[GH-ISSUE #6302] Stepwise decoding via websocket, replacing server-side GBNF and JSON schema constraints #50462

Open
opened 2026-04-28 15:57:18 -05:00 by GiteaMirror · 1 comment
Owner

Originally created by @James4Ever0 on GitHub (Aug 10, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/6302

Let the client accesses the token map, server gives the full tensor to client and let the client choose the next token, then continues decoding, or rollbacks previous decoding step.

This eliminates the complexity of GBNF and handles all grammar and decision making steps on the client side. It can even integrates with online learning transformers.

The server shall provide the following commands:

  • Submit continuation task
  • Accept input tensor instead of text
  • Request token map
  • One token forward
  • One token backward
  • Current decoding status
Originally created by @James4Ever0 on GitHub (Aug 10, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/6302 Let the client accesses the token map, server gives the full tensor to client and let the client choose the next token, then continues decoding, or rollbacks previous decoding step. This eliminates the complexity of GBNF and handles all grammar and decision making steps on the client side. It can even integrates with online learning transformers. The server shall provide the following commands: - Submit continuation task - Accept input tensor instead of text - Request token map - One token forward - One token backward - Current decoding status
GiteaMirror added the feature request label 2026-04-28 15:57:18 -05:00
Author
Owner

@James4Ever0 commented on GitHub (Aug 11, 2024):

This would definitely be helpful for those LangGraph IDE users and developers.

<!-- gh-comment-id:2282671326 --> @James4Ever0 commented on GitHub (Aug 11, 2024): This would definitely be helpful for those [LangGraph IDE](https://blog.langchain.dev/langgraph-studio-the-first-agent-ide/) users and developers.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#50462