operations are lazy and are only executed when you realize the tensor using `.realize()` or `.numpy()`.
|
||||
|
||||
```python
|
||||
print(t6.numpy())
|
||||
# [-56. -48. -36. -20. 0.]
|
||||
```
|
||||
|
||||
There are a lot more operations that can be performed on tensors, you can find them in the [Tensor Ops](tensor/ops.md) file.
|
||||
Additionally reading through [abstractions2.py](https://github.com/tinygrad/tinygrad/blob/master/docs/abstractions2.py) will help you understand how operations on these tensors make their way down to your hardware.
|
||||
|
||||
## Models
|
||||
|
||||
Neural networks in tinygrad are really just represented by the operations performed on tensors.
|
||||
These operations are commonly grouped into the `__call__` method of a class which allows modularization and reuse of these groups of operations.
|
||||
These classes do not need to inherit from any base class, in fact if they don't need any trainable parameters they don't even need to be a class!
|
||||
|
||||
An example of this would be the `nn.Linear` class which represents a linear layer in a neural network.
|
||||
|
||||
```python
|
||||
class Linear:
|
||||
def __init__(self, in_features, out_features, bias=True, initialization: str='kaiming_uniform'):
|
||||
self.weight = getattr(Tensor, initialization)(out_features, in_features)
|
||||
self.bias = Tensor.zeros(out_features) if bias else None
|
||||
|
||||
def __call__(self, x):
|
||||
return x.linear(self.weight.transpose(), self.bias)
|
||||
```
|
||||
|
||||
There are more neural network modules already implemented in [nn](nn.md), and you can also implement your own.
|
||||
|
||||
We will be implementing a simple neural network that can classify handwritten digits from the MNIST dataset.
|
||||
Our classifier will be a simple 2 layer neural network with a Leaky ReLU activation function.
|
||||
It will use a hidden layer size of 128 and an output layer size of 10 (one for each digit) with no bias on either Linear layer.
|
||||
|
||||
```python
|
||||
class TinyNet:
|
||||
def __init__(self):
|
||||
self.l1 = Linear(784, 128, bias=False)
|
||||
self.l2 = Linear(128, 10, bias=False)
|
||||
|
||||
def __call__(self, x):
|
||||
x = self.l1(x)
|
||||
x = x.leakyrelu()
|
||||
x = self.l2(x)
|
||||
return x
|
||||
|
||||
net = TinyNet()
|
||||
```
|
||||
|
||||
We can see that the forward pass of our neural network is just the sequence of operations performed on the input tensor `x`.
|
||||
We can also see that functional operations like `leakyrelu` are not defined as classes and instead are just methods we can just call.
|
||||
Finally, we just initialize an instance of our neural network, and we are ready to start training it.
|
||||
|
||||
## Training
|
||||
|
||||
Now that we have our neural network defined we can start training it.
|
||||
Training neural networks in tinygrad is super simple.
|
||||
All we need to do is define our neural network, define our loss function, and then call `.backward()` on the loss function to compute the gradients.
|
||||
They can then be used to update the parameters of our neural network using one of the many [Optimizers](nn.md#optimizers).
|
||||
|
||||
For our loss function we will be using sparse categorical cross entropy loss. The implementation below is taken from [tensor.py](https://github.com/tinygrad/tinygrad/blob/master/tinygrad/tensor.py), it's copied below to highlight an important detail of tinygrad.
|
||||
|
||||
```python
|
||||
def sparse_categorical_crossentropy(self, Y, ignore_index=-1) -> Tensor:
|
||||
loss_mask = Y != ignore_index
|
||||
y_counter = Tensor.arange(self.shape[-1], dtype=dtypes.int32, requires_grad=False, device=self.device).unsqueeze(0).expand(Y.numel(), self.shape[-1])
|
||||
y = ((y_counter == Y.flatten().reshape(-1, 1)).where(-1.0, 0) * loss_mask.reshape(-1, 1)).reshape(*Y.shape, self.shape[-1])
|
||||
return self.log_softmax().mul(y).sum() / loss_mask.sum()
|
||||
```
|
||||
|
||||
As we can see in this implementation of cross entropy loss, there are certain operations that tinygrad does not support natively.
|
||||
Load/store ops are not supported in tinygrad natively because they add complexity when trying to port to different backends, 90% of the models out there don't use/need them, and they can be implemented like it's done above with an `arange` mask.
|
||||
|
||||
For our optimizer we will be using the traditional stochastic gradient descent optimizer with a learning rate of 3e-4.
|
||||
|
||||
```python
|
||||
from tinygrad.nn.optim import SGD
|
||||
|
||||
opt = SGD([net.l1.weight, net.l2.weight], lr=3e-4)
|
||||
```
|
||||
|
||||
We can see that we are passing in the parameters of our neural network to the optimizer.
|
||||
This is due to the fact that the optimizer needs to know which parameters to update.
|
||||
There is a simpler way to do this just by using `get_parameters(net)` from `tinygrad.nn.state` which will return a list of all the parameters in the neural network.
|
||||
The parameters are just listed out explicitly here for clarity.
|
||||
|
||||
Now that we have our network, loss function, and optimizer defined all we are missing is the data to train on!
|
||||
There are a couple of dataset loaders in tinygrad located in [/extra/datasets](https://github.com/tinygrad/tinygrad/blob/master/extra/datasets).
|
||||
We will be using the MNIST dataset loader.
|
||||
|
||||
```python
|
||||
from extra.datasets import fetch_mnist
|
||||
```
|
||||
|
||||
Now we have everything we need to start training our neural network.
|
||||
We will be training for 1000 steps with a batch size of 64.
|
||||
|
||||
We use `with Tensor.train()` to set the internal flag `Tensor.training` to `True` during training.
|
||||
Upon exit, the flag is restored to its previous value by the context manager.
|
||||
|
||||
```python
|
||||
X_train, Y_train, X_test, Y_test = fetch_mnist()
|
||||
|
||||
with Tensor.train():
|
||||
for step in range(1000):
|
||||
# random sample a batch
|
||||
samp = np.random.randint(0, X_train.shape[0], size=(64))
|
||||
batch = Tensor(X_train[samp], requires_grad=False)
|
||||
# get the corresponding labels
|
||||
labels = Tensor(Y_train[samp])
|
||||
|
||||
# forward pass
|
||||
out = net(batch)
|
||||
|
||||
# compute loss
|
||||
loss = sparse_categorical_crossentropy(out, labels)
|
||||
|
||||
# zero gradients
|
||||
opt.zero_grad()
|
||||
|
||||
# backward pass
|
||||
loss.backward()
|
||||
|
||||
# update parameters
|
||||
opt.step()
|
||||
|
||||
# calculate accuracy
|
||||
pred = out.argmax(axis=-1)
|
||||
acc = (pred == labels).mean()
|
||||
|
||||
if step % 100 == 0:
|
||||
print(f"Step {step+1} | Loss: {loss.numpy()} | Accuracy: {acc.numpy()}")
|
||||
```
|
||||
|
||||
## Evaluation
|
||||
|
||||
Now that we have trained our neural network we can evaluate it on the test set.
|
||||
We will be using the same batch size of 64 and will be evaluating for 1000 of those batches.
|
||||
|
||||
```python
|
||||
with Timing("Time: "):
|
||||
avg_acc = 0
|
||||
for step in range(1000):
|
||||
# random sample a batch
|
||||
samp = np.random.randint(0, X_test.shape[0], size=(64))
|
||||
batch = Tensor(X_test[samp], requires_grad=False)
|
||||
# get the corresponding labels
|
||||
labels = Y_test[samp]
|
||||
|
||||
# forward pass
|
||||
out = net(batch)
|
||||
|
||||
# calculate accuracy
|
||||
pred = out.argmax(axis=-1).numpy()
|
||||
avg_acc += (pred == labels).mean()
|
||||
print(f"Test Accuracy: {avg_acc / 1000}")
|
||||
```
|
||||
|
||||
## And that's it
|
||||
|
||||
Highly recommend you check out the [examples/](https://github.com/tinygrad/tinygrad/blob/master/examples) folder for more examples of using tinygrad.
