Report the xgrammar draft-token rejection with MTP k=2 to the vllm-backport maintainers; report the GNmac TP4 sm80 attention long-context corruption to GNmac with the six-context scorer; ...

  • Swimlane: Default swimlane
  • Column: Idea
  • Position: 32
  • Assignee: not assigned
  • Creator: Ivan
  • Started:
  • Created: 10/30/2026 09:18
  • Modified: 10/30/2026 09:18
  • Moved: 10/30/2026 09:18
  • 8 GB
  • 10 GB
Description

Full title: Report the xgrammar draft-token rejection with MTP k=2 to the vllm-backport maintainers; report the GNmac TP4 sm80 attention long-context corruption to GNmac with the six-context scorer; post the PR #60477 confirmation comment

Reported by / on it: @darkmaniac7

Source: https://github.com/vllm-project/vllm/pull/60477

Wiki page: Known bad combinations

Cards: both

Card D-116, seeded from the cmpunlocker Discord with credit (reference github:vllm-project/vllm/2).

Sub-Tasks
Internal links
Comments