pytorch/xla

Pull Requests Commits

[torch_xla2] Support `aten.avg_pool2d` and `aten.avg_pool3d` with `ceil_mode=True` (#7806)

chunnienc committed 1 year ago

Verified dd04c588

Allow user to pass a name to torch_xla.compile (#7802)

JackCaoG committed 1 year ago

Verified 6250c02d

Attach callback to a Tensor's underlying buffer (#7793)

will-cromar committed 1 year ago

Verified 02d83221

[benchmarks] Fix verifier error handling and OOM. (#7777)

ysiraichi committed 1 year ago

Verified 5bbc4b3d

Move `_xla_register_custom_call_target` implementation into `PjRtComputationClient` (#7801)

will-cromar committed 1 year ago

Verified 7135a7e9

[torch/xla2] Lower aten_copysign and corrcoef (#7713)

wenxindongwork committed 1 year ago

Verified a3dd4d81

Refresh xla2 doc (#7804)

qihqi committed 1 year ago

Verified 8969043e

Update XLA Pin (#7797)

lsy323 committed 1 year ago

Verified dd3b00ca

make StepMarker's wait block for async device execution (#7794)

JackCaoG committed 1 year ago

Verified dfd7b00d

Use reduce-scatter coalescing for FSDP (#6024)

jeffhataws committed 1 year ago

Verified 7fe070a6

Update function doc strings (#7757)

JackCaoG committed 1 year ago

Verified d29b7611

disable sdpa amp (#7800)

lsy323 committed 1 year ago

Verified 225ec7b3

Add 1-layer gradient accumulation test to check aliasing (#7692)

jeffhataws committed 1 year ago

Verified a5520e5d

Add .rst for quantized op doc (#7760)

lsy323 committed 1 year ago

Verified bc9b6061

Handle SPMD case inside of `ComputationClient::WaitDeviceOps` (#7796)

will-cromar committed 1 year ago

Verified 2766f9f7

Improve device auto-detection (#7787)

zpcore committed 1 year ago

Verified 2ba4e057

Using dummy store when user skips barrier to save open file descriptors (#6834)

barry-jin committed 1 year ago

Verified eef7bb4e

Use std::make_optional (#7795)

cyyever committed 1 year ago

Verified b9e75390

Implement torch.ops._c10d_functional.broadcast (#7770)

JackCaoG committed 1 year ago

Verified e4ca4956

Update the `PT_XLA_DEBUG` to make it recognize the `torch_xla.compile` (#7768)

JackCaoG committed 1 year ago

Verified 4e94ff67

fix comment in #7776 (#7789)

JackCaoG committed 1 year ago

Verified 52ea89ff

Add `+default` to libtpu package name (#7792)

will-cromar committed 1 year ago

Verified ff307735

Use launch() for all multiprocessing (#7764)

zpcore committed 1 year ago

Verified 6c18d50c

Support fullgraph for torch_xla.compile (#7776)

JackCaoG committed 1 year ago

Verified 731eb08b

[Pallas] Fix the doc (#7788)

alanwaketan committed 1 year ago

Verified 942ba7a6

Update libtpu_storage_path (#7786)

bhuvanpkaruturi committed 1 year ago

Verified ef49ddd8

Update GPU Development Doc for Hermetic Python setup (#7781)

lsy323 committed 1 year ago

Verified af7d2ea6

Further reduce number of compiling procs (#7779)

lsy323 committed 1 year ago

Verified e52b53f1

reenable dynamo dynamic shape test (#7775)

JackCaoG committed 1 year ago

Verified b09c9699

Update eager doc to use torch_xla.compile (#7778)

JackCaoG committed 1 year ago

Verified 28de4eb3

Newer Older