|
||||
Reading the source code of tinygrad is also a great way to learn how it works.
|
||||
Specifically the tests in [test/](https://github.com/tinygrad/tinygrad/blob/master/test) are a great place to see how to use and the semantics of the different operations.
|
||||
There are also a bunch of models implemented in [models/](https://github.com/tinygrad/tinygrad/blob/master/extra/models) that you can use as a reference.
|
||||
|
||||
Additionally, feel free to ask questions in the `#learn-tinygrad` channel on the [discord](https://discord.gg/beYbxwxVdx). Don't ask to ask, just ask!
|
||||
|
||||
## Extras
|
||||
|
||||
### JIT
|
||||
|
||||
Additionally, it is possible to speed up the computation of certain neural networks by using the JIT.
|
||||
Currently, this does not support models with varying input sizes and non tinygrad operations.
|
||||
|
||||
To use the JIT we just need to add a function decorator to the forward pass of our neural network and ensure that the input and output are realized tensors.
|
||||
Or in this case we will create a wrapper function and decorate the wrapper function to speed up the evaluation of our neural network.
|
||||
|
||||
```python
|
||||
from tinygrad import TinyJit
|
||||
|
||||
@TinyJit
|
||||
def jit(x):
|
||||
return net(x).realize()
|
||||
|
||||
with Timing("Time: "):
|
||||
avg_acc = 0
|
||||
for step in range(1000):
|
||||
# random sample a batch
|
||||
samp = np.random.randint(0, X_test.shape[0], size=(64))
|
||||
batch = Tensor(X_test[samp], requires_grad=False)
|
||||
# get the corresponding labels
|
||||
labels = Y_test[samp]
|
||||
|
||||
# forward pass with jit
|
||||
out = jit(batch)
|
||||
|
||||
# calculate accuracy
|
||||
pred = out.argmax(axis=-1).numpy()
|
||||
avg_acc += (pred == labels).mean()
|
||||
print(f"Test Accuracy: {avg_acc / 1000}")
|
||||
```
|
||||
|
||||
You will find that the evaluation time is much faster than before and that your accelerator utilization is much higher.
|
||||
|
||||
### Saving and Loading Models
|
||||
|
||||
The standard weight format for tinygrad is [safetensors](https://github.com/huggingface/safetensors). This means that you can load the weights of any model also using safetensors into tinygrad.
|
||||
There are functions in [state.py](https://github.com/tinygrad/tinygrad/blob/master/tinygrad/nn/state.py) to save and load models to and from this format.
|
||||
|
||||
```python
|
||||
from tinygrad.nn.state import safe_save, safe_load, get_state_dict, load_state_dict
|
||||
|
||||
# first we need the state dict of our model
|
||||
state_dict = get_state_dict(net)
|
||||
|
||||
# then we can just save it to a file
|
||||
safe_save(state_dict, "model.safetensors")
|
||||
|
||||
# and load it back in
|
||||
state_dict = safe_load("model.safetensors")
|
||||
load_state_dict(net, state_dict)
|
||||
```
|
||||
|
||||
Many of the models in the [models/](https://github.com/tinygrad/tinygrad/tree/master/extra/models) folder have a `load_from_pretrained` method that will download and load the weights for you. These usually are pytorch weights meaning that you would need pytorch installed to load them.
|
||||
|
||||
### Environment Variables
|
||||
|
||||
There exist a bunch of environment variables that control the runtime behavior of tinygrad.
|
||||
Some of the commons ones are `DEBUG` and the different backend enablement variables.
|
||||
|
||||
You can find a full list and their descriptions in [env_vars.md](env_vars.md).
|
||||
|
||||
### Visualizing the Computation Graph
|
||||
|
||||
It is possible to visualize the computation graph of a neural network using VIZ=1.
|
||||
@@ -1,14 +0,0 @@
|
||||
# Runtimes
|
||||
|
||||
tinygrad supports various runtimes, enabling your code to scale across a wide range of devices. The default runtime can be automatically selected based on the available hardware, or you can force a specific runtime to be default using environment variables (e.g., `CLANG=1`).
|
||||
|
||||
| Runtime | Description | Requirements |
|
||||
|---------|-------------|--------------|
|
||||
| [NV](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_nv.py) | Provides acceleration for NVIDIA GPUs | Ampere/Ada series GPUs |
|
||||
| [AMD](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_amd.py) | Provides acceleration for AMD GPUs | RDNA2/RDNA3 series GPUs |
|
||||
| [QCOM](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_qcom.py) | Provides acceleration for QCOM GPUs | 6xx series GPUs |
|
||||
| [METAL](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_metal.py) | Utilizes Metal for acceleration on Apple devices | M1+ Macs; Metal 3.0+ for `bfloat` support |
|
||||
| [CUDA](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_cuda.py) | Utilizes CUDA for acceleration on NVIDIA GPUs | NVIDIA GPU with CUDA support |
|
||||
| [GPU (OpenCL)](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_gpu.py) | Accelerates computations using OpenCL on GPUs | OpenCL 2.0 compatible device |
|
||||
| [CLANG (C Code)](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_clang.py) | Runs on CPU using the clang compiler | `clang` compiler in system `PATH` |
|
||||
| [LLVM](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/runtime/ops_llvm.py) | Runs on CPU using the LLVM compiler infrastructure | `llvmlite` package installed |
|
||||
@@ -1,62 +0,0 @@
|
||||
# Showcase
|
||||
|
||||
Despite being a tiny library, tinygrad is capable of doing a lot of things. From state-of-the-art [vision](https://arxiv.org/abs/1905.11946) to state-of-the-art [language](https://arxiv.org/abs/1706.03762) models.
|
||||
|
||||
## Vision
|
||||
|
||||
### EfficientNet
|
||||
|
||||
You can either pass in the URL of a picture to discover what it is:
|
||||
```sh
|
||||
python3 examples/efficientnet.py ./test/models/efficientnet/Chicken.jpg
|
||||
```
|
||||
Or, if you have a camera and OpenCV installed, you can detect what is in front of you:
|
||||
```sh
|
||||
python3 examples/efficientnet.py webcam
|
||||
```
|
||||
|
||||
### YOLOv8
|
||||
|
||||
Take a look at [yolov8.py](https://github.com/tinygrad/tinygrad/tree/master/examples/yolov8.py).
|
||||
|
||||
|
||||
|
||||
## Audio
|
||||
|
||||
### Whisper
|
||||
|
||||
Take a look at [whisper.py](https://github.com/tinygrad/tinygrad/tree/master/examples/whisper.py). You need pyaudio and torchaudio installed.
|
||||
|
||||
```sh
|
||||
SMALL=1 python3 examples/whisper.py
|
||||
```
|
||||
|
||||
## Generative
|
||||
|
||||
### Stable Diffusion
|
||||
|
||||
```sh
|
||||
python3 examples/stable_diffusion.py
|
||||
```
|
||||
|
||||
|
||||
|
||||
*"a horse sized cat eating a bagel"*
|
||||
|
||||
### LLaMA
|
||||
|
||||
You will need to download and put the weights into the `weights/LLaMA` directory, which may need to be created.
|
||||
|
||||
Then you can have a chat with Stacy:
|
||||
```sh
|
||||
python3 examples/llama.py
|
||||
```
|
||||
|
||||
### Conversation
|
||||
|
||||
Make sure you have espeak installed and `PHONEMIZER_ESPEAK_LIBRARY` set.
