pytorch/xla

Pull Requests Commits

Added softmax to autocast policy (#8509)

avizon-aws committed 1 year ago

e2d16b6c

When modifying IR node, make sure to not lose the read_only bit (#8505) (#8508)

mcuiaws committed 1 year ago

Verified 38ed80e8

xm.save() should not set sync_xla_data=True when sync'ing. (#8484) (#8504)

mcuiaws committed 1 year ago

Verified ef857715

Add xm.xla_device_kind() to return XLA device kind string. (#8493) (#8506)

mcuiaws committed 1 year ago

Verified de9a01ec

Compute and hash buffer_donor_indices for step marker (#8467) (#8503)

mcuiaws committed 1 year ago

Verified cb880ca8

Fix a DDP graph capture issue (#8500)

Chengji Yao committed 1 year ago

Verified 32e3c609

More branch cut changes (#8488)

tengyifei committed 1 year ago

Verified cfe54bf3

Update CI to use PyTorch 2.6 branch (#8487)

tengyifei committed 1 year ago

Verified 142e7027

Update libtpu/xla/jax pin (#8480)

lsy323 committed 1 year ago

Verified 7261c794

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

Older