openrlhf-training

openrlhf-training is a skill for Claude Code, Codex from davila7/claude-code-templates. It costs 72 tokens per session (2,537 once invoked), scanned B, original, MIT.

A distributed framework for reinforcement learning from human feedback (RLHF), where a language model is improved using ratings or reward signals. It combines Ray for coordinating jobs, vLLM for generating answers, and ZeRO-3 for splitting model work across GPUs.

In plain words
What is it for?
Use it to train 7B–70B+ language models with PPO, GRPO, RLOO, or DPO across multiple GPUs.
Why use it?
It handles the coordination and memory demands of training large models, which are difficult to manage on a single machine.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one.

Needs its repository: it runs a file that does not travel with it, so clone the repository first. The line is --save_path ./output/llama3-8b-rlhf \.

About the project

Claude Code Templates is a command-line tool and catalogue for configuring Anthropic’s Claude Code with agents, commands, settings, hooks, integrations, skills, and project templates. Developers use it to browse and install reusable components for their coding workflows. The catalogue includes many of these Claude Code components.

davila7/claude-code-templates · 30,545 stars · on GitHub · aitmpl.com

Install

Getting it into your agent

It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.

Clone the repo
git clone --depth 1 https://github.com/davila7/claude-code-templates
agentmods
npx agentmods add skills/davila7/claude-code-templates/post-training-openrlhf

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for openrlhf-training

README.md
[![agentmods](https://agentmods.dev/badge/skills/davila7/claude-code-templates/post-training-openrlhf.svg)](https://agentmods.dev/skills/davila7/claude-code-templates/post-training-openrlhf)
Your own site
<a href="https://agentmods.dev/skills/davila7/claude-code-templates/post-training-openrlhf"><img src="https://agentmods.dev/badge/skills/davila7/claude-code-templates/post-training-openrlhf.svg" alt="Measured on agentmods" height="20"></a>
Per session 72 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,537 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00072 $0.02537
Opus 5 $0.00036 $0.01269
Sonnet 5 $0.00014 $0.00507
Haiku 4.5 $0.00007 $0.00254

Measured 3d ago against content hash eb9da6a1a81c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade B, and why

openrlhf-training scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Asks for rootmediumPrivilege escalation

A mod that escalates privileges can change anything on the machine, not only the project.

sudo pip uninstall xgboost transformer_engine flash_attn pynvml -y
Origin

Copies of this mod

3 near-identical copies found in the catalogue:

cli-tool/components/skills/ai-research/post-training-openrlhf/SKILL.md · 250 lines

How it starts

The opening of the file, as written. The whole thing — 250 lines — stays where its author put it; the contents beside it link to each section on GitHub.

OpenRLHF - High-Performance RLHF Training

Quick start

OpenRLHF is a Ray-based RLHF framework optimized for distributed training with vLLM inference acceleration.

Installation:

# Launch Docker container
docker run --runtime=nvidia -it --rm --shm-size="10g" --cap-add=SYS_ADMIN \
  -v $PWD:/openrlhf nvcr.io/nvidia/pytorch:25.02-py3 bash

# Uninstall conflicts
sudo pip uninstall xgboost transformer_engine flash_attn pynvml -y

# Install OpenRLHF with vLLM
pip install openrlhf[vllm]

PPO Training (Hybrid Engine):

ray start --head --node-ip-address 0.0.0.0 --num-gpus 8

ray job submit --address="http://127.0.0.1:8265" \
  --runtime-env-json='{"working_dir": "/openrlhf"}' \
  -- python3 -m openrlhf.cli.train_ppo_ray \
  --ref_num_nodes 1 --ref_num_gpus_per_node 8 \
  --reward_num_nodes 1 --reward_num_gpus_per_node 8 \
  --critic_num_nodes 1 --critic_num_gpus_per_node 8 \
  --actor_num_nodes 1 --actor_num_gpus_per_node 8 \
  --vllm_num_engines 4 --vllm_tensor_parallel_size 2 \
  --colocate_all_models \
  --vllm_gpu_memory_utilization 0.5 \
  --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
  --reward_pretrain OpenRLHF/Llama-3-8b-rm-700k \
  --save_path ./output/llama3-8b-rlhf \
  --micro_train_batch_size 8 --train_batch_size 128 \
  --micro_rollout_batch_size 16 --rollout_batch_size 1024 \
  --max_epochs 1 --prompt_max_len 1024 --generate_max_len 1024 \
  --zero_stage 3 --bf16 \
  --actor_learning_rate 5e-7 --critic_learning_rate 9e-6 \
  --init_kl_coef 0.01 --normalize_reward \
  --gradient_checkpointing --packing_samples \
  --vllm_enable_sleep --deepspeed_enable_sleep

GRPO Training (Group Normalized Policy Optimization):

# Same command as PPO, but add:
--advantage_estimator group_norm

Common workflows

Workflow 1: Full RLHF pipeline (SFT → Reward Model → PPO)

Step 1: Train reward model (DPO):

deepspeed --module openrlhf.cli.train_rm \
  --save_path ./output/llama3-8b-rm \
  --save_steps -1 --logging_steps 1 \
  --eval_steps -1 --train_batch_size 256 \
  --micro_train_batch_size 1 --pretrain meta-llama/Meta-Llama-3-8B \
  --bf16 --max_epochs 1 --max_len 8192 \
  --zero_stage 3 --learning_rate 9e-6 \
  --dataset OpenRLHF/preference_dataset_mixture2_and_safe_pku \
  --apply_chat_template --chosen_key chosen \
  --rejected_key rejected --flash_attn --gradient_checkpointing

Read the full file on GitHub · 250 lines

Files

What ships with it

4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 250 lines · 72 tokens per session scan B eb9da6a1a81c

Subscribe to this mod's changes

openrlhf-training is a skill published in the GitHub repository davila7/claude-code-templates (30,545 stars, last pushed today), licensed MIT. It adds 72 tokens to every session and 2,537 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it B with 1 finding (asks for root). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.

Related

Other skills, from other repositories

openrlhf-training

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

OpenLAIR/dr-claw · 72 tokens

openrlhf-training

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

Orchestra-Research/AI-Research-SKILLs · 72 tokens

openrlhf-training

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

liortesta/ClawdAgent · 72 tokens

openrlhf-training

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

ihatesea69/HieuNghi-AI-Skills · 72 tokens

grpo-rl-training

Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training.

OpenLAIR/dr-claw · 26 tokens

fine-tuning-with-trl

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

OpenLAIR/dr-claw · 69 tokens