|
||||
|
||||
Then you can talk to Stacy:
|
||||
```sh
|
||||
python3 examples/conversation.py
|
||||
```
|
||||
@@ -1,30 +0,0 @@
|
||||
## Creation (basic)
|
||||
|
||||
::: tinygrad.Tensor.empty
|
||||
::: tinygrad.Tensor.zeros
|
||||
::: tinygrad.Tensor.ones
|
||||
::: tinygrad.Tensor.full
|
||||
::: tinygrad.Tensor.arange
|
||||
::: tinygrad.Tensor.linspace
|
||||
::: tinygrad.Tensor.eye
|
||||
::: tinygrad.Tensor.full_like
|
||||
::: tinygrad.Tensor.zeros_like
|
||||
::: tinygrad.Tensor.ones_like
|
||||
|
||||
## Creation (external)
|
||||
|
||||
::: tinygrad.Tensor.from_blob
|
||||
::: tinygrad.Tensor.from_url
|
||||
|
||||
## Creation (random)
|
||||
|
||||
::: tinygrad.Tensor.manual_seed
|
||||
::: tinygrad.Tensor.rand
|
||||
::: tinygrad.Tensor.randn
|
||||
::: tinygrad.Tensor.randint
|
||||
::: tinygrad.Tensor.normal
|
||||
::: tinygrad.Tensor.uniform
|
||||
::: tinygrad.Tensor.scaled_uniform
|
||||
::: tinygrad.Tensor.glorot_uniform
|
||||
::: tinygrad.Tensor.kaiming_uniform
|
||||
::: tinygrad.Tensor.kaiming_normal
|
||||
@@ -1,81 +0,0 @@
|
||||
Elementwise ops operate on a per element basis. They don't change the shape of the tensor.
|
||||
|
||||
## Unary Ops (math)
|
||||
|
||||
::: tinygrad.Tensor.logical_not
|
||||
::: tinygrad.Tensor.neg
|
||||
::: tinygrad.Tensor.log
|
||||
::: tinygrad.Tensor.log2
|
||||
::: tinygrad.Tensor.exp
|
||||
::: tinygrad.Tensor.exp2
|
||||
::: tinygrad.Tensor.sqrt
|
||||
::: tinygrad.Tensor.rsqrt
|
||||
::: tinygrad.Tensor.sin
|
||||
::: tinygrad.Tensor.cos
|
||||
::: tinygrad.Tensor.tan
|
||||
::: tinygrad.Tensor.asin
|
||||
::: tinygrad.Tensor.acos
|
||||
::: tinygrad.Tensor.atan
|
||||
::: tinygrad.Tensor.trunc
|
||||
::: tinygrad.Tensor.ceil
|
||||
::: tinygrad.Tensor.floor
|
||||
::: tinygrad.Tensor.round
|
||||
::: tinygrad.Tensor.isinf
|
||||
::: tinygrad.Tensor.isnan
|
||||
::: tinygrad.Tensor.lerp
|
||||
::: tinygrad.Tensor.square
|
||||
::: tinygrad.Tensor.clamp
|
||||
::: tinygrad.Tensor.clip
|
||||
::: tinygrad.Tensor.sign
|
||||
::: tinygrad.Tensor.abs
|
||||
::: tinygrad.Tensor.reciprocal
|
||||
|
||||
## Unary Ops (activation)
|
||||
|
||||
::: tinygrad.Tensor.relu
|
||||
::: tinygrad.Tensor.sigmoid
|
||||
::: tinygrad.Tensor.hardsigmoid
|
||||
::: tinygrad.Tensor.elu
|
||||
::: tinygrad.Tensor.celu
|
||||
::: tinygrad.Tensor.selu
|
||||
::: tinygrad.Tensor.swish
|
||||
::: tinygrad.Tensor.silu
|
||||
::: tinygrad.Tensor.relu6
|
||||
::: tinygrad.Tensor.hardswish
|
||||
::: tinygrad.Tensor.tanh
|
||||
::: tinygrad.Tensor.sinh
|
||||
::: tinygrad.Tensor.cosh
|
||||
::: tinygrad.Tensor.atanh
|
||||
::: tinygrad.Tensor.asinh
|
||||
::: tinygrad.Tensor.acosh
|
||||
::: tinygrad.Tensor.hardtanh
|
||||
::: tinygrad.Tensor.erf
|
||||
::: tinygrad.Tensor.gelu
|
||||
::: tinygrad.Tensor.quick_gelu
|
||||
::: tinygrad.Tensor.leakyrelu
|
||||
::: tinygrad.Tensor.mish
|
||||
::: tinygrad.Tensor.softplus
|
||||
::: tinygrad.Tensor.softsign
|
||||
|
||||
## Elementwise Ops (broadcasted)
|
||||
|
||||
::: tinygrad.Tensor.add
|
||||
::: tinygrad.Tensor.sub
|
||||
::: tinygrad.Tensor.mul
|
||||
::: tinygrad.Tensor.div
|
||||
::: tinygrad.Tensor.xor
|
||||
::: tinygrad.Tensor.lshift
|
||||
::: tinygrad.Tensor.rshift
|
||||
::: tinygrad.Tensor.pow
|
||||
::: tinygrad.Tensor.maximum
|
||||
::: tinygrad.Tensor.minimum
|
||||
::: tinygrad.Tensor.where
|
||||
|
||||
## Casting Ops
|
||||
|
||||
::: tinygrad.Tensor.cast
|
||||
::: tinygrad.Tensor.bitcast
|
||||
::: tinygrad.Tensor.float
|
||||
::: tinygrad.Tensor.half
|
||||
::: tinygrad.Tensor.int
|
||||
::: tinygrad.Tensor.bool
|
||||
@@ -1,7 +0,0 @@
|
||||
# Tensor
|
||||
|
||||
::: tinygrad.Tensor
|
||||
options:
|
||||
heading_level: 2
|
||||
members: false
|
||||
show_source: false
|
||||
@@ -1,26 +0,0 @@
|
||||
## Movement (low level)
|
||||
|
||||
::: tinygrad.Tensor.view
|
||||
::: tinygrad.Tensor.reshape
|
||||
::: tinygrad.Tensor.expand
|
||||
::: tinygrad.Tensor.permute
|
||||
::: tinygrad.Tensor.flip
|
||||
::: tinygrad.Tensor.shrink
|
||||
::: tinygrad.Tensor.pad
|
||||
|
||||
## Movement (high level)
|
||||
|
||||
::: tinygrad.Tensor.gather
|
||||
::: tinygrad.Tensor.cat
|
||||
::: tinygrad.Tensor.stack
|
||||
::: tinygrad.Tensor.repeat
|
||||
::: tinygrad.Tensor.repeat_interleave
|
||||
::: tinygrad.Tensor.split
|
||||
::: tinygrad.Tensor.chunk
|
||||
::: tinygrad.Tensor.meshgrid
|
||||
::: tinygrad.Tensor.squeeze
|
||||
::: tinygrad.Tensor.unsqueeze
|
||||
::: tinygrad.Tensor.T
|
||||
::: tinygrad.Tensor.transpose
|
||||
::: tinygrad.Tensor.flatten
|
||||
::: tinygrad.Tensor.unflatten
|
||||
@@ -1,49 +0,0 @@
|
||||
## Reduce
|
||||
|
||||
::: tinygrad.Tensor.sum
|
||||
::: tinygrad.Tensor.prod
|
||||
::: tinygrad.Tensor.max
|
||||
::: tinygrad.Tensor.min
|
||||
::: tinygrad.Tensor.any
|
||||
::: tinygrad.Tensor.all
|
||||
::: tinygrad.Tensor.mean
|
||||
::: tinygrad.Tensor.var
|
||||
::: tinygrad.Tensor.std
|
||||
::: tinygrad.Tensor.std_mean
|
||||
::: tinygrad.Tensor.softmax
|
||||
::: tinygrad.Tensor.log_softmax
|
||||
::: tinygrad.Tensor.logsumexp
|
||||
::: tinygrad.Tensor.logcumsumexp
|
||||
::: tinygrad.Tensor.argmax
|
||||
::: tinygrad.Tensor.argmin
|
||||
|
||||
## Processing
|
||||
|
||||
::: tinygrad.Tensor.avg_pool2d
|
||||
::: tinygrad.Tensor.max_pool2d
|
||||
::: tinygrad.Tensor.conv2d
|
||||
::: tinygrad.Tensor.conv_transpose2d
|
||||
::: tinygrad.Tensor.dot
|
||||
::: tinygrad.Tensor.matmul
|
||||
::: tinygrad.Tensor.einsum
|
||||
::: tinygrad.Tensor.cumsum
|
||||
::: tinygrad.Tensor.cummax
|
||||
::: tinygrad.Tensor.triu
|
||||
::: tinygrad.Tensor.tril
|
||||
::: tinygrad.Tensor.interpolate
|
||||
::: tinygrad.Tensor.scatter
|
||||
|
||||
## Neural Network (functional)
|
||||
|
||||
::: tinygrad.Tensor.linear
|
||||
::: tinygrad.Tensor.sequential
|
||||
