pytorch/benchmark

Pull Requests Commits

Add flash attention v3 (#2381)

xuzhao9 committed 1 year ago

6dfede78

enforce style for empty lines in import segments in `benchmarks/` (#129754)

XuehaiPan committed 1 year ago

429137dd

typing: convert_frame (#130670)

aorenste committed 1 year ago

97770e96

Run performance test non-alternately (#130136)

xuzhao9 committed 1 year ago

736bd91d

Mark nn_module params and buffers as static in dynamo (#130391)

mlazos committed 1 year ago

fbfbe989

dynamo add support for `defaultdict(set)` (#130745)

alexcdennis committed 1 year ago

40ee53b1

Add TMA+persistent bf16 gemm (#2373)

bertmaher committed 1 year ago

0ec9f60e

Add persistent+TMA matmul to fp8 gemm benchmark (#2377)

bertmaher committed 1 year ago

4f245a87

Fix servicelab test failures

xuzhao9 committed 1 year ago

8a3bf01a

Add torch.sum benchmark for jagged_layernorm operator (#2376)

jananisriram committed 1 year ago

e26cd75d

Update FBGEMM commit (#2375)

xuzhao9 committed 1 year ago

c70ccb37

Fix torch_onnx patch accuracy bug in benchmark (#130586)

titaiwangms committed 1 year ago

8359e1aa

add total wall time in calculate_time_spent (#130611)

dshi7 committed 1 year ago

35489f10

apply autofix for ruff rules

XuehaiPan committed 1 year ago

59596d44

Enhancements to recompiles logs (#130043)

ezyang committed 1 year ago

b74c016d

Add ncu_rep_ir metric for TTGIR source attribution

int3 committed 1 year ago

643fa347

tweak shapes (#2372)

bertmaher committed 1 year ago

475e123a

Simplify and improve scope of UserDefinedObject var_getattr (#130169)

anijain2305 committed 1 year ago

9783803c

Add help for run_benchmark (#2361)

xuzhao9 committed 1 year ago

bb529409

Automatically collect best_config

int3 committed 1 year ago

555d05a3

Add RoBERTa to the model suite (#2279)

Rohanjames1997 committed 1 year ago

afdc319b

Add usage logger to track user usage

xuzhao9 committed 1 year ago

ff1b8eb0

Fallback to delete-package-versions v4 to use node16 (#2367)

xuzhao9 committed 1 year ago

b35d2970

raise tolerance for squeezenet1_1 (#130165)

shunting314 committed 1 year ago

36f880d9

use eval mode for vision_maskrcnn (#130163)

shunting314 committed 1 year ago

4b434db1

fix accuracy failure for beit_base_patch16_224 during training (#130005)

shunting314 committed 1 year ago

6712815f

fix accuracy failure for a few models (#129941)

shunting314 committed 1 year ago

4b2105b9

pass acc test if ref is NaN (#129996)

shunting314 committed 1 year ago

06dd5b8c

Skip side effect tracking for c wrappers/descriptors (#129914)

anijain2305 committed 1 year ago

8649e6cd

Add `artifacts_dir` to torch-onnx-patch in benchmark (#130069)

titaiwangms committed 1 year ago

7cbf4872

Newer Older