skills.rest
Supporter09
Supporter09COMMUNITY
Explore all skills by this author →
GitHub

torchforge-rl-training

Train PyTorch-native agentic reinforcement learning workflows with GRPO and multi-GPU distributed setups.

Updated May 4, 2026
Search Tags:#llm#reinforcement learning#pytorch#distributed training#vllm#grpo#torchtitan
One-click install
npx skills add https://github.com/Supporter09/Face_Anti_Spoofing_Biometric --skill torchforge-rl-training-supporter09
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: torchforge-rl-training
Source: https://github.com/Supporter09/Face_Anti_Spoofing_Biometric/tree/main/.claude/skills/torchforge
Command: npx skills add https://github.com/Supporter09/Face_Anti_Spoofing_Biometric --skill torchforge-rl-training-supporter09

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps you design and run PyTorch-native agentic reinforcement learning workflows without mixing algorithm logic with distributed infrastructure concerns.

Core Features & Use Cases

  • Algorithm experimentation: Build or adapt RL objectives such as GRPO, DAPO, SAPO, and REINFORCE-style losses while keeping infrastructure separate.
  • Scalable training: Coordinate trainer, generator, and reference-model services across single-GPU, multi-GPU, or multi-node setups.
  • Practical workflows: Use it for SFT pretraining, GRPO reasoning runs, custom reward functions, and distributed model synchronization.
  • Use case: A research team can launch a math-reasoning GRPO experiment, monitor KL stability, and iterate on reward shaping without rewriting the serving stack.

Quick Start

Ask the Skill to help you set up a torchforge GRPO training workflow for your model, dataset, and GPU budget.

Frequently Asked Questions about torchforge-rl-training

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

FAQPage Schema
How do I set up distributed GRPO training for LLMs using PyTorch?▼

You can run PyTorch-native agentic reinforcement learning workflows by using this Skill to coordinate trainer, generator, and reference-model services across multi-GPU or multi-node setups with asynchronous weight synchronization.

Can I use vLLM for inference during reinforcement learning experimentation?▼

Yes, vLLM inference is supported natively. The Skill provides asynchronous service abstractions that integrate vLLM to handle generation workloads while keeping the distributed training infrastructure separate from your RL algorithm logic.

How does Monarch-based coordination work for multi-GPU RL training?▼

Monarch-based coordination manages scalable training by providing asynchronous service abstractions and configurable resource allocation, allowing you to scale PyTorch RL research across single-GPU, multi-GPU, or multi-node environments without rewriting infrastructure code.

Does TorchTitan integration support custom reward functions for GRPO?▼

Yes, TorchTitan integration supports custom reward functions. You can iterate on reward shaping and monitor KL stability for GRPO reasoning runs without mixing your custom loss design with the underlying distributed serving stack.

What's the best way to decouple RL algorithm logic from distributed infrastructure in PyTorch?▼

The best way to decouple them is using clean service abstractions for trainer, generator, and reference-model components, enabling you to build RL objectives like DAPO and SAPO while the system handles distributed model synchronization.

When do I need asynchronous weight synchronization for LLM agents?▼

You need asynchronous weight synchronization when running scalable distributed training workflows across multiple GPUs or nodes, ensuring that trainer and generator services stay coordinated without blocking your reinforcement learning experimentation.

Related Skills

Explore More Software Engineering Skills→
Clay-HHKClay-HHK
2

torchforge-rl-training

Orchestrate PyTorch-native distributed RL training and vLLM inference for agentic models.

Community
Advanced
JKhyroJKhyro

grpo-rl-training

Build GRPO reinforcement-learning training workflows for language models.

Community
Advanced
jackspacejackspace
16

pufferlib

Optimize high-throughput PPO+LSTM reinforcement learning training with vectorized environments.

Community
Advanced
Tianyi-Billy-MaTianyi-Billy-Ma

pytorch-lightning

Automate PyTorch training workflows with LightningModule, Trainer, and LightningDataModule.

Community
Advanced