::: tinygrad.Tensor.layernorm
|
||||
::: tinygrad.Tensor.batchnorm
|
||||
::: tinygrad.Tensor.dropout
|
||||
::: tinygrad.Tensor.one_hot
|
||||
::: tinygrad.Tensor.scaled_dot_product_attention
|
||||
::: tinygrad.Tensor.binary_crossentropy
|
||||
::: tinygrad.Tensor.binary_crossentropy_logits
|
||||
::: tinygrad.Tensor.sparse_categorical_crossentropy
|
||||
::: tinygrad.Tensor.cross_entropy
|
||||
::: tinygrad.Tensor.nll_loss
|
||||
@@ -1,34 +0,0 @@
|
||||
## Basic
|
||||
|
||||
::: tinygrad.Tensor.shape
|
||||
::: tinygrad.Tensor.dtype
|
||||
::: tinygrad.Tensor.device
|
||||
::: tinygrad.Tensor.ndim
|
||||
::: tinygrad.Tensor.numel
|
||||
::: tinygrad.Tensor.element_size
|
||||
::: tinygrad.Tensor.nbytes
|
||||
::: tinygrad.Tensor.is_floating_point
|
||||
::: tinygrad.Tensor.size
|
||||
|
||||
## Data Access
|
||||
|
||||
::: tinygrad.Tensor.data
|
||||
::: tinygrad.Tensor.item
|
||||
::: tinygrad.Tensor.tolist
|
||||
::: tinygrad.Tensor.numpy
|
||||
|
||||
## tinygrad ops
|
||||
|
||||
::: tinygrad.Tensor.schedule_with_vars
|
||||
::: tinygrad.Tensor.schedule
|
||||
::: tinygrad.Tensor.realize
|
||||
::: tinygrad.Tensor.replace
|
||||
::: tinygrad.Tensor.assign
|
||||
::: tinygrad.Tensor.detach
|
||||
::: tinygrad.Tensor.to
|
||||
::: tinygrad.Tensor.to_
|
||||
::: tinygrad.Tensor.shard
|
||||
::: tinygrad.Tensor.shard_
|
||||
::: tinygrad.Tensor.contiguous
|
||||
::: tinygrad.Tensor.contiguous_backward
|
||||
::: tinygrad.Tensor.backward
|
||||
@@ -1,54 +0,0 @@
|
||||
# tinybox
|
||||
|
||||
Although these docs live in tinygrad, they pertain to deep learning hardware sold by the tiny corp. tinyboxes are used heavily in tinygrad's CI, and are the best tested platform to use tinygrad with. They appeared running tinygrad on [MLPerf Training 4.0](https://public.tableau.com/views/MLCommons-Training_16993769118290/MLCommons-Training)
|
||||
|
||||
If you don't have a tinybox and you want one, see [tinygrad.org](https://tinygrad.org). If you don't want one, that's okay too.
|
||||
|
||||
## Welcome
|
||||
|
||||
Welcome to your tinybox! The tinybox is the universal system purpose-built for all AI infrastructure and workloads, from training to inference. The red box includes six 7900XTX GPUs, and the green box includes six 4090 GPUs. Whether you bought a red one or a green one, we want you to love it.
|
||||
|
||||
We don't have a stupid cloud service, you don't have to create a tiny account to set it up, and we aren't tracking how you use the box. We're just happy you bought one. This petaflop is your petaflop.
|
||||
|
||||
## Plugging it in
|
||||
|
||||
tinybox has two 1600W PSUs, which together exceed the capacity of most 120V household circuits. Fortunately, it comes with two plugs. You'll want to plug each plug into a different circuit. You can verify that they are different circuits by flipping the breaker and seeing what turns off. If you have at least a 120V 30A or 220V 20A circuit, you are welcome to use only that one.
|
||||
|
||||
You'll also want to connect the Ethernet port without a rubber stopper to your home network.
|
||||
|
||||
While it's designed primarily for the home or office, the tinybox is 12U rack mountable using [these rails](https://rackmountmart.store.turbify.net/26slidrailfo.html).
|
||||
|
||||
## Power limiting the box
|
||||
|
||||
While a tinybox should ideally be run without power limits, there are cases where you might want to run the box off of a single outlet.
|
||||
|
||||
In such cases, it is possible to power limit the box using the provided `power-limit` script, which will power limit all of the GPUs to a specified wattage.
|
||||
|
||||
`sudo power-limit 150` should be good to run off of a single 120V 15A outlet.
|
||||
|
||||
## Connecting to the box
|
||||
|
||||
tinybox ships with a relatively basic install of Ubuntu 22.04. To do initial setup, you can either plug in a VGA monitor and keyboard, or you can connect remotely to the machine using the BMC. The BMC IP and password are displayed on the screen.
|
||||
|
||||
`ipmitool -H <BMC IP> -U admin -P <BMC PW> -I lanplus sol activate`
|
||||
|
||||
The default username is `tiny` and the default password is `tiny`. Once you are logged in, you can add an SSH key to authorized keys to connect over SSH (on the normal IP). Exit `ipmitool` with `~.` after a newline.
|
||||
|
||||
The BMC also has a web interface you can use if you find that easier.
|
||||
|
||||
## Changing the BMC password
|
||||
|
||||
It is recommended that you change the BMC password after setting up the box, as the password on the screen is only the initial password.
|
||||
|
||||
If you do decide to change the BMC password and no longer want the initial password to be displayed, remove the `/root/.bmc_password` file.
|
||||
Reboot after making these changes or restart the `displayservice.service` service.
|
||||
|
||||
## What do I use it for?
|
||||
|
||||
The [default tinybox image](https://github.com/tinygrad/tinyos) ships with tinygrad and PyTorch. While we develop tinygrad, the box is universal hardware. Use whatever framework you desire, run notebooks, download demos, install more things, train, inference, live, laugh, love, you aren't paying per hour for this box so the only limit is your imagination.
|
||||
|
||||
## tinychat
|
||||
|
||||
Since LLMs are so popular, we ship with a built in tinygrad based chatbot using a LLaMA-3 finetune. Visit the IP (not the BMC IP) of your tinybox in a web browser on your computer or phone, and you'll find a friendly looking chat interface. This chatbot also provides an OpenAI compatible LLM API on that port, so you can script it.
|
||||
|
||||
The conversations you have with this chatbot are between you and your tinybox. Also, the history in the web app is saved on the client, not the tinybox.
