SPS interleaves RL and IRL to counteract probability squeezing in LLM reasoning trajectories, improving Pass@k on five benchmarks while identifying an empirical upper bound on multi-sample performance.
https://arxiv.org/abs/2503.06594
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
fields
cs.CL 3roles
background 2polarities
background 2representative citing papers
NiuTrans.LMT introduces Strategic Downsampling and Parallel Multilingual Prompting to mitigate Directional Degeneration in multilingual MT and releases competitive open models for 60 languages and 234 directions.
A literature survey that organizes prompting, fine-tuning, preference optimization, and context-aware techniques for LLM-based machine translation with emphasis on low-resource languages.
citing papers explorer
-
SPS: Steering Probability Squeezing for Better Exploration in Reinforcement Learning for Large Language Models
SPS interleaves RL and IRL to counteract probability squeezing in LLM reasoning trajectories, improving Pass@k on five benchmarks while identifying an empirical upper bound on multi-sample performance.
-
NiuTrans.LMT: Toward Inclusive and Scalable Multilingual Machine Translation with LLMs
NiuTrans.LMT introduces Strategic Downsampling and Parallel Multilingual Prompting to mitigate Directional Degeneration in multilingual MT and releases competitive open models for 60 languages and 234 directions.
-
Bridging the Linguistic Divide: A Survey on Leveraging Large Language Models for Machine Translation
A literature survey that organizes prompting, fine-tuning, preference optimization, and context-aware techniques for LLM-based machine translation with emphasis on low-resource languages.