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

grpo

Configures and runs GRPO reinforcement learning workflows with the Tinker API for math, code, and custom environments.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Set up and run reinforcement learning with verifiable rewards (RLVR/GRPO) for math, code, or custom environments using the Tinker API. Use when the user wants to do RL training, GRPO, reward-based optimization, or train with verifiable rewards.

Core Features & Use Cases

  • End-to-end GRPO workflow orchestration using the Tinker API for RL with verifiable rewards.
  • Supports math, code, and custom environments via well-defined env builders and existing recipes in tinker_cookbook.
  • Provides guidance to configure models, environments, rewards, and training loops, reusing patterns from the RL ecosystem.

Quick Start

Configure your model, environment, and reward signal, then run the GRPO workflow to begin RL training with verifiable rewards.

Frequently Asked Questions about grpo

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

FAQPage Schema
How do I set up reinforcement learning with verifiable rewards using the Tinker API?▼

To set up reinforcement learning with verifiable rewards using the Tinker API, configure your model, environment, and reward signal, then run the GRPO workflow to begin training. This orchestrates the end-to-end RLVR pipeline.

Can I train models with GRPO on custom environments?▼

Yes, GRPO supports training on custom environments. The workflow uses well-defined environment builders and existing recipes, allowing you to apply both programmatic verifiable rewards and learned reward signals.

What is needed to run a GRPO training pipeline for math and code tasks?▼

Running a GRPO training pipeline for math and code tasks requires defining the environment and reward design, then configuring the model using Tinker API patterns to execute the reinforcement learning loop.

Does this GRPO workflow support learned reward signals as well as verifiable rewards?▼

Yes, the GRPO workflow supports both verifiable programmatic rewards and learned reward signals. This allows flexible reward design when training models across math, code, or custom environments via the Tinker API.

When should I use GRPO with verifiable rewards instead of standard reinforcement learning?▼

Use GRPO with verifiable rewards when you have programmatic metrics for math or code tasks. It integrates environment builders and reward design to orchestrate end-to-end RL training using the Tinker API patterns.

Are there existing recipes for environment design in the GRPO training pipeline?▼

Yes, the GRPO training pipeline reuses patterns and existing recipes from the tinker_cookbook RL ecosystem. This provides guidance for configuring environments, rewards, and training loops effectively.

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
qcmuuqcmuu

grpo-rl-training

Fine-tune language models with GRPO using TRL and custom reward functions.

Community
Advanced
ovachieverovachiever
52

grpo-rl-training

Apply GRPO/RL training with TRL to optimize language models.

Community
Advanced
InverterNetworkInverterNetwork

fine-tuning-with-trl

Fine-tune large language models with TRL for SFT, DPO, and GRPO workflows.

Official
Advanced