|
||||
+4157
File diff suppressed because it is too large
Load Diff
+1763
File diff suppressed because it is too large
Load Diff
@@ -1,9 +0,0 @@
|
||||
import globals from "globals";
|
||||
import pluginJs from "@eslint/js";
|
||||
import pluginHtml from "eslint-plugin-html";
|
||||
|
||||
export default [
|
||||
{files: ["**/*.html"], plugins: {html: pluginHtml}, rules:{"max-len": ["error", {"code": 150}]}},
|
||||
{languageOptions: {globals: globals.browser}},
|
||||
pluginJs.configs.recommended,
|
||||
];
|
||||
@@ -1,129 +0,0 @@
|
||||
from typing import Tuple
|
||||
import time
|
||||
from tinygrad import Tensor, TinyJit, nn
|
||||
import gymnasium as gym
|
||||
from tinygrad.helpers import trange
|
||||
import numpy as np # TODO: remove numpy import
|
||||
|
||||
ENVIRONMENT_NAME = 'CartPole-v1'
|
||||
#ENVIRONMENT_NAME = 'LunarLander-v2'
|
||||
|
||||
#import examples.rl.lightupbutton
|
||||
#ENVIRONMENT_NAME = 'PressTheLightUpButton-v0'
|
||||
|
||||
# *** hyperparameters ***
|
||||
# https://github.com/llSourcell/Unity_ML_Agents/blob/master/docs/best-practices-ppo.md
|
||||
|
||||
BATCH_SIZE = 256
|
||||
ENTROPY_SCALE = 0.0005
|
||||
REPLAY_BUFFER_SIZE = 2000
|
||||
PPO_EPSILON = 0.2
|
||||
HIDDEN_UNITS = 32
|
||||
LEARNING_RATE = 1e-2
|
||||
TRAIN_STEPS = 5
|
||||
EPISODES = 40
|
||||
DISCOUNT_FACTOR = 0.99
|
||||
|
||||
class ActorCritic:
|
||||
def __init__(self, in_features, out_features, hidden_state=HIDDEN_UNITS):
|
||||
self.l1 = nn.Linear(in_features, hidden_state)
|
||||
self.l2 = nn.Linear(hidden_state, out_features)
|
||||
|
||||
self.c1 = nn.Linear(in_features, hidden_state)
|
||||
self.c2 = nn.Linear(hidden_state, 1)
|
||||
|
||||
def __call__(self, obs:Tensor) -> Tuple[Tensor, Tensor]:
|
||||
x = self.l1(obs).tanh()
|
||||
act = self.l2(x).log_softmax()
|
||||
x = self.c1(obs).relu()
|
||||
return act, self.c2(x)
|
||||
|
||||
def evaluate(model:ActorCritic, test_env:gym.Env) -> float:
|
||||
(obs, _), terminated, truncated = test_env.reset(), False, False
|
||||
total_rew = 0.0
|
||||
while not terminated and not truncated:
|
||||
act = model(Tensor(obs))[0].argmax().item()
|
||||
obs, rew, terminated, truncated, _ = test_env.step(act)
|
||||
total_rew += float(rew)
|
||||
return total_rew
|
||||
|
||||
if __name__ == "__main__":
|
||||
env = gym.make(ENVIRONMENT_NAME)
|
||||
|
||||
model = ActorCritic(env.observation_space.shape[0], int(env.action_space.n)) # type: ignore
|
||||
opt = nn.optim.Adam(nn.state.get_parameters(model), lr=LEARNING_RATE)
|
||||
|
||||
@TinyJit
|
||||
def train_step(x:Tensor, selected_action:Tensor, reward:Tensor, old_log_dist:Tensor) -> Tuple[Tensor, Tensor, Tensor]:
|
||||
with Tensor.train():
|
||||
log_dist, value = model(x)
|
||||
action_mask = (selected_action.reshape(-1, 1) == Tensor.arange(log_dist.shape[1]).reshape(1, -1).expand(selected_action.shape[0], -1)).float()
|
||||
|
||||
# get real advantage using the value function
|
||||
advantage = reward.reshape(-1, 1) - value
|
||||
masked_advantage = action_mask * advantage.detach()
|
||||
|
||||
# PPO
|
||||
ratios = (log_dist - old_log_dist).exp()
|
||||
unclipped_ratio = masked_advantage * ratios
|
||||
clipped_ratio = masked_advantage * ratios.clip(1-PPO_EPSILON, 1+PPO_EPSILON)
|
||||
action_loss = -unclipped_ratio.minimum(clipped_ratio).sum(-1).mean()
|
||||
|
||||
entropy_loss = (log_dist.exp() * log_dist).sum(-1).mean() # this encourages diversity
|
||||
critic_loss = advantage.square().mean()
|
||||
opt.zero_grad()
|
||||
(action_loss + entropy_loss*ENTROPY_SCALE + critic_loss).backward()
|
||||
opt.step()
|
||||
return action_loss.realize(), entropy_loss.realize(), critic_loss.realize()
|
||||
|
||||
@TinyJit
|
||||
def get_action(obs:Tensor) -> Tensor:
|
||||
# TODO: with no_grad
|
||||
Tensor.no_grad = True
|
||||
ret = model(obs)[0].exp().multinomial().realize()
|
||||
Tensor.no_grad = False
|
||||
return ret
|
||||
|
||||
st, steps = time.perf_counter(), 0
|
||||
Xn, An, Rn = [], [], []
|
||||
for episode_number in (t:=trange(EPISODES)):
|
||||
get_action.reset() # NOTE: if you don't reset the jit here it captures the wrong model on the first run through
|
||||
|
||||
obs:np.ndarray = env.reset()[0]
|
||||
rews, terminated, truncated = [], False, False
|
||||
# NOTE: we don't want to early stop since then the rewards are wrong for the last episode
|
||||
while not terminated and not truncated:
|
||||
# pick actions
|
||||
# TODO: what's the temperature here?
|
||||
act = get_action(Tensor(obs)).item()
|
||||
|
||||
# save this state action pair
|
||||
# TODO: don't use np.copy here on the CPU, what's the tinygrad way to do this and keep on device? need __setitem__ assignment
|
||||
Xn.append(np.copy(obs))
|
||||
An.append(act)
|
||||
|
||||
obs, rew, terminated, truncated, _ = env.step(act)
|
||||
rews.append(float(rew))
|
||||
steps += len(rews)
|
||||
|
||||
# reward to go
|
||||
# TODO: move this into tinygrad
|
||||
discounts = np.power(DISCOUNT_FACTOR, np.arange(len(rews)))
|
||||
Rn += [np.sum(rews[i:] * discounts[:len(rews)-i]) for i in range(len(rews))]
|
||||
|
||||
Xn, An, Rn = Xn[-REPLAY_BUFFER_SIZE:], An[-REPLAY_BUFFER_SIZE:], Rn[-REPLAY_BUFFER_SIZE:]
|
||||
X, A, R = Tensor(Xn), Tensor(An), Tensor(Rn)
|
||||
|
||||
# TODO: make this work
|
||||
#vsz = Variable("sz", 1, REPLAY_BUFFER_SIZE-1).bind(len(Xn))
|
||||
#X, A, R = Tensor(Xn).reshape(vsz, None), Tensor(An).reshape(vsz), Tensor(Rn).reshape(vsz)
|
||||
|
||||
old_log_dist = model(X)[0].detach() # TODO: could save these instead of recomputing
|
||||
for i in range(TRAIN_STEPS):
|
||||
samples = Tensor.randint(BATCH_SIZE, high=X.shape[0]).realize() # TODO: remove the need for this
