xla
Add features needed for vllm
#9092
Open
Overview
Commits
16
Changes
View On GitHub

Add features needed for vllm #9092

wenxindongwork wants to merge 16 commits into pytorch:master from wenxindongwork:torchax-vllm
wenxindongwork
wenxindongwork105 days ago (edited 101 days ago)

Works with

torch-xla 2.8.0.dev20250508

pip3 install torch==2.8.0.dev20250508+cpu --index-url https://download.pytorch.org/whl/nightly/cpu
 pip3 install https://storage.googleapis.com/pytorch-xla-releases/wheels/tpuvm/torch_xla-2.8.0.dev20250508-cp310-cp310-linux_x86_64.whl
Collecting torch-xla==2.8.0.dev20250508
wenxindongwork wenxindongwork marked this pull request as draft 105 days ago
wenxindongwork intercept torch.ops.xla.ragged_paged_attention and dispatch to ragged…
b7686241
wenxindongwork add data @property to View class
9a5f8a2f
wenxindongwork narrow() returns a View
5df82990
wenxindongwork flatten returns a view
0c34344a
wenxindongwork support index_copy_
d37bed0c
wenxindongwork delete commented out code
1f4239cf
wenxindongwork simplify logic
27d5d660
wenxindongwork not needed
f291a3ac
wenxindongwork not needed
bb3763b0
wenxindongwork fix type check
12a157a9
wenxindongwork intercept torch.ops.xla.dynamo_set_buffer_donor_
f9d1df16
wenxindongwork intercept torch._sync as no-op
b673d4de
wenxindongwork import ragged_attention from torch_xla
0f4b40ed
wenxindongwork wrap xla ops in try catch
55b339f2
wenxindongwork wenxindongwork force pushed from f96b9735 to 55b339f2 98 days ago
wenxindongwork formatting
25eb9388
wenxindongwork fix some unit tests
eb0258b4
miladm miladm requested a review from bhavya01 bhavya01 97 days ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Terms Of Use Privacy Policy Refund Policy Impressum