pytorch/xla

Pull Requests Commits

Addressing comments from Mike

yaoshiang committed 1 year ago

3da81651

yaoshiang committed 1 year ago

d89f44cf

yaoshiang committed 1 year ago

18ce9114

Guide to debugging in PyTorch

yaoshiang committed 1 year ago

6f95d544

Set scoped vmem for paged attention (#8988)

zpcore committed 1 year ago

Verified 366f248a

Fix merge conflict (#8991)

zpcore committed 1 year ago

Verified 083934fd

Add an option for JittableModule to dedup parameters. (#8965)

qihqi committed 1 year ago

Verified cce7cc07

Add a helper class to handle mesh and sharding (#8967)

qihqi committed 1 year ago

Verified 78824755

Fix splash attention test (#8978)

zpcore committed 1 year ago

Verified 13155c9c

Fix call_jax hashing (#8981)

zpcore committed 1 year ago

Verified a48d6a97

Update to ubuntu latest (#8979)

qihqi committed 1 year ago

Verified a37a5852

@assume_pure (#8962)

tengyifei committed 1 year ago

Verified cf764ffa

Avoid re-computing computation hashes (#8976)

rpsilva-aws committed 1 year ago

Verified bcca5eba

Disable one splash attention test (#8970)

zpcore committed 1 year ago

Verified ba5cd4cb

Add tuned parameters for Qwen/Qwen2.5-32B (#8966)

yarongmu-google committed 1 year ago

Verified 4eea5e13

Adapt Splash Attention from TorchPrime (#8911)

zpcore committed 1 year ago

Verified 45830519

Make scan-based GRU support `batch_first` parameter. (#8964)

iwknow committed 1 year ago

Verified aad87e61

[call_jax] support returning PyTree from the JAX function (#8957)

tengyifei committed 1 year ago

Verified 297f5de6

Remove deprecated typing module in ansible flow (#8955)

tengyifei committed 1 year ago

Verified e77bd2f6

Add test to ensure scan-based and the standard GRU are interchangeable. (#8949)

iwknow committed 1 year ago

Verified d4d997dc

Update to pytorch 2.6 (#8944)

qihqi committed 1 year ago

Verified 085f8bc1

Pin update to 20250406 (#8945)

tengyifei committed 1 year ago

Verified 0837fac3

Use pages_per_seq * page_size instead of directly passing max_model_len (#8950)

yaochengji committed 1 year ago

Verified 9ec626e6

Fix ragged_paged_attention op signature (#8943)

yaochengji committed 1 year ago

Verified 80929a91

add rc4 trigger (#8941)

zpcore committed 1 year ago

Verified ac9a39f4

Add block size table for ragged_paged_attention (#8942)

yaochengji committed 1 year ago

Verified cbff76f5

Add alternative dynamo backend (#8893)

qihqi committed 1 year ago

Verified fb6038de

Trigger build for 2.6.1 patch (#8938)

zpcore committed 1 year ago

Verified 379ebd5f

Update and rename torch_xla2.yml to torchax.yml (#8936)

tengyifei committed 1 year ago

Verified 9e23ca85

Rewrite scan-based GRU based on nn.GRU (#8914)

iwknow committed 1 year ago

Verified 145c33ae

Older