strisunshinewentingwang/pytorch

Pull Requests Commits

Refactor test_inductor_benchmark into test_single_dynamo_benchmark helper (#92665)

ezyang committed 1 year ago

1bf51201

Add a warning about performance cost of set_default_device (#92703)

ezyang committed 1 year ago

85a1f022

Implement SymBool (#92149)

ezyang committed 1 year ago

5c6f5439

Dispatch the auxiliary frobenius_norm and nuclear_norm to better implementations and deprecate them (#81763)

lezcano committed 1 year ago

34e8eb22

[cublas][cublasLt] Fall back to unfused `addmm` for 2-byte-aligned inputs (#92201)

eqy committed 1 year ago

1af40d51

[quant][fx][pt2e][be] Store node_name_to_target_dtype to node.meta["target_dtype_info"] (#92574)

jerryzh168 committed 1 year ago

a74c8df7

[optim][foreach] Do NOT inplace modify gradients (#92706)

janeyx99 committed 1 year ago

de0375e7

[optim][NAdam] Fix discrepancy between mt vs st impl (#92699)

janeyx99 committed 1 year ago

2b885e1f

fix the formatting of runtime error msg in prims _cat_meta (#92124)

dshi7 committed 1 year ago

896b6d87

Shard mac to 3 (#91277)

clee2000 committed 1 year ago

703265e5

Don't allow recomputing a node that *must* be materialized in the backwards pass (#90896)

Chillee committed 1 year ago

d6c3468f

Fix GroupNorm backward prop on CUDA (#92671)

malfet committed 1 year ago

97b7e4cd

[CUDA][CUBLAS][BFloat16] Tenatively disable reduced precision reductions for some matmul tests (#92599)

eqy committed 1 year ago

8c0289a6

[inductor] Lower torch.exp2 and use it for torch.pow(2, x) (#92632)

peterbell10 committed 1 year ago

56440594

Add GHA side support for ciflow/inductor-perf-test-nightly (#92693)

ezyang committed 1 year ago

5a134440

Create autograd Function for aot_autograd backward only when needed (#92688)

soulitzer committed 1 year ago

a3efa9d7

[PT-D][checkpoint] Resolve no such file or directory issue when checkpointing on multi hosts (#92553)

wz337 committed 1 year ago

eee2869e

Foreach gradient clipping (#91846)

milesial committed 1 year ago

e4d83d54

Clean up argparser help (benchmarks/dynamo/distributed.py) (#92687)

wconstab committed 1 year ago

44b7a0b7

Deprecate capture hooks except distributed use case (#92653)

soulitzer committed 1 year ago

9db4323e

Delete get_pyobj() entirely (#92638)

ezyang committed 1 year ago

c4501593

Mark test monitoring as an optional process (#92658)

huydhn committed 1 year ago

56107660

Revert "Run dynamo/test_dynamic_shapes serially (#92215)"

pytorchmergebot committed 1 year ago

8b3e35ea

update vmap to accept nones (#91644)

srossross committed 1 year ago

fb3d9f39

[Dynamo] Preserve source_fn in node.meta (#92399)

SherlockNoMad committed 1 year ago

2fb328eb

[decomp] Use new squeeze.dims overload in decompositions (#91602)

peterbell10 committed 1 year ago

dd760c98

logaddexp2: Use log1p and exp2 (#92116)

peterbell10 committed 1 year ago

2af2952c

lint fix (#92685)

ZainRizvi committed 1 year ago

67bb5236

Revert "Add meta kernel coverage for aten.unsafe_split, aten.unsafe_chunk (#92608)"

pytorchmergebot committed 1 year ago

2891cecd

Update android/README.md, how to build pytorch android from source (#92356)

vfdev-5 committed 1 year ago

215f4fc3

Newer Older