pytorch/benchmark

Pull Requests Commits

stable_diffusion_unet allow customize bs

xmfan committed 2 years ago

9010216d

Move ShapeEnv config out of dynamo (#112933)

peterbell10 committed 2 years ago

101fae3f

Rename torch.onnx.ExportOutput* to ONNXProgram* (#112263)

Thiago Crepaldi committed 2 years ago

f5b502a5

detectron2: require coco2017-minimal data (#2022)

davidberard98 committed 2 years ago

1e03c239

Add custom treespec fqn field

angelayi committed 2 years ago

94f54e87

Log non-pt2_compliant ops encountered by Dynamo

zou3519 committed 2 years ago

bd6a7668

metric table (#109245)

shunting314 committed 2 years ago

fec28961

remove #ops comparison to fx.symbolic_trace from dynamo standard_test (#112420)

williamwen42 committed 2 years ago

b7d90f73

Update how Dynamo decides to graph break on an OpOverloadPacket (#112200)

zou3519 committed 2 years ago

cdbfa9c4

Use `pytree.tree_leaves` everywhere (#112324)

peterbell10 committed 2 years ago

9221c797

Remove VariableTracker.as_specialized (#112363)

jansel committed 2 years ago

f13ea812

Repair `num_batch` field access and cleanup code (#2020)

gs-olive committed 2 years ago

2d7c7708

Make torch context manager a TorchCtxManagerClassVariable (#111622)

yanboliang committed 2 years ago

619f1594

Enable typechecking for testing.py (#112129)

int3 committed 2 years ago

94126be6

Move Set to dicts.py (#110522)

lezcano committed 2 years ago

97e7f0d0

Support calling __torch_function__ attribute access (#111737)

mlazos committed 2 years ago

808d8be4

Enable bf16 on all models

xuzhao9 committed 2 years ago

adc4b0c8

Default to supported model batch size instead of metadata if batch size not specified (#2015)

eellison committed 2 years ago

f2351447

yolov3: reduce batch size due to OOM (#111959)

xmfan committed 2 years ago

c533c93a

Add autotune_max_gemm and dump_triton options

xuzhao9 committed 2 years ago

edf71156

Dynamo runner: add FSDP handcrafted module wrapping policy (#111505)

xmfan committed 2 years ago

f2534640

Ensure Dynamo uses this graph's fakes for `Tensor` `example_value`s (#111954)

jon-chuang committed 2 years ago

42a3c07a

Pass TorchIR to AOTInductor

angelayi committed 2 years ago

9cf38eca

fix regression which creates a new fake tensor (#111864)

jon-chuang committed 2 years ago

65c0b7de

Delete deepcopied model after use in benchmark to reduce memory consumption (#111868)

BowenBao committed 2 years ago

085167e5

Apply same 'pick_grad' on generating fp64 reference outputs (#111593)

BowenBao committed 2 years ago

8a4b1540

Enable onnx inlining in benchmark for >2GB models (#111867)

BowenBao committed 2 years ago

92932909

Add error logs to GHA (#2007)

gs-olive committed 2 years ago

fd88c7ac

Update Torch-TRT latency report (#2006)

gs-olive committed 2 years ago

e6e8a603

disable issue creation temporarily (#2000)

janeyx99 committed 2 years ago

3b7eac98

Older