pytorch/xla

Pull Requests Commits

Update test_train_spmd_linear_model.py

ManfeiBai committed 1 year ago

Verified 51c05db9

Update test_train_spmd_linear_model.py

ManfeiBai committed 1 year ago

Verified 712855a6

Update test_train_spmd_linear_model.py to show clear loss value

ManfeiBai committed 1 year ago

Verified 00de5066

Make it easy to switch to nightly libtpu and jax (#8468)

tengyifei committed 1 year ago

Verified 7dd2697a

[AdHoc] Test fix: clear metrics and simplify the replicated counter (#8478)

rpsilva-aws committed 1 year ago

Verified 7ef4a120

Update torch_xla wheels to have the suffix +cxx11 instead of .cxx11 (#8472)

bhavya01 committed 1 year ago

Verified aa56abcc

Register offloading ops as torch ops (#8477)

tengyifei committed 1 year ago

Verified e8720ba6

flash_attention: support also cross attention. (#8427)

dudulightricks committed 1 year ago

Verified 5790092f

[1/N] Use std::string_view (#7297)

cyyever committed 1 year ago

Verified 58be354d

Add autocast support for XlaPatchedLinear (#8421)

aws-nm9 committed 1 year ago

Verified 2021c0fe

[LoweringContext] SPMD propagation (#8471)

rpsilva-aws committed 1 year ago

Verified b068cab9

Modify the deprecation plan for XLA_USE_BF16 (#8474)

rpsilva-aws committed 1 year ago

Verified 18128175

Fix #8475: add libtpu-wheels registry when building docker (#8476)

tengyifei committed 1 year ago

Verified 47f737a7

Add autocast support for einsum (#8420)

aws-nm9 committed 1 year ago

Verified 7b5aca68

Remove Jack and Will from code owner (#8473)

JackCaoG committed 1 year ago

Verified 5c062ea3

Support S32/U32 indices for BWD embedding & Neuron implicit downcast (#8462)

rpsilva-aws committed 1 year ago

Verified 00c0e966

[LoweringContext] Support an optimized parameter mapping for SPMD (#8460)

rpsilva-aws committed 1 year ago

Verified 5d11f660

Improve wheel naming logic in ansible (#8465)

tengyifei committed 1 year ago

Verified cff26e5b

Update libtpu pin to 2024/12/05 / 0.0.5 stable (#8461)

tengyifei committed 1 year ago

Verified 85cf18ac

Skip the computation if the current block lies above the causal mask diagonal. (#8448)

vanbasten23 committed 1 year ago

Verified 5270d7c4

Stricter testing of sharding annotations after scan (#8456)

tengyifei committed 1 year ago

Verified a48ff6a1

DLPack: fix test using PyTorch API. (#8348)

ysiraichi committed 1 year ago

Verified 4c99d21c

Fix an edge case issue when storing output in the paged attention. (#8431)

vanbasten23 committed 1 year ago

Verified a1a11457

Update torch version to 2.5.1 (#8452)

qihqi committed 1 year ago

Verified 1c65ad74

Updates Contributing.md (#8343)

mikegre-google committed 1 year ago

Verified 19c4d632

[Documentation] Added a section pointing readers to the AI-Hypercomputer/tpu-recipies repo for reference model implementations. (#8412)

mikegre-google committed 1 year ago

Verified 8e10036b

Align rrelu_with_noise schema to reflect noise mutation [branch] (#8363)

IvanKobzarev committed 1 year ago

Verified 73f54ba5

[scan] Add a test under SPMD (#8419)

tengyifei committed 1 year ago

Verified 2ba14fcd

Switch to libtpu package from the libtpu-wheels registry (#8409)

tengyifei committed 1 year ago

Verified ae172982

Update `train_resnet_benchmark.py` reference (#8428)

emmanuel-ferdman committed 1 year ago

Verified b66c2494

Older