Search papers, labs, and topics across Lattice.
Molt is a PyTorch-native training framework designed to streamline the process of agentic reinforcement learning by minimizing the complexity and overhead associated with algorithm modifications and pipeline adjustments. This framework allows researchers to maintain a clear understanding of the entire algorithm flow while training multimodal and mixture-of-experts policies in a fully asynchronous manner. Notably, Molt achieves performance on par with state-of-the-art Megatron-based stacks while maintaining a compact and accessible codebase, making it easier for researchers to iterate and innovate.
Molt reduces the cognitive load on researchers by offering a clean and compact codebase that maintains high performance, enabling faster iterations in agentic reinforcement learning.
Agentic reinforcement learning research is constant algorithm modification, new estimators, new pipeline stages, new rollout schemes, and in mainstream frameworks each change threads through layers of trainer, distributed backend, and rollout glue: the cost lands on the researcher at every iteration. Molt is a PyTorch-native training framework built to keep that cost small: a codebase compact and clean enough for a researcher to hold in their head, and for an AI coding assistant to read and reason about in its entirety, so the algorithm flow can be traced and changed end to end. The agent is an ordinary program, and one asynchronous loop trains multimodal and mixture-of-experts policies while never training on a token it did not generate, consistent in tokens, policy versions, and model semantics. Leanness does not cost performance: under a matched, fully asynchronous protocol, Molt is statistically comparable to a state-of-the-art Megatron-based stack. Molt is open source and provides recipes and containers at https://github.com/NVIDIA-NeMo/labs-molt.
NVIDIA