benchmark data, training data, and STT-Agent from our paper "STT-Arena: A More Realistic Environment for Tool-Using with Spatio-Temporal Dynamics"
AI & ML interests
LLM reasoning
Recent Activity
benchmark data, training data, and STT-Agent from our paper "STT-Arena: A More Realistic Environment for Tool-Using with Spatio-Temporal Dynamics"
Training and evaluation dataset, model checkpoints in 'Step-Level Sparse Autoencoder for Reasoning Process Interpretation'
RL training and evaluation datasets, and checkpoints in 'Not All Steps are Informative: On the Linearity of LLMs’ RLVR Training'