skills.rest
thinking-machines-lab
thinking-machines-labOFFICIAL
Explore all skills by this author →
WebsiteGitHub

multiturn-rl

Coordinate multi-turn RL training workflows against interactive environments using Tinker APIs.

4.0k|507|Updated Jul 14, 2025
Search Tags:#environment#rl#multiturn#llm-training#tinker#async-config
One-click install
npx skills add https://github.com/thinking-machines-lab/tinker-cookbook --skill multiturn-rl
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: multiturn-rl
Source: https://github.com/thinking-machines-lab/tinker-cookbook/tree/main/.claude/skills/multiturn-rl
Command: npx skills add https://github.com/thinking-machines-lab/tinker-cookbook --skill multiturn-rl

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Multi-turn RL training for interactive environments using the Tinker API, enabling agents to reason across turns, manage tool use, and optimize policies in dynamic tasks.

Core Features & Use Cases

  • Orchestrates Harbor terminal RL, Search-RAG, and multiplayer RL pipelines for end-to-end training
  • Provides guidance on environment types, turn structures, and configuration patterns (HarborTask, HarborDatasetBuilder, AsyncConfig, Config)
  • Includes steps to run, test, and extend multi-turn RL experiments across custom environments

Quick Start

Run the multi-turn RL training workflow by executing the Harbor RL, Search-R1, or multiplayer RL trainer script.

Frequently Asked Questions about multiturn-rl

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run multi-turn RL training for LLM agents in interactive environments?▼

You can run multi-turn RL training by executing Harbor RL, Search-R1, or multiplayer trainer scripts to coordinate end-to-end workflows against interactive environments using Tinker APIs.

What is multi-turn reinforcement learning and when do I need it for LLM training?▼

Multi-turn reinforcement learning trains agents to reason across multiple interactions, manage tool use, and optimize policies in dynamic tasks like terminal sandboxes or multiplayer games.

Can I configure custom turn structures and tool outputs for RL rollouts?▼

Yes, multi-turn RL training supports configurable turn structures and tool outputs, allowing you to define custom environments and manage dynamic task interactions.

Does Tinker support async rollout configurations for reinforcement learning?▼

Yes, Tinker supports async rollout configurations through AsyncConfig definitions, enabling coordinated multi-turn RL training workflows across interactive environments.

What environments are compatible with multi-turn RL training pipelines?▼

Compatible environments include terminal sandbox tasks, tool-assisted search and RAG pipelines, multiplayer games, and custom user-defined environments configured via HarborTask.

Related Skills

Explore More Software Engineering Skills→
thinking-machines-labthinking-machines-lab
4.0k

distillation

Automate knowledge distillation from teacher to student models via the Tinker API.

Official
Advanced
Lord1EgyptLord1Egypt
2

transformers

Load, run, and fine-tune pre-trained transformer models for NLP, vision, and audio tasks.

Community
Advanced
carterwayneskhizeinecarterwayneskhizeine
19

hermes-atropos-environments

Implements Hermes-AgentRL environments integrating multi-turn tool-calling loops with the Atropos training framework.

Community
Advanced
uv-xiaouv-xiao
1

uv-slime-rl-training

Run GRPO reinforcement learning training for LLMs with Megatron-LM and SGLang.

Community
Advanced