Skip to content

[VLLM][unified attention] Updating VLLM pin leads to verification issues for the upstream code #5547

@Egor-Krivov

Description

@Egor-Krivov

Updating VLLM pin to d6953beb91da4e9c99be4c0a1304a2d24189535c leads to verification issues for the upstream triton code. The changes includes key optimization for sliding attention performance.

Need to investigate the issue

Sub-issues

Metadata

Metadata

Assignees

Type

Projects

No projects

Relationships

None yet

Development

No branches or pull requests

Issue actions