v0.6.1.post1 Release Tracker

[simon-mo]

Anything you want to discuss about vllm.

• Fix the AMD weight loading tests #8390
• [Bugfix] Fix InternVL2 inference with various num_patches #8375
• [Hotfix][VLM] Fixing max position embeddings for Pixtral #8399
• [Bugfix] Fix async log stats #8417
• [Hotfix][Pixtral] Fix multiple images bugs #8415
• [Bugfix] Offline mode fix #8376
• [Hotfix][Core][VLM] Disable chunked prefill by default and prefix caching for multimodal models #8425

[ywang96]

#8425

[SolitaryThinker]

Perhaps we can add the flashinfer+multi-step backend PR #7928

[ywang96]

#8428 not a real release blocker, but could help with main branch CI

[SolitaryThinker]

Should add this PR that fixed cudagraph+multistep #8340.

[SolitaryThinker]

#8354

[simon-mo]

Done https://github.com/vllm-project/vllm/releases/tag/v0.6.1.post1