Skip to content

Fix #189: apply per-role decoding config in run_trajectory#220

Open
marktech0813 wants to merge 1 commit into
mini-router:mainfrom
marktech0813:fix/189-gittensor
Open

Fix #189: apply per-role decoding config in run_trajectory#220
marktech0813 wants to merge 1 commit into
mini-router:mainfrom
marktech0813:fix/189-gittensor

Conversation

@marktech0813

Copy link
Copy Markdown

Summary

Fixes #189 — per-role decoding: config is now applied.

  • Add OpenAICompatiblePool.decoding_for(role) with fallback defaults
  • run_trajectory selects temperature/top_p/max_tokens per role from that resolver
  • Behavior unchanged when no decoding: block is present

Test plan

  • pytest tests/test_decoding_for.py
  • With decoding.worker.temperature=0.2, worker turns no longer force 0.0

@github-actions github-actions Bot added the train Training or routing changes label Jul 16, 2026
@marktech0813
marktech0813 force-pushed the fix/189-gittensor branch 2 times, most recently from a3d1553 to 4168139 Compare July 16, 2026 16:03
OpenAICompatiblePool loaded decoding: but never read it, so every role
used session defaults. Add decoding_for(role) and wire it into
run_trajectory while preserving backward-compatible defaults.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

train Training or routing changes

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Per-role decoding config is loaded but never applied (dead config)

1 participant