Researchers have developed AnyJev, a method to extract typed decisions from pre-trained language models without requiring gradient steps or parameter changes. AnyJev addresses two key defects: the model's tendency to assign higher probabilities to certain labels or positions, and its inability to directly output typed decisions. By dividing out a label prior and averaging log-probabilities over cyclic rotations of option lists, AnyJev significantly improves accuracy and reduces order-flip rates on multi-option tasks. The system also incorporates a stopping rule to optimize performance and speed, serving decisions up to 2.2 times faster on vLLM. AI
IMPACT Improves efficiency and accuracy of typed decision extraction from LLMs, potentially speeding up applications requiring structured outputs.
RANK_REASON The cluster describes a technical report detailing a new method for extracting typed decisions from language models, which is a research contribution.
Read on Hugging Face Daily Papers →
- alphaXiv
- AnyJev
- arXiv
- CatalyzeX
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- IArxiv Recommender
- ScienceCast
- vLLM
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →