sirhcm/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Activity

llama: use env vars (#15978)

Browse Source
This commit is contained in:
wozeparrot
2026-04-29 12:37:15 -07:00
committed by GitHub
parent a37b605523
commit 0080489abe
3 changed files with 12 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama8b/implementations/tinybox_8xMI350X/dev_beam.sh
+4
View File
@@ -16,6 +16,10 @@ export MASTER_WEIGHTS=${MASTER_WEIGHTS:-1}
export FP8=${FP8:-1}
export ALLREDUCE_CAST=${ALLREDUCE_CAST:-1}
export FAST_CE=${FASE_CE:-1}
export FUSED_INPUT_QUANTIZE=${FUSED_INPUT_QUANTIZE:-1}
export FUSED_ADD_NORM_MUL_QUANTIZE=${FUSED_ADD_NORM_MUL_QUANTIZE:-1}
export FUSED_SILU_W13=${FUSED_SILU_W13:-1}
export FUSED_PAD_GRAD_ACCUM=${FUSED_PAD_GRAD_ACCUM:-1}
export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
export DP=${DP:-8} MP=${MP:-1} BS=${BS:-16} EVAL_BS=${EVAL_BS:-8} GRADIENT_ACC_STEPS=${GRADIENT_ACC_STEPS:-2}
examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama8b/implementations/tinybox_8xMI350X/dev_run.sh
+4
View File
@@ -16,6 +16,10 @@ export MASTER_WEIGHTS=${MASTER_WEIGHTS:-1}
export FP8=${FP8:-1}
export ALLREDUCE_CAST=${ALLREDUCE_CAST:-1}
export FAST_CE=${FASE_CE:-1}
export FUSED_INPUT_QUANTIZE=${FUSED_INPUT_QUANTIZE:-1}
export FUSED_ADD_NORM_MUL_QUANTIZE=${FUSED_ADD_NORM_MUL_QUANTIZE:-1}
export FUSED_SILU_W13=${FUSED_SILU_W13:-1}
export FUSED_PAD_GRAD_ACCUM=${FUSED_PAD_GRAD_ACCUM:-1}
export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
export DP=${DP:-8} MP=${MP:-1} BS=${BS:-16} EVAL_BS=${EVAL_BS:-8} GRADIENT_ACC_STEPS=${GRADIENT_ACC_STEPS:-2}
examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama8b/implementations/tinybox_8xMI350X/run_and_time.sh
+4
View File
@@ -17,6 +17,10 @@ export MASTER_WEIGHTS=1
export FP8=1
export ALLREDUCE_CAST=1
export FAST_CE=1
export FUSED_INPUT_QUANTIZE=1
export FUSED_ADD_NORM_MUL_QUANTIZE=1
export FUSED_SILU_W13=1
export FUSED_PAD_GRAD_ACCUM=1
export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
export DP=8 MP=1 BS=16 EVAL_BS=8 GRADIENT_ACC_STEPS=2