slime-user

What problem does it solve?

SLIME User Guide provides a structured, extensible workflow for RL training of language models, unifying rollout, reward modeling, and backend integrations to reduce setup complexity and accelerate experimentation.

Core Features & Use Cases

• Multi-backend training with Megatron and FSDP backends for scalable RL training.
• Multi-turn interactions and tool calling with custom generation and reward functions.
• Dynamic sampling and partial rollout to improve sample efficiency across datasets.
• Data management with JSONL prompts, labels, and metadata for RL tasks.
• Troubleshooting and customization hooks for bespoke RL workflows. Use cases include training language models with RL, evaluating reward models, and deploying tool-enabled generation.

Quick Start

Launch a SLIME training session with a minimal setup and provide a dataset and HF checkpoint path.