pytorch/xla

Pull Requests Commits

reduce dcoder model size to make eager runable on v2-8

JackCaoG committed 1 year ago

7d183365

Clean up and add doc for dynamo_mark_sharding (#7281)

wonjoo-wj committed 1 year ago

Verified 61e03890

[Doc] Add an instruction to install nightly wheels. (#7275)

vanbasten23 committed 1 year ago

Verified 62015ee1

Disable remote cache for PRs from forks (#7282)

will-cromar committed 1 year ago

Verified a7976d07

Make a dynamo inference test that runs on TPU (#7264)

JackCaoG committed 1 year ago

Verified 1d27c2a3

Document process to create a new PJRT plugin (#7277)

will-cromar committed 1 year ago

Verified 84dd2f2b

Avoid log spam (#7278)

JackCaoG committed 1 year ago

Verified feb1e63e

Create common `setup` action to wrap setup steps (#7272)

will-cromar committed 1 year ago

Verified a5eba0fa

[SPMD] Fix reduce_scatter for TPUv2 (#7248)

alanwaketan committed 1 year ago

Verified 0025ca76

Integrate dlpack to dynamo. (#7173)

vanbasten23 committed 1 year ago

Verified c216d26c

Update docs/gpu.md (#7273)

Ricktho1 committed 1 year ago

Verified b196bf07

Updated the examples folder readme file (#7208)

sitammeur committed 1 year ago

Verified ca72f1a5

Revert the mul change (#7271)

JackCaoG committed 1 year ago

Verified 286b31f0

Enable PJRT plugins by default (#7268)

will-cromar committed 1 year ago

Verified 1cad403a

support eager mode with torch.compile (#7256)

JackCaoG committed 1 year ago

Verified 192dec65

Increase max in flight operation to accommodate eager mode (#7263)

JackCaoG committed 1 year ago

Verified 0548ec3d

Revert "Modularize zero step function and make it customizable" (#7259)

JackCaoG committed 1 year ago

Verified ea2a6f77

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

karakusc committed 1 year ago

Verified 51f8d3e6

Bump JAX version (#7258)

vanbasten23 committed 1 year ago

Verified 1e3e6b83

Update the XLA pin (#7254)

vanbasten23 committed 1 year ago

Verified a2edc217

Supporting passing ML framework name and version through the PjRt Client Create options in PyTorch/xla. (#7249)

aman2930 committed 1 year ago

Verified 81f657a2

add torch_xla.experimental.compile for eager mode (#7246)

JackCaoG committed 1 year ago

Verified 90168e8e

Revert "Remove stack dump for compiler failure which can happen often" (#7226)

JackCaoG committed 1 year ago

Verified f4665a73

Move nightly version to 2.5.0 (#7245)

JackCaoG committed 1 year ago

Verified 283abdab

Register custom call with CUDA plugin (#7219)

bhavya01 committed 1 year ago

Verified 86d8967c

Change #include <c10/util/Optional.h> to #include <optional> (#7241)

cyyever committed 1 year ago

Verified c65e87b3

Refactor a bit so that the ops/ is a standalone module (#7250)

qihqi committed 1 year ago

Verified 8cd6ad43

Updating ml_framework_name and version to pytorch/xla2 in jax_configs. (#7211)

aman2930 committed 1 year ago

Verified 3f54fa21

Enable eager mode for PyTorch/XLA (#7234)

JackCaoG committed 1 year ago

Verified 4d0b94fd

Use dest_offsets directly in LoadPlanner (#7155)

jonb377 committed 1 year ago

Verified 7be1d3d0

Older