|
||||
# TODO: is this recompiling based on the shape?
|
||||
action_loss, entropy_loss, critic_loss = train_step(X[samples], A[samples], R[samples], old_log_dist[samples])
|
||||
t.set_description(f"sz: {len(Xn):5d} steps/s: {steps/(time.perf_counter()-st):7.2f} action_loss: {action_loss.item():7.3f} entropy_loss: {entropy_loss.item():7.3f} critic_loss: {critic_loss.item():8.3f} reward: {sum(rews):6.2f}")
|
||||
|
||||
test_rew = evaluate(model, gym.make(ENVIRONMENT_NAME, render_mode='human'))
|
||||
print(f"test reward: {test_rew}")
|
||||
@@ -1,168 +0,0 @@
|
||||
import time
|
||||
start_tm = time.perf_counter()
|
||||
import math
|
||||
from typing import Tuple, cast
|
||||
import numpy as np
|
||||
from tinygrad import Tensor, nn, GlobalCounters, TinyJit, dtypes
|
||||
from tinygrad.helpers import partition, trange, getenv, Context
|
||||
from extra.lr_scheduler import OneCycleLR
|
||||
|
||||
dtypes.default_float = dtypes.half
|
||||
|
||||
# from https://github.com/tysam-code/hlb-CIFAR10/blob/main/main.py
|
||||
batchsize = getenv("BS", 1024)
|
||||
bias_scaler = 64
|
||||
hyp = {
|
||||
'opt': {
|
||||
'bias_lr': 1.525 * bias_scaler/512, # TODO: Is there maybe a better way to express the bias and batchnorm scaling? :'))))
|
||||
'non_bias_lr': 1.525 / 512,
|
||||
'bias_decay': 6.687e-4 * batchsize/bias_scaler,
|
||||
'non_bias_decay': 6.687e-4 * batchsize,
|
||||
'scaling_factor': 1./9,
|
||||
'percent_start': .23,
|
||||
'loss_scale_scaler': 1./32, # * Regularizer inside the loss summing (range: ~1/512 - 16+). FP8 should help with this somewhat too, whenever it comes out. :)
|
||||
},
|
||||
'net': {
|
||||
'whitening': {
|
||||
'kernel_size': 2,
|
||||
'num_examples': 50000,
|
||||
},
|
||||
'batch_norm_momentum': .4, # * Don't forget momentum is 1 - momentum here (due to a quirk in the original paper... >:( )
|
||||
'cutmix_size': 3,
|
||||
'cutmix_epochs': 6,
|
||||
'pad_amount': 2,
|
||||
'base_depth': 64 ## This should be a factor of 8 in some way to stay tensor core friendly
|
||||
},
|
||||
'misc': {
|
||||
'ema': {
|
||||
'epochs': 10, # Slight bug in that this counts only full epochs and then additionally runs the EMA for any fractional epochs at the end too
|
||||
'decay_base': .95,
|
||||
'decay_pow': 3.,
|
||||
'every_n_steps': 5,
|
||||
},
|
||||
'train_epochs': 12,
|
||||
#'train_epochs': 12.1,
|
||||
'device': 'cuda',
|
||||
'data_location': 'data.pt',
|
||||
}
|
||||
}
|
||||
|
||||
scaler = 2. ## You can play with this on your own if you want, for the first beta I wanted to keep things simple (for now) and leave it out of the hyperparams dict
|
||||
depths = {
|
||||
'init': round(scaler**-1*hyp['net']['base_depth']), # 32 w/ scaler at base value
|
||||
'block1': round(scaler** 0*hyp['net']['base_depth']), # 64 w/ scaler at base value
|
||||
'block2': round(scaler** 2*hyp['net']['base_depth']), # 256 w/ scaler at base value
|
||||
'block3': round(scaler** 3*hyp['net']['base_depth']), # 512 w/ scaler at base value
|
||||
'num_classes': 10
|
||||
}
|
||||
whiten_conv_depth = 3*hyp['net']['whitening']['kernel_size']**2
|
||||
|
||||
class ConvGroup:
|
||||
def __init__(self, channels_in, channels_out):
|
||||
self.conv1 = nn.Conv2d(channels_in, channels_out, kernel_size=3, padding=1, bias=False)
|
||||
self.conv2 = nn.Conv2d(channels_out, channels_out, kernel_size=3, padding=1, bias=False)
|
||||
self.norm1 = nn.BatchNorm(channels_out, track_running_stats=False, eps=1e-12, momentum=hyp['net']['batch_norm_momentum'])
|
||||
self.norm2 = nn.BatchNorm(channels_out, track_running_stats=False, eps=1e-12, momentum=hyp['net']['batch_norm_momentum'])
|
||||
cast(Tensor, self.norm1.weight).requires_grad = False
|
||||
cast(Tensor, self.norm2.weight).requires_grad = False
|
||||
def __call__(self, x:Tensor) -> Tensor:
|
||||
x = self.norm1(self.conv1(x).max_pool2d().float()).cast(dtypes.default_float).quick_gelu()
|
||||
return self.norm2(self.conv2(x).float()).cast(dtypes.default_float).quick_gelu()
|
||||
|
||||
class SpeedyConvNet:
|
||||
def __init__(self):
|
||||
self.whiten = nn.Conv2d(3, 2*whiten_conv_depth, kernel_size=hyp['net']['whitening']['kernel_size'], padding=0, bias=False)
|
||||
self.conv_group_1 = ConvGroup(2*whiten_conv_depth, depths['block1'])
|
||||
self.conv_group_2 = ConvGroup(depths['block1'], depths['block2'])
|
||||
self.conv_group_3 = ConvGroup(depths['block2'], depths['block3'])
|
||||
self.linear = nn.Linear(depths['block3'], depths['num_classes'], bias=False)
|
||||
def __call__(self, x:Tensor) -> Tensor:
|
||||
x = self.whiten(x).quick_gelu()
|
||||
x = x.sequential([self.conv_group_1, self.conv_group_2, self.conv_group_3])
|
||||
return self.linear(x.max(axis=(2,3))) * hyp['opt']['scaling_factor']
|
||||
|
||||
if __name__ == "__main__":
|
||||
# *** dataset ***
|
||||
X_train, Y_train, X_test, Y_test = nn.datasets.cifar()
|
||||
# TODO: without this line indexing doesn't fuse!
|
||||
X_train, Y_train, X_test, Y_test = [x.contiguous() for x in [X_train, Y_train, X_test, Y_test]]
|
||||
cifar10_std, cifar10_mean = X_train.float().std_mean(axis=(0, 2, 3))
|
||||
def preprocess(X:Tensor, Y:Tensor) -> Tuple[Tensor, Tensor]:
|
||||
return ((X - cifar10_mean.view(1, -1, 1, 1)) / cifar10_std.view(1, -1, 1, 1)).cast(dtypes.default_float), Y.one_hot(depths['num_classes'])
|
||||
|
||||
# *** model ***
|
||||
model = SpeedyConvNet()
|
||||
state_dict = nn.state.get_state_dict(model)
|
||||
|
||||
#for k,v in nn.state.torch_load("/tmp/cifar_net.pt").items(): print(k)
|
||||
|
||||
params_bias, params_non_bias = partition(state_dict.items(), lambda x: 'bias' in x[0])
|
||||
opt_bias = nn.optim.SGD([x[1] for x in params_bias], lr=0.01, momentum=.85, nesterov=True, weight_decay=hyp['opt']['bias_decay'])
|
||||
opt_non_bias = nn.optim.SGD([x[1] for x in params_non_bias], lr=0.01, momentum=.85, nesterov=True, weight_decay=hyp['opt']['non_bias_decay'])
|
||||
opt = nn.optim.OptimizerGroup(opt_bias, opt_non_bias)
|
||||
|
||||
num_steps_per_epoch = X_train.size(0) // batchsize
|
||||
total_train_steps = math.ceil(num_steps_per_epoch * hyp['misc']['train_epochs'])
|
||||
loss_batchsize_scaler = 512/batchsize
|
||||
|
||||
pct_start = hyp['opt']['percent_start']
|
||||
initial_div_factor = 1e16 # basically to make the initial lr ~0 or so :D
|
||||
final_lr_ratio = .07 # Actually pretty important, apparently!
