MSA reduces attention computation for million-token contexts by a factor of 28.4 through blockwise sparse selection and achieves practical speedups via co-design of algorithm and GPU kernel.
Agent-EvalKit automates the evaluation of AI agents through structured test-case generation, observability instrumentation, and combined code and LLM-based metrics directly in the development environment.
Grammar-Constrained Decoding (GCD), a technique for ensuring syntactically correct code, opens a new jailbreak method for attackers with a success rate over 30 percentage points higher than previous approaches.
Aligning router rows with the principal singular directions of their associated expert matrices improves the efficiency and stability of Mixture-of-Experts models.
The Claw-SWE-Bench framework demonstrates that adapter design is critical for code agents: with a minimal adapter, OpenClaw achieves 19.1% Pass@1, with a complete adapter 73.4%.
Arbor enables AI-driven research through systematic hypothesis management and achieved an average of 2.5x higher improvements than existing code models on six test tasks.
Bebop uses rejection sampling and TV loss optimization to maintain stable MTP acceptance rates during RL training and accelerates rollouts by up to 1.8x.
