Concept检索片段训练 / 检索片段微调
Retrieved-Span Training
A fine-tuning method for retrieval-augmented generation models that aligns model training with the input format of retrieved passages. This alignment adapts the model to accept retrieval results as input, recovering performance losses caused by input format mismatches in tasks such as query-focused summarization.
Timeline (last 90 days)
Sep 23
After fine-tuning the 406M model specifically for the 2,000-word retrieved segment input format, the previously lost 6.30 points were fully recovered.
Unverified50%