radixark/miles

Overview

Miles is an enterprise-focused reinforcement learning framework aimed at post-training large language and vision-language models, with tooling for reward modeling, evaluation, and deployment.[3] It targets teams that need to adapt foundation models to domain-specific behaviors using RL rather than pure supervised fine-tuning. Developers should care because it packages complex RLHF-style workflows into a more reusable, production-ready framework.