…learning with verifiable rewards by usingLorem Ipsum perturbations to enhance exploration in large language model training. AI-generated summary Reinforcement learning with verifiable rewards, particularly Group Relative Policy Optimization (GRPO), has significantly…
…Deferring the Distiller's training step (backward pass and weight update) to the CPU-bound post-processing intervals to hide the latency is a fantastic practical touch, keeping the throughput overhead under…
…AI-generated summary In settings where labeled verifiable training data is the binding constraint, each checked example should be allocated carefully. The standard practice is to use this data directly on the…
…fixed budget--turning merging from heuristic practice into a computationally efficient, planable alternative to multitask training . This suggests a scaling principle for distributed generative AI : predictable gains can be achieved by composing…
…beyond their training length. Extensive ablation studies validate the contribution of each component and provide guidance for practical configuration. Our code is publicly available at https://github.com/Musubi-ai/Mela Get…
…AI-generated summary Continual post-training aims to extend large language models (LLMs) with new knowledge, skills, and behaviors, yet it remains unclear when sequential updates enable capability transfer and when they…
…We present AEM, a supervision-free credit assignment method that adaptively modulates entropy dynamics during RL training to improve the exploration-exploitation trade-off . Since in agentic RL the environment is typically…
…We address this bottleneck in the Byte Latent Transformer (BLT) through new training and generation techniques. First, we introduce BLT Diffusion (BLT-D), a new model and our fastest BLT variant, trained…
…Existing approaches either rely on manual skill curation , prescribe heuristic skill operations, or train for short-horizon skill operations. However, they still struggle to learn complex long-term curation policies from indirect…
…that trains a single policy to simultaneously evolve skill selection, utilization, and distillation capabilities using a shared task-outcome objective, demonstrating superior performance over existing baselines in complex task environments. AI-generated…