pytorch/xla

Pull Requests Commits

Introduce FSDPv2

alanwaketan committed 1 year ago

1261d5b5

Fix precision issue of pow(int, float) (#6103) (#6172)

qihqi committed 1 year ago

Verified a2a7ff64

Remove redundant log message (#6056) (#6185)

qihqi committed 1 year ago

Verified 94c07f7e

[Backport] Don't explcitly include virtual device in hash (#6167)

jonb377 committed 1 year ago

Verified dbfe5c35

[SPMD] Support mark_shard scalar tensors (#6158)

alanwaketan committed 1 year ago

Verified 98df3d1a

[Cherry-pick] Do manual type conversion for isnan op (#6110) (#6173)

wonjoo-wj committed 1 year ago

Verified 1cba38fe

Handled lifted constants in torch export. (#6111) (#6164)

qihqi committed 1 year ago

Verified 8b3d7a1f

[SPMD] Add apply_backward_optimization_barrier (#6157)

alanwaketan committed 1 year ago

Verified e854d07b

Make r2.2 release wheel compatible with PyPI (#6145)

will-cromar committed 1 year ago

Verified 419dd870

[Backport] Fix bad error message when `PjRtComputationClient` throws exception (#6144)

will-cromar committed 1 year ago

Verified dec978dc

Remove flaky test (#6120)

jonb377 committed 1 year ago

Verified 41d0fe58

[Backport]Handle negative dim for Diagonal Scatter (#6123) (#6129)

JackCaoG committed 1 year ago

Verified f87e9718

[Backport]Add out-of-place all-gather coalesced (#6059) (#6108)

JackCaoG committed 1 year ago

Verified 9a54abab

Update Troubleshotting doc with new PT_XLA_DEBUG (#6039) (#6095)

JackCaoG committed 1 year ago

Verified 53579b1d

[Backport]Add out-of-place reduce-scatter coalescing (#6109)

JackCaoG committed 1 year ago

Verified d1a7e9a0

Increase delay in trace_detached test (#6091)

jonb377 committed 1 year ago

Verified cd186bcd

[Backpoort]Move 0-dimensional tensors to CPU before copying to XLA. (#6071) (#6094)

JackCaoG committed 1 year ago

Verified c2cacdab

[Cherry-pick] [OpenXLA-pin] Update OpenXLA-pin to Nov29 (#6053) (#6079)

ManfeiBai committed 1 year ago

Verified 17cf6741

Add documentation and python API for persistent cache (#6074)

jonb377 committed 1 year ago

Verified a61ddef3

[Backport] Move CUDA tensors to CPU before moving to XLA (#6066)

will-cromar committed 1 year ago

Verified f77d1fac

[Backport]Add a spmd for application to pass in topology (#5931) (#6054)

JackCaoG committed 1 year ago

Verified 45742d2d

Enable persistent compilation caching (#6065)

jonb377 committed 1 year ago

Verified 01405e18

[Backport] Add reduce-scatter coalescing for FSDP/ZeRO1 (#5956) (#6055)

JackCaoG committed 1 year ago

Verified 8472abdf

[Cherry-Pick] Incorporate compilation environment in hash (#6050)

jonb377 committed 1 year ago

Verified b576d85f

[Backport][dynamo] Move CPU tensor constructor nodes to XLA. (#5922) (#6052)

JackCaoG committed 1 year ago

Verified aaccf540

Make PT_XLA_DEBUG analysis's number of frame printed configurable (#6037)

JackCaoG committed 1 year ago

Verified 616e91e6

Add a persistent cache (#5800)

jonb377 committed 1 year ago

Verified f06f6c00

Fix bug with IsTpuDevice for PJRT_DEVICE=TPU_C_API (#6031)

yeounoh committed 1 year ago

Verified 216e07dc

Enable Cumsum with long input (#5974)

JackCaoG committed 1 year ago

Verified 0287a3b5

Enable reduce scatter test on gpu (#6019)

JackCaoG committed 1 year ago

Verified b7cf37af

Older