pytorch/xla

Pull Requests Commits

Make matmul faster

alanwaketan committed 2 years ago

b61fb961

alanwaketan committed 2 years ago

0ff65b16

remove reshape in linear backward

alanwaketan committed 2 years ago

84b86617

alanwaketan committed 2 years ago

8c12dd09

Introudce XLAPatchedMatmul that has a backward

alanwaketan committed 2 years ago

0fc3da72

introduce _xla_matmul

alanwaketan committed 2 years ago

7917c37e

alanwaketan committed 2 years ago

37dcf561

Patch nn.linear

alanwaketan committed 2 years ago

aed42f7e

Do custom call in _propagate_xla_data

alanwaketan committed 2 years ago

ffcc94c7

alanwaketan committed 2 years ago

bedffa83

alanwaketan committed 2 years ago

9aa2f921

Make it in-place op

alanwaketan committed 2 years ago

b0736bf2

Don't use XLAShardedTensor

alanwaketan committed 2 years ago

eee78554

alanwaketan committed 2 years ago

8d2e76bf

Fix the custom call hlo

alanwaketan committed 2 years ago

36ab2bc8

Modify test and small fix

alanwaketan committed 2 years ago

13359c1c

connect the pybind

alanwaketan committed 2 years ago

2f1aded9

add tensor methods

alanwaketan committed 2 years ago

123290d9

alanwaketan committed 2 years ago

f558b236

alanwaketan committed 2 years ago

148e6c92

alanwaketan committed 2 years ago

8a09a2bc

alanwaketan committed 2 years ago

9e6b6170

Add repo for libcudnn8=8.7.0.84 and CUDA 11.8 (#5425)

mateuszlewko committed 2 years ago

Verified f06c31f5

[PJRT|TPU] Update `test_xla_devices_single_process_all_chips` for expected device number (#5421)

ManfeiBai committed 2 years ago

Verified 2d97d3a1

[SPMD] named partition spec support (#5415)

yeounoh committed 2 years ago

Verified 85453d27

Fix the incorect clone arg condition in dynamo bridge (#5414)

JackCaoG committed 2 years ago

Verified 1835771c

SPMD cross slice-replication using partial_replication sharding (#5411)

yeounoh committed 2 years ago

Verified 6852df1d

Revert "Partially replicate lower-rank tensors (#5409)" (#5412)

yeounoh committed 2 years ago

Verified 628ba698

Partially replicate lower-rank tensors (#5409)

jonb377 committed 2 years ago

Verified 56a6a02a

Fix TPU collective ops test for multi-host TPUs (#5408)

will-cromar committed 2 years ago

Verified 2658c882

Older