github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-02-17 10:02:00 -05:00
Code Issues Packages Projects Releases Wiki Activity

fix rand_like for multi (#8880)

Browse Source
This commit is contained in:
chenyu
2025-02-03 19:00:14 -05:00
committed by GitHub
parent 746d899dbd
commit 836cf42c2e
3 changed files with 18 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
.github/workflows/benchmark.yml vendored
View File

@@ -300,9 +300,8 @@ jobs:
- name: Run 10 MLPerf ResNet50 training steps (6 gpu)
run: NV=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=1536 GPUS=6 MODEL=resnet python3 examples/mlperf/model_train.py | tee train_resnet.txt
- name: Run 10 MLPerf Bert training steps (6 gpu)
# TODO: remove DISABLE_DROPOUT once dropout is fixed
# TODO: remove BERT_LAYERS once scheduler is fast
run: NV=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=66 GPUS=6 DISABLE_DROPOUT=1 BERT_LAYERS=2 MODEL=bert python3 examples/mlperf/model_train.py | tee train_bert.txt
run: NV=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=66 GPUS=6 BERT_LAYERS=2 MODEL=bert python3 examples/mlperf/model_train.py | tee train_bert.txt
- uses: actions/upload-artifact@v4
with:
name: Speed (NVIDIA Training)
@@ -498,9 +497,8 @@ jobs:
- name: Run 10 MLPerf ResNet50 training steps (6 gpu)
run: AMD=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=1536 GPUS=6 MODEL=resnet python3 examples/mlperf/model_train.py | tee train_resnet.txt
- name: Run 10 MLPerf Bert training steps (6 gpu)
# TODO: remove DISABLE_DROPOUT once dropout is fixed
# TODO: remove BERT_LAYERS once scheduler is fast
run: AMD=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=66 GPUS=6 DISABLE_DROPOUT=1 BERT_LAYERS=2 MODEL=bert python3 examples/mlperf/model_train.py | tee train_bert.txt
run: AMD=1 CAPTURE_PROCESS_REPLAY=0 DEFAULT_FLOAT=HALF BENCHMARK=10 BS=66 GPUS=6 BERT_LAYERS=2 MODEL=bert python3 examples/mlperf/model_train.py | tee train_bert.txt
- uses: actions/upload-artifact@v4
with:
name: Speed (AMD Training)