|
||||
lr_sched_bias = OneCycleLR(opt_bias, max_lr=hyp['opt']['bias_lr'], pct_start=pct_start, div_factor=initial_div_factor, final_div_factor=1./(initial_div_factor*final_lr_ratio), total_steps=total_train_steps)
|
||||
lr_sched_non_bias = OneCycleLR(opt_non_bias, max_lr=hyp['opt']['non_bias_lr'], pct_start=pct_start, div_factor=initial_div_factor, final_div_factor=1./(initial_div_factor*final_lr_ratio), total_steps=total_train_steps)
|
||||
|
||||
def loss_fn(out, Y):
|
||||
return out.cross_entropy(Y, reduction='none', label_smoothing=0.2).mul(hyp['opt']['loss_scale_scaler']*loss_batchsize_scaler).sum().div(hyp['opt']['loss_scale_scaler'])
|
||||
|
||||
@TinyJit
|
||||
@Tensor.train()
|
||||
def train_step(idxs:Tensor) -> Tensor:
|
||||
with Context(SPLIT_REDUCEOP=0, FUSE_ARANGE=1):
|
||||
X = X_train[idxs]
|
||||
Y = Y_train[idxs].realize(X)
|
||||
X, Y = preprocess(X, Y)
|
||||
out = model(X)
|
||||
loss = loss_fn(out, Y)
|
||||
opt.zero_grad()
|
||||
loss.backward()
|
||||
opt.step()
|
||||
lr_sched_bias.step()
|
||||
lr_sched_non_bias.step()
|
||||
return loss / (batchsize*loss_batchsize_scaler)
|
||||
|
||||
eval_batchsize = 2500
|
||||
@TinyJit
|
||||
@Tensor.test()
|
||||
def val_step() -> Tuple[Tensor, Tensor]:
|
||||
# TODO with Tensor.no_grad()
|
||||
Tensor.no_grad = True
|
||||
loss, acc = [], []
|
||||
for i in range(0, X_test.size(0), eval_batchsize):
|
||||
X, Y = preprocess(X_test[i:i+eval_batchsize], Y_test[i:i+eval_batchsize])
|
||||
out = model(X)
|
||||
loss.append(loss_fn(out, Y))
|
||||
acc.append((out.argmax(-1).one_hot(depths['num_classes']) * Y).sum() / eval_batchsize)
|
||||
ret = Tensor.stack(*loss).mean() / (batchsize*loss_batchsize_scaler), Tensor.stack(*acc).mean()
|
||||
Tensor.no_grad = False
|
||||
return ret
|
||||
|
||||
np.random.seed(1337)
|
||||
for epoch in range(math.ceil(hyp['misc']['train_epochs'])):
|
||||
# TODO: move to tinygrad
|
||||
gst = time.perf_counter()
|
||||
idxs = np.arange(X_train.shape[0])
|
||||
np.random.shuffle(idxs)
|
||||
tidxs = Tensor(idxs, dtype='int')[:num_steps_per_epoch*batchsize].reshape(num_steps_per_epoch, batchsize) # NOTE: long doesn't fold
|
||||
train_loss:float = 0
|
||||
for epoch_step in (t:=trange(num_steps_per_epoch)):
|
||||
st = time.perf_counter()
|
||||
GlobalCounters.reset()
|
||||
loss = train_step(tidxs[epoch_step].contiguous()).float().item()
|
||||
t.set_description(f"*** loss: {loss:5.3f} lr: {opt_non_bias.lr.item():.6f}"
|
||||
f" tm: {(et:=(time.perf_counter()-st))*1000:6.2f} ms {GlobalCounters.global_ops/(1e9*et):7.0f} GFLOPS")
|
||||
train_loss += loss
|
||||
gmt = time.perf_counter()
|
||||
GlobalCounters.reset()
|
||||
val_loss, acc = [x.float().item() for x in val_step()]
|
||||
get = time.perf_counter()
|
||||
print(f"\033[F*** epoch {epoch:3d} tm: {(gmt-gst):5.2f} s val_tm: {(get-gmt):5.2f} s train_loss: {train_loss/num_steps_per_epoch:5.3f} val_loss: {val_loss:5.3f} eval acc: {acc*100:5.2f}% @ {get-start_tm:6.2f} s ")
|
||||
@@ -1,50 +0,0 @@
|
||||
# model based off https://towardsdatascience.com/going-beyond-99-mnist-handwritten-digits-recognition-cfff96337392
|
||||
from typing import List, Callable
|
||||
from tinygrad import Tensor, TinyJit, nn, GlobalCounters
|
||||
from tinygrad.helpers import getenv, colored, trange
|
||||
from tinygrad.nn.datasets import mnist
|
||||
|
||||
class Model:
|
||||
def __init__(self):
|
||||
self.layers: List[Callable[[Tensor], Tensor]] = [
|
||||
nn.Conv2d(1, 32, 5), Tensor.relu,
|
||||
nn.Conv2d(32, 32, 5), Tensor.relu,
|
||||
nn.BatchNorm(32), Tensor.max_pool2d,
|
||||
nn.Conv2d(32, 64, 3), Tensor.relu,
|
||||
nn.Conv2d(64, 64, 3), Tensor.relu,
|
||||
nn.BatchNorm(64), Tensor.max_pool2d,
|
||||
lambda x: x.flatten(1), nn.Linear(576, 10)]
|
||||
|
||||
def __call__(self, x:Tensor) -> Tensor: return x.sequential(self.layers)
|
||||
|
||||
if __name__ == "__main__":
|
||||
X_train, Y_train, X_test, Y_test = mnist(fashion=getenv("FASHION"))
|
||||
|
||||
model = Model()
|
||||
opt = nn.optim.Adam(nn.state.get_parameters(model))
|
||||
|
||||
@TinyJit
|
||||
@Tensor.train()
|
||||
def train_step() -> Tensor:
|
||||
opt.zero_grad()
|
||||
samples = Tensor.randint(getenv("BS", 512), high=X_train.shape[0])
|
||||
# TODO: this "gather" of samples is very slow. will be under 5s when this is fixed
|
||||
loss = model(X_train[samples]).sparse_categorical_crossentropy(Y_train[samples]).backward()
|
||||
opt.step()
|
||||
return loss
|
||||
|
||||
@TinyJit
|
||||
@Tensor.test()
|
||||
def get_test_acc() -> Tensor: return (model(X_test).argmax(axis=1) == Y_test).mean()*100
|
||||
|
||||
test_acc = float('nan')
|
||||
for i in (t:=trange(getenv("STEPS", 70))):
|
||||
GlobalCounters.reset() # NOTE: this makes it nice for DEBUG=2 timing
|
||||
loss = train_step()
|
||||
if i%10 == 9: test_acc = get_test_acc().item()
|
||||
t.set_description(f"loss: {loss.item():6.2f} test_accuracy: {test_acc:5.2f}%")
|
||||
|
||||
# verify eval acc
|
||||
if target := getenv("TARGET_EVAL_ACC_PCT", 0.0):
|
||||
if test_acc >= target and test_acc != 100.0: print(colored(f"{test_acc=} >= {target}", "green"))
|
||||
else: raise ValueError(colored(f"{test_acc=} < {target}", "red"))
|
||||
@@ -1,56 +0,0 @@
|
||||
# model based off https://towardsdatascience.com/going-beyond-99-mnist-handwritten-digits-recognition-cfff96337392
|
||||
from typing import List, Callable
|
||||
from tinygrad import Tensor, TinyJit, nn, GlobalCounters, Device
|
||||
from tinygrad.helpers import getenv, colored, trange
|
||||
from tinygrad.nn.datasets import mnist
|
||||
|
||||
GPUS = tuple(f'{Device.DEFAULT}:{i}' for i in range(getenv("GPUS", 2)))
|
||||
|
||||
class Model:
|
||||
def __init__(self):
|
||||
self.layers: List[Callable[[Tensor], Tensor]] = [
|
||||
nn.Conv2d(1, 32, 5), Tensor.relu,
|
||||
nn.Conv2d(32, 32, 5), Tensor.relu,
|
||||
nn.BatchNorm2d(32), Tensor.max_pool2d,
|
||||
nn.Conv2d(32, 64, 3), Tensor.relu,
|
||||
nn.Conv2d(64, 64, 3), Tensor.relu,
|
||||
nn.BatchNorm2d(64), Tensor.max_pool2d,
|
||||
lambda x: x.flatten(1), nn.Linear(576, 10)]
|
||||
|
||||
def __call__(self, x:Tensor) -> Tensor: return x.sequential(self.layers)
|
||||
|
||||
if __name__ == "__main__":
|
||||
X_train, Y_train, X_test, Y_test = mnist()
|
||||
# we shard the test data on axis 0
|
