pytorch/xla

Pull Requests Commits

rename segment_id for pallas kernel to make code less confusing

JackCaoG committed 338 days ago

c4db5985

JackCaoG committed 338 days ago

5e2cb30e

Support SegmentID when doing data prallel SPMD

JackCaoG committed 338 days ago

0453dd1e

Use regular `torch.Tensor` for CPU tensors (#8416)

qihqi committed 338 days ago

Verified 20f51661

Update XLA pin to 11/22 (#8411)

JackCaoG committed 340 days ago

Verified 39e67b5d

scan and scan_layers (#7901)

tengyifei committed 343 days ago

Verified 51575db4

Add an example that train the torchtitan version of llama. (#8400)

qihqi committed 343 days ago

Verified 31d348e5

Force specialize_float = True for torch_xla (#8404)

ezyang committed 344 days ago

Verified d503ca54

Only write to HBM at the last iteration. (#8393)

vanbasten23 committed 345 days ago

Verified d572aeb5

Unify how cpp test is being run (#8391)

JackCaoG committed 346 days ago

Verified 2ec22641

Update ARC runner version to 2.321.0 (#8397)

mbzomowski committed 346 days ago

Verified fbed916a

Update XLA pin to 11/17 (#8392)

JackCaoG committed 347 days ago

Verified 0c3d54bf

Fix the fallback op in SPMD (#8386)

JackCaoG committed 347 days ago

Verified defd0a9d

[torch_xla2] add `nn.functional.multilabel_margin_loss` ... `randint` (#8378)

ManfeiBai committed 350 days ago

Verified 91f5c8ad

Fix incorrect xla type for view_as_complex (#8382)

JackCaoG committed 350 days ago

Verified bd4006ef

update the path to tsl/protobuf/error_codes.pb.h (#8383)

JackCaoG committed 350 days ago

Verified 5b3c3315

Support the case where query_len>context_len in the multi-queries paged attention. (#8356)

vanbasten23 committed 352 days ago

Verified 102cd489

Expose lazy all number special scalars (#8361)

rpsilva-aws committed 352 days ago

Verified f610f704

[torch_xla2] Enable nn.functional.cosine_embedding_loss (#8368)

ManfeiBai committed 352 days ago

Verified d4c90c91

Fix op info test for svd_lowrank (#8376)

dvhg committed 352 days ago

Verified 7220aee4

Revert S64 and U64 type change for Neuron device due to hardcoded dtype checks in torch-xla (#8375)

aws-nm9 committed 352 days ago

Verified bc227f7f

Remove unnecessary prints in PagedAttention (#8374)

WoosukKwon committed 353 days ago

Verified baa14f36

Fix XLA type of the view_as_real (#8370)

JackCaoG committed 353 days ago

Verified 071ddfa6

Improve some typing annotations (#8369)

tengyifei committed 354 days ago

Verified c4f27718

Add tengyifei to docs code owners (#8365)

tengyifei committed 358 days ago

Verified 5a4641b4

Add mikegre-google as docs owner (#8364)

tengyifei committed 358 days ago

Verified 09f36bee

XLA downcast for S64 and U64 for Neuron (#8347)

aws-nm9 committed 358 days ago

Verified 03f07e2a

Add aten::lift_fresh to the exception list that converts input to torch tensor (#8360)

barney-s committed 359 days ago

Verified 6454b42f

Add inplace ops for sigmoid and tanh (#8358)

barney-s committed 359 days ago

Verified 77abab25

Revive "Modularize zero step function and make it customizable" #7233 (#8332)

jeffhataws committed 360 days ago

Verified 7d6a4f25

Older