Skip to content

Conversation

@flyinglandlord
Copy link
Collaborator

@flyinglandlord flyinglandlord commented Jan 7, 2025

Change:

  • change launch argument from --simple_constraint_mode to --output_constraint_mode, now user can choose the constriant decode backend from ['outlines', 'xgrammar']
  • add XgrammarBackend used for xgrammar constraint decode, maybe we should merge it with SimpleConstraintBackend later?
  • now we adopt the same request body, the same as vLLM with xgrammar(https://docs.vllm.ai/en/stable/serving/openai_compatible_server.html)
    • user can add guided_grammar to pass a EBNF grammar and execute the constraint decode
    • user can add guided_json to pass a standard json schema and do the constraint decode

@hiworldwzj hiworldwzj merged commit c8c892a into ModelTC:main Feb 26, 2025
1 check passed
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