strisunshinewentingwang/pytorch

Pull Requests Commits

Flash Attention v2 (#105602)

drisspg committed 1 year ago

add45aea

[PyTorch][Vulkan] Add a matrix multiplication performance test binary and fix GPU latency measurement (#108266)

liuk22 committed 1 year ago

234f00e1

add Half support for GroupNorm on CPU (#100234)

CaoE committed 1 year ago

8f028845

NT support for matmul of (B, *, C, D) NT with dense (D, E) (#108370)

jbschlosser committed 1 year ago

54dcb0ea

[DTensor][random] add DTensor constructor: randn (#108285)

XilunWu committed 1 year ago

a78b78cd

Put logging in run_tests (#107987)

clee2000 committed 1 year ago

c67ebae3

Fix `full` on symbolic value. (#108166)

ysiraichi committed 1 year ago

29f17e1f

[export] Properly handle duplicated params. (#108415)

zhxchen17 committed 1 year ago

fc1c862e

enabled AT_USE_JITERATOR() for `tan` and `tanh` kernels. (#102427)

parth-desai committed 1 year ago

2d9a8289

[ONNX] Show sarif_report_path (#108398)

justinchuby committed 1 year ago

6ba2b6e1

[inductor] Generalize pointless_cumsum_replacement pattern (#108373)

peterbell10 committed 1 year ago

e58d3ed8

[CI] Enable max-autotune for Sunday dashboard run (#108386)

desertfire committed 1 year ago

0f1a225f

[dynamo] Avoid recompilation when the PyTorch function accepts scalars (#108162)

lezcano committed 1 year ago

2a6ef9b0

[FSDP][state_dict][optim_state_dict] Log slow optim and model state_dict paths (#108290)

fegin committed 1 year ago

591cb776

[NCCL][CUDA][CUDA Graphs] Flush enqueued work before starting a graph capture (#104487)

eqy committed 1 year ago

db63bf3d

[PyPer][BE] Fix test_scripted_module in StatCollector (#108232)

PaulZhang12 committed 1 year ago

4a9c6f1b

[export] Fix duplicated params for AOTInductor. (#108354)

zhxchen17 committed 1 year ago

d96446b9

Update accuracy checking for nan, floats (#108202)

eellison committed 1 year ago

e18f512b

[DeviceMesh] Add unique mesh_dim_name check in init_device_mesh() (#108326)

wz337 committed 1 year ago

90ef3b82

dynamo: trace autograd.Function with tensor subclass input (#108093)

vkuzo committed 1 year ago

37029807

NT support for cat with dim=0 (#108361)

jbschlosser committed 1 year ago

414cb26d

[quant][pt2e] Move propagate_annotation from quant flow to quantizer (#108320)

jerryzh168 committed 1 year ago

a9fe0b5b

Revert "[HSDP] Add device_mesh to FSDP and add dtensor state_dict support for HSDP (#107533)"

pytorchmergebot committed 1 year ago

ab5b4c44

Support is_mtia attribute. (#108307) (#108310)

egienvalue committed 1 year ago

8289ad8e

Revert "Flash Attention v2 (#105602)"

pytorchmergebot committed 1 year ago

d569e506

Allow float dtype when Autocast CPU Disabled (#107348)

leslie-fang-intel committed 1 year ago

ee0e04ac

Revert PR-107951 to only support new graph capture API in Quantization (#108317)

leslie-fang-intel committed 1 year ago

6c342ec3

x86_inductor_quantizer switches to new graph capture API (#108214)

leslie-fang-intel committed 1 year ago

fb808c30

[DCP]Add unit test for tp checkpoint (#108286)

wz337 committed 1 year ago

aadd86b1

Add Drq to BF16 Higher Tolernace (#108368)

eellison committed 1 year ago

63eee52b

Newer Older