Skip to content

feat: BRGS helpers#9

Open
sheshansh-ctx wants to merge 8 commits intomainfrom
sheshansh/brgs_helpers
Open

feat: BRGS helpers#9
sheshansh-ctx wants to merge 8 commits intomainfrom
sheshansh/brgs_helpers

Conversation

@sheshansh-ctx
Copy link
Copy Markdown
Collaborator

No description provided.

sheshansh-ctx and others added 8 commits March 21, 2026 21:51
When the LLM returns a malformed ranking (missing or duplicate passage
IDs), retry up to 5 times by appending the bad response and a correction
prompt as new conversation turns. Accumulates token/latency stats across
retries. Falls back gracefully if all retries fail.

Also forwards **kwargs through get_response to chat, enabling callers
to pass additional parameters like response_format.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Unified the two failure paths (JSON parse failure vs ranking validation
failure) through a single _validate_ranking call. Previously, consecutive
JSON parse failures would skip validation entirely, leaving `permutation`
uninitialized. Now:
- Both paths always produce a validated permutation
- On JSON failure: extract integers via regex from raw response, pass
  through _validate_ranking which deduplicates and appends missing
  indices in identity order
- On complete failure: falls back to identity permutation
- Cache is NOT populated for degraded results (existing guard on
  missing_indices)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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