set SPLIT_W13=0 for 8b DP by default (#16302)

This commit is contained in:
qazal 2026-05-21 16:09:10 +03:00 committed by GitHub
commit e33e058d34
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 2 additions and 0 deletions

View file

@ -24,6 +24,7 @@ export FUSED_GRAD_QUANTIZE=${FUSED_GRAD_QUANTIZE:-1}
export FUSED_ADD_NORM_MUL_QUANTIZE=${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 SPLIT_W13=${SPLIT_W13:-0}
export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
export DP=${DP:-8} MP=${MP:-1} BS=${BS:-16} EVAL_BS=${EVAL_BS:-8} GRADIENT_ACC_STEPS=${GRADIENT_ACC_STEPS:-2}

View file

@ -24,6 +24,7 @@ export FUSED_GRAD_QUANTIZE=${FUSED_GRAD_QUANTIZE:-1}
export FUSED_ADD_NORM_MUL_QUANTIZE=${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 SPLIT_W13=${SPLIT_W13:-0}
export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
export DP=${DP:-8} MP=${MP:-1} BS=${BS:-16} EVAL_BS=${EVAL_BS:-8} GRADIENT_ACC_STEPS=${GRADIENT_ACC_STEPS:-2}