Skip to content

remove stale issue #119 xfail for greedy paged-parity test#149

Merged
LxYuan0420 merged 1 commit intovllm-project:mainfrom
LxYuan0420:chore/remove-stale-issue119-xfail
Mar 11, 2026
Merged

remove stale issue #119 xfail for greedy paged-parity test#149
LxYuan0420 merged 1 commit intovllm-project:mainfrom
LxYuan0420:chore/remove-stale-issue119-xfail

Conversation

@LxYuan0420
Copy link
Collaborator

This PR is:

  • To remove a stale xfail on test_greedy_output_matches that was originally added for issue Metal paged-attention parity mismatch vs standard path #119.
  • To align test expectation with current main behavior after paged-path fixes already merged.
  • To keep parity tracking accurate while leaving batched behavior to its own tracking path.

Context

Issue #119 reported token mismatch parity failures between:

  • standard MLX KV cache path, and
  • Metal paged-attention path.

Since then, two key fixes landed:

With those changes, the old greedy mismatch from #119 no longer reproduces on main, so the greedy xfail is stale.

Verification

pytest -q tests/test_metal_kernel_paged.py::TestMetalKernelPagedVsStandard::test_greedy_output_matches -s
pytest -m slow -q tests/test_metal_kernel_paged.py

Signed-off-by: Yuan Lik Xun <lxyuan0420@gmail.com>
@LxYuan0420 LxYuan0420 self-assigned this Mar 9, 2026
@LxYuan0420 LxYuan0420 requested a review from ericcurtin March 9, 2026 06:39
Copy link
Collaborator

@WindChimeRan WindChimeRan left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. safe to merge. (seems like we still need @ericcurtin to merge these)

@LxYuan0420 LxYuan0420 merged commit 97a2844 into vllm-project:main Mar 11, 2026
5 checks passed
renqHIT added a commit to renqHIT/vllm-metal that referenced this pull request Mar 13, 2026
The xfail on `test_batched_decode_matches` was added for issue vllm-project#119
(B=2 batched GEMM producing different floats than B=1). The test now
passes consistently on main after recent paged kernel fixes (vllm-project#146, vllm-project#151).
This follows PR vllm-project#149 which removed the same stale xfail for the
greedy single-request test.

Signed-off-by: Qiang <qren@integralads.com>
LxYuan0420 pushed a commit that referenced this pull request Mar 17, 2026
…167)

`test_metal_kernel_paged.py` re-implements vllm-metal internals (cache
setup, prefill/decode orchestration, context management) to compare two
paths. This scaffolding introduces additional complexity, making
failures hard to attribute.

Delete it and add its prompts to `test_paged_deterministic.py`, which
does the same comparison end-to-end through the real vLLM stack against
golden tokens.

Related: 
#158 
#149
#119

---------

Signed-off-by: ran <hzz5361@psu.edu>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants