Skip to content

OpenConceptLab/ocl_issues#2601 | Qwen model logit score and LRU cache#889

Open
snyaggarwal wants to merge 1 commit into
masterfrom
issues#2601
Open

OpenConceptLab/ocl_issues#2601 | Qwen model logit score and LRU cache#889
snyaggarwal wants to merge 1 commit into
masterfrom
issues#2601

Conversation

@snyaggarwal

Copy link
Copy Markdown
Contributor

Linked Issue

Closes OpenConceptLab/ocl_issues#2601

@snyaggarwal snyaggarwal self-assigned this Jul 3, 2026

@filiperochalopes filiperochalopes left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

RERANKER_SCORE_PROFILES could work as a router for what function to use to normalize the score 0-1. Use Qwen/ looks too specific. But works

Comment thread core/common/search.py
def _get_activation_fn(self):
"""Return the score activation required by the configured reranker model."""
model_name = self.model_name or self.default_model
if isinstance(model_name, str) and model_name.startswith(self.QWEN_RERANKER_PREFIX):

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For me this approach is too hardcoded, maybe its interesting to do somethins like:

RERANKER_SCORE_PROFILES = {
    "Qwen/": "logit_sigmoid",
    "BAAI/bge-reranker": "logit_sigmoid",
    "cross-encoder/ms-marco-": "raw_rank_score",
    "ncbi/MedCPT-Cross-Encoder": "logit_sigmoid",
}

But it also could be too YAGNI, so let's hear @paynejd thoughts

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.

Qwen/Qwen3-Reranker-0.6B memory load and score between 0-1

2 participants