||||
X_test.shard_(GPUS, axis=0)
|
||||
Y_test.shard_(GPUS, axis=0)
|
||||
|
||||
model = Model()
|
||||
for k, x in nn.state.get_state_dict(model).items(): x.to_(GPUS) # we put a copy of the model on every GPU
|
||||
opt = nn.optim.Adam(nn.state.get_parameters(model))
|
||||
|
||||
@TinyJit
|
||||
def train_step() -> Tensor:
|
||||
with Tensor.train():
|
||||
opt.zero_grad()
|
||||
samples = Tensor.randint(getenv("BS", 512), high=X_train.shape[0])
|
||||
Xt, Yt = X_train[samples].shard_(GPUS, axis=0), Y_train[samples].shard_(GPUS, axis=0) # we shard the data on axis 0
|
||||
# TODO: this "gather" of samples is very slow. will be under 5s when this is fixed
|
||||
loss = model(Xt).sparse_categorical_crossentropy(Yt).backward()
|
||||
opt.step()
|
||||
return loss
|
||||
|
||||
@TinyJit
|
||||
def get_test_acc() -> Tensor: return (model(X_test).argmax(axis=1) == Y_test).mean()*100
|
||||
|
||||
test_acc = float('nan')
|
||||
for i in (t:=trange(getenv("STEPS", 70))):
|
||||
GlobalCounters.reset() # NOTE: this makes it nice for DEBUG=2 timing
|
||||
loss = train_step()
|
||||
if i%10 == 9: test_acc = get_test_acc().item()
|
||||
t.set_description(f"loss: {loss.item():6.2f} test_accuracy: {test_acc:5.2f}%")
|
||||
|
||||
# verify eval acc
|
||||
if target := getenv("TARGET_EVAL_ACC_PCT", 0.0):
|
||||
if test_acc >= target: print(colored(f"{test_acc=} >= {target}", "green"))
|
||||
else: raise ValueError(colored(f"{test_acc=} < {target}", "red"))
|
||||
@@ -1,95 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
import os, sys, traceback
|
||||
sys.path.append(os.getcwd())
|
||||
|
||||
from io import StringIO
|
||||
from contextlib import redirect_stdout
|
||||
from tinygrad import Tensor, nn, Device, dtypes
|
||||
from tinygrad.helpers import Timing, colored, getenv, fetch
|
||||
from extra.models.llama import Transformer, convert_from_huggingface, fix_bf16
|
||||
from sentencepiece import SentencePieceProcessor
|
||||
|
||||
def create_fixed_tokenizer(output_file):
|
||||
print("creating fixed tokenizer")
|
||||
import extra.junk.sentencepiece_model_pb2 as spb2
|
||||
mp = spb2.ModelProto()
|
||||
mp.ParseFromString(fetch("https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B/resolve/main/tokenizer.model?download=true").read_bytes())
|
||||
mp.pieces.append(spb2.ModelProto.SentencePiece(piece="<|im_end|>", score=0))
|
||||
mp.pieces.append(spb2.ModelProto.SentencePiece(piece="<|im_start|>", score=0))
|
||||
with open(output_file, "wb") as f:
|
||||
f.write(mp.SerializeToString())
|
||||
|
||||
# example:
|
||||
# echo -en "write 2+2\nwrite hello world\ny\n" | TEMP=0 python3 examples/coder.py
|
||||
|
||||
if __name__ == "__main__":
|
||||
Tensor.no_grad = True
|
||||
|
||||
# https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B/blob/main/config.json
|
||||
with Timing("create model: "):
|
||||
model = Transformer(4096, 14336, n_heads=32, n_layers=32, norm_eps=1e-5, vocab_size=32002, n_kv_heads=8, max_context=4096, jit=getenv("JIT", 1))
|
||||
|
||||
with Timing("download weights: "):
|
||||
part1 = nn.state.torch_load(fetch("https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B/resolve/main/pytorch_model-00001-of-00002.bin?download=true"))
|
||||
part2 = nn.state.torch_load(fetch("https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B/resolve/main/pytorch_model-00002-of-00002.bin?download=true"))
|
||||
|
||||
with Timing("weights -> model: "):
|
||||
nn.state.load_state_dict(model, fix_bf16(convert_from_huggingface(part1, model, 32, 8)), strict=False)
|
||||
nn.state.load_state_dict(model, fix_bf16(convert_from_huggingface(part2, model, 32, 8)), strict=False)
|
||||
|
||||
if not os.path.isfile("/tmp/tokenizer.model"): create_fixed_tokenizer("/tmp/tokenizer.model")
|
||||
spp = SentencePieceProcessor(model_file="/tmp/tokenizer.model")
|
||||
|
||||
# https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B/blob/main/tokenizer_config.json
|
||||
# "chat_template": "{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
|
||||
IM_END = 32000
|
||||
IM_START = 32001
|
||||
def encode_prompt(k, v): return [IM_START]+spp.encode(f"{k}\n{v}")+[IM_END]+spp.encode("\n")
|
||||
def start_prompt(k): return [IM_START]+spp.encode(f"{k}\n")
|
||||
def output(outputted, toks, color):
|
||||
cur = spp.decode(toks)[len(outputted):]
|
||||
sys.stdout.write(colored(cur, color))
|
||||
sys.stdout.flush()
|
||||
outputted += cur
|
||||
return outputted
|
||||
|
||||
# *** app below this line ***
|
||||
|
||||
toks = [spp.bos_id()] + encode_prompt("system", "You are Quentin. Quentin is a useful assistant who writes Python code to answer questions. He keeps the code as short as possible and doesn't read from user input")
|
||||
|
||||
PROMPT = getenv("PROMPT", 1)
|
||||
temperature = getenv("TEMP", 0.7)
|
||||
|
||||
start_pos = 0
|
||||
outputted = output("", toks, "green")
|
||||
turn = True
|
||||
while 1:
|
||||
if PROMPT:
|
||||
toks += encode_prompt("user", input("Q: ")) + start_prompt("assistant")
|
||||
else:
|
||||
toks += start_prompt("user" if turn else "assistant")
|
||||
turn = not turn
|
||||
old_output_len = len(outputted)
|
||||
while 1:
|
||||
tok = model(Tensor([toks[start_pos:]]), start_pos, temperature).item()
|
||||
start_pos = len(toks)
|
||||
toks.append(tok)
|
||||
outputted = output(outputted, toks, "blue" if not turn else "cyan")
|
||||
if tok == IM_END: break
|
||||
if tok == spp.eos_id(): break
|
||||
new_output = outputted[old_output_len:]
|
||||
|
||||
if new_output.endswith("```") and '```python\n' in new_output:
|
||||
python_code = new_output.split('```python\n')[1].split("```")[0]
|
||||
# AI safety. Warning to user. Do not press y if the AI is trying to do unsafe things.
|
||||
if input(colored(f" <-- PYTHON DETECTED, RUN IT? ", "red")).lower() == 'y':
|
||||
my_stdout = StringIO()
|
||||
try:
|
||||
with redirect_stdout(my_stdout): exec(python_code)
|
||||
result = my_stdout.getvalue()
|
||||
except Exception as e:
|
||||
result = ''.join(traceback.format_exception_only(e))
|
||||
toks += spp.encode(f"\nOutput:\n```\n{result}```")
|
||||
outputted = output(outputted, toks, "yellow")
|
||||
old_output_len = len(outputted)
|
||||
print("")
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user