tinygrad/tinygrad
2
0
Fork 3
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-09-03 00:45:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

CACHELEVEL=0 -> IGNORE_BEAM_CACHE=1 in benchmark (#5191)

Browse Source 
ignoring beam cache but using compile cache should be fine, saved some benchmark time.

also updated `beam_search` to check flag value before accessing diskcache
This commit is contained in:
chenyu
2024-06-27 13:15:18 -04:00
committed by GitHub
parent c12de4f47d
commit 0c6c7c5f7b
2 changed files with 10 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/benchmark.yml
+9 -9
View File
@@ -56,7 +56,7 @@ jobs:
JIT=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_unjitted.txt
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: JIT=1 JITBEAM=2 CACHELEVEL=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_beam.txt
run: JIT=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 quantized LLaMA
run: |
JIT=1 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing --quantize int8 | tee llama_int8.txt
@@ -70,7 +70,7 @@ jobs:
- name: Run GPT2 w HALF
run: JIT=1 HALF=1 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half.txt
- name: Run GPT2 w HALF/BEAM
run: JIT=1 HALF=1 JITBEAM=2 CACHELEVEL=0 CAST_BEFORE_VIEW=0 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
run: JIT=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
- name: Train MNIST
run: time PYTHONPATH=. TARGET_EVAL_ACC_PCT=97.3 python3 examples/beautiful_mnist.py | tee beautiful_mnist.txt
- name: Run 10 CIFAR training steps
@@ -145,7 +145,7 @@ jobs:
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 (CUDA) with BEAM
run: BEAM=4 CUDA=1 HALF=1 CACHELEVEL=0 DEBUG=2 python3 extra/gemm/simple_matmul.py
run: BEAM=4 CUDA=1 HALF=1 IGNORE_BEAM_CACHE=1 DEBUG=2 python3 extra/gemm/simple_matmul.py
- name: Run Tensor Core GEMM (PTX)
run: CUDA=1 PTX=1 HALF=1 DEBUG=2 python3 extra/gemm/simple_matmul.py | tee matmul_ptx.txt
- name: Run Tensor Core GEMM (NV)
@@ -161,13 +161,13 @@ jobs:
CUDA=1 JIT=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_unjitted.txt
CUDA=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: CUDA=1 JIT=1 JITBEAM=2 CACHELEVEL=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_beam.txt
run: CUDA=1 JIT=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: CUDA=1 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: CUDA=1 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 CACHELEVEL=0 python3 examples/llama3.py --model weights/LLaMA-3/8B-SF-DPO/ --benchmark | tee llama3_beam.txt
run: NV=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 python3 examples/llama3.py --model weights/LLaMA-3/8B-SF-DPO/ --benchmark | tee llama3_beam.txt
- name: Run LLaMA-3 8B on 4 GPUs
run: NV=1 python3 examples/llama3.py --shard 4 --model weights/LLaMA-3/8B-SF-DPO/ --benchmark | tee llama3_four_gpu.txt
- name: Run LLaMA-3 8B on 6 GPUs
@@ -183,7 +183,7 @@ jobs:
- name: Run GPT2 w HALF
run: CUDA=1 JIT=1 HALF=1 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half.txt
- name: Run GPT2 w HALF/BEAM
run: CUDA=1 JIT=1 HALF=1 JITBEAM=2 CACHELEVEL=0 CAST_BEFORE_VIEW=0 JIT_BATCH_SIZE=4 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
run: CUDA=1 JIT=1 HALF=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 CAST_BEFORE_VIEW=0 JIT_BATCH_SIZE=4 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
- uses: actions/upload-artifact@v4
with:
name: Speed (NVIDIA)
@@ -314,13 +314,13 @@ jobs:
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 JIT=1 JITBEAM=2 CACHELEVEL=0 python3 examples/llama.py --gen 1 --prompt "Hello." --count 10 --temperature 0 --timing | tee llama_beam.txt
run: AMD=1 JIT=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 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 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 CACHELEVEL=0 python3 examples/llama3.py --model weights/LLaMA-3/8B-SF-DPO/ --benchmark | tee llama3_beam.txt
run: AMD=1 JITBEAM=2 IGNORE_BEAM_CACHE=1 python3 examples/llama3.py --model weights/LLaMA-3/8B-SF-DPO/ --benchmark | tee llama3_beam.txt
- name: Run LLaMA-3 8B on 4 GPUs
run: AMD=1 python3 examples/llama3.py --shard 4 --model weights/LLaMA-3/8B-SF-DPO/ --benchmark | tee llama3_four_gpu.txt
- name: Run LLaMA-3 8B on 6 GPUs
@@ -336,7 +336,7 @@ jobs:
- name: Run GPT2 w HALF
run: AMD=1 JIT=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 JIT=1 HALF=1 JITBEAM=2 CACHELEVEL=0 CAST_BEFORE_VIEW=0 python3 examples/gpt2.py --count 10 --temperature 0 --timing | tee gpt2_half_beam.txt
run: AMD=1 JIT=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)
tinygrad/engine/search.py
+1 -1
View File
@@ -118,7 +118,7 @@ beam_pool, BEAM_DEBUG = None, getenv("BEAM_DEBUG")
def beam_search(lin:Linearizer, rawbufs:List[Buffer], amt:int, allow_test_size=True) -> Linearizer:
global beam_pool
key = {"ast": lin.ast[0].key, "amt": amt, "allow_test_size": allow_test_size, "device": lin.opts.device, "suffix": lin.opts.suffix}
if (val:=diskcache_get("beam_search", key)) is not None and not getenv("IGNORE_BEAM_CACHE") and CACHELEVEL >= 1:
if not getenv("IGNORE_BEAM_CACHE") and CACHELEVEL >= 1 and (val:=diskcache_get("beam_search", key)) is not None:
ret = lin.copy()
for o in val[len(lin.applied_opts):]: ret.apply_opt(o)
return ret