Baosong Yang

SASFT: Sparse Autoencoder-guided Supervised Finetuning to Mitigate Unexpected Code-Switching in LLMs

mGRPO: Unlocking LLM Reasoning through Multilingual Thinking

Sampling-Efficient Test-Time Scaling: Self-Estimating the Best-of-N Sampling in Early Decoding

ConText: Driving In-context Learning for Text Removal and Segmentation

Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluation