skills.rest
Codekiing
CodekiingCOMMUNITY
Explore all skills by this author →
GitHub

verl-agent-training

Automate VERL two-GPU PPO training with dataset conversion and repair workflows.

Updated Aug 23, 2026
Search Tags:#agent training#qwen#huggingface#verl#dataset conversion#ppo training#gpu automation
One-click install
npx skills add https://github.com/Codekiing/Auto_VeRL_PPO_Skill --skill verl-agent-training
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: verl-agent-training
Source: https://github.com/Codekiing/Auto_VeRL_PPO_Skill/tree/main
Command: npx skills add https://github.com/Codekiing/Auto_VeRL_PPO_Skill --skill verl-agent-training

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It eliminates the manual, multi-step setup required to train a domain-specific PPO agent by automating environment preparation, dataset acquisition/conversion, and the VERL PPO training run with guardrails.

Core Features & Use Cases

  • End-to-end VERL PPO training pipeline: Clones VERL, installs a fixed compatible dependency set, prepares dataset parquet files, then launches python3 -m verl.trainer.main_ppo.
  • Interactive dataset selection by agent type: Uses your provided AGENT_TYPE to search and choose an appropriate Hugging Face dataset, then converts it into VERL-ready train.parquet and test.parquet.
  • Crash recovery via “immediate repair” rules: Provides deterministic fallback steps for common failures, including OOM mitigation by reducing response length and micro-batch sizes.

Quick Start

Run this Skill in a blank directory by first saving the provided script as run_verl_skill.sh, then executing chmod +x run_verl_skill.sh && bash run_verl_skill.sh after setting AGENT_TYPE to your desired domain (e.g., math).

Frequently Asked Questions about verl-agent-training

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

FAQPage Schema
How do I automate PPO training for a Qwen2.5 agent using Hugging Face datasets?▼

Yes, VERL PPO training requires strict usage through the python3 -m verl.trainer.main_ppo command. The pipeline clones VERL, installs compatible dependencies, and generates train/test parquet files to launch the training run.

How do I convert Hugging Face datasets to parquet for VERL PPO training?▼

The pipeline automatically searches Hugging Face datasets using your specified AGENT_TYPE domain variable, then selects an appropriate dataset and converts it into VERL-ready train.parquet and test.parquet files.

Do I need two GPUs to run VERL PPO training for a domain agent?▼

This pipeline automates two-GPU VERL PPO training for a domain-specific agent by orchestrating environment setup, dataset conversion, and PPO execution using a Qwen2.5-0.5B-based reward model policy.

How do I fix OOM errors during VERL PPO agent training?▼

The pipeline enforces an immediate repair workflow for OOM failures by automatically reducing response length and micro-batch sizes to mitigate memory constraints and resume PPO training.

Can I train a domain-specific PPO agent end-to-end with a single command?▼

You can train a domain PPO agent end-to-end with one command by executing the provided bash script after setting your desired AGENT_TYPE domain variable, which automates the entire VERL training pipeline.

Related Skills

Explore More Software Engineering Skills→
tankygranny05tankygranny05
2

clone-worktree-port-bump

Clone a repository into ~/temp and bump *_PORT values in .env by 100 * N.

Community
Advanced
jacardljacardl

grpo-rl-training

Implement GRPO-based RL fine-tuning for language models with TRL.

Community
Advanced
yanochka11yanochka11

huggingface-datasets

Query Hugging Face dataset metadata, rows, statistics, and parquet resources.

Community
Intermediate
Clay-HHKClay-HHK
2

verl-rl-training

Train large language models with reinforcement learning using verl and Ray.

Community
Advanced