Skip to content

Backport vLLM structured-output reasoning boundary fix (#44297 + #44993) - #19

Draft
danielmccannsayles wants to merge 1 commit into
mainfrom
dmccanns/structured-output-reasoning-boundary
Draft

Backport vLLM structured-output reasoning boundary fix (#44297 + #44993)#19
danielmccannsayles wants to merge 1 commit into
mainfrom
dmccanns/structured-output-reasoning-boundary

Conversation

@danielmccannsayles

@danielmccannsayles danielmccannsayles commented Jul 16, 2026

Copy link
Copy Markdown
Member

Fixes the GLM-5.2 tool-call 500s and malformed JSON under MTP spec decode + reasoning. Neither patch is in v0.24.0 yet — #44297 merged to main after the cut, #44993 is still open. Both required (44993 stacks on 44297).

…oundary

Fixes GLM-5.2 tool-call 500s and malformed JSON under MTP spec decode +
reasoning. #44297 (merged, post-v0.24.0) constrains the reasoning-end
bitmask and trims grammar advance; #44993 (open, approved, stacked on
#44297) advances the grammar across the boundary for all constraint types.

Both verified to apply cleanly to v0.24.0 with patch -p1 --fuzz=0.
@danielmccannsayles danielmccannsayles changed the title Backport vLLM #44297 + #44993: structured-output reasoning boundary fix Backport vLLM structured-output reasoning boundary fix (#44297 + #44993) Jul 16, 2026
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.

1 participant