Entity · technique

Logit-Contribution Scoring

techniqueactiveprovisionallogit-contribution-scoring-15379b50·1 events·first seen 42h ago

Aliases: Logit-Contribution Scoring

Co-occurring entities

MuSiQue OLMo-3 Gemma-3-4B-IT NoLiMa Qwen3 BABI-Long

More like this (12)

logit lens Log Probability Bias Analysis LALS (Latent Association Leaning Score)LogbQuant logistic regression probes LLM-judged explanation score Exact Posterior Score Estimation for Solving Linear Inverse Problems isomorphism-based scoring reward-induced maximum likelihood 2-Parameter Logistic IRT Model LLM-judge scoring concrete score function

Recent events (1)

5arXiv · cs.CL·42h ago·source ↗

LOCOS: Logit-Contribution Scoring identifies non-literal retrieval heads in long-context LLMs

A new arXiv preprint introduces Logit-Contribution Scoring (LOCOS), a method for identifying attention heads responsible for non-literal retrieval in long-context LLMs — cases where models synthesize answers from meaning rather than copying tokens verbatim. Existing detectors fail at this task because they rely on a literal-copy criterion that misses the output-value (OV) circuit mechanism. Evaluated across Qwen3, Gemma-3, and OLMo-3.1, LOCOS outperforms prior attention-based detectors on the NoLiMa benchmark, with ablation of 50 heads on Qwen3-8B collapsing ROUGE-L from 0.401 to 0.000 while the best baseline retains 0.292. The identified heads are retrieval-specific, leaving parametric recall and arithmetic reasoning unaffected.

Long Context Evolution Evaluation and Benchmarking MuSiQue OLMo-3 Gemma-3-4B-IT +4 more