Abstract: Class-Incremental Learning (CIL) enables models to continuously learn new classes while mitigating catastrophic forgetting. Recently, Pre-Trained Models (PTMs) have greatly enhanced CIL ...

⚠️ Building flash-attn v2.7.4 with CUDA 12.8 on Windows cannot be completed because of GitHub Actions’ processing-time limits. In the future, I plan to add a self-hosted Windows runner to resolve this ...

We propose TraceRL, a trajectory-aware reinforcement learning method for diffusion language models, which demonstrates the best performance among RL approaches for DLMs. We also introduce a ...

A worm appears on the playground. A block tower collapses, again. Two children want the same toy, and voices begin to rise. In early childhood classrooms, spontaneous moments like these happen ...