pytorch/xla

Pull Requests Commits

Test the einsum linear a bit

alanwaketan committed 2 years ago

4959c293

Use einsum for linear

alanwaketan committed 2 years ago

0f5bbb9c

support grad shard

alanwaketan committed 2 years ago

89924ddf

alanwaketan committed 2 years ago

511e7581

Disable allow_spmd_sharding_propagation_to_output

alanwaketan committed 2 years ago

c61e0468

modifications to support output sharding

alanwaketan committed 2 years ago

47717709

Update XLA pin to 2023/08/08

JackCaoG committed 2 years ago

c372c670

alanwaketan committed 2 years ago

1f661545

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

Older