A DPO-tuned summarizer compresses QA contexts to 12-20% of their length, raising EM-per-token 6.8x but lowering absolute EM.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Context Filtering with Reward Modeling in Question Answering
A DPO-tuned summarizer compresses QA contexts to 12-20% of their length, raising EM-per-token 6.8x but lowering absolute EM.