skills.rest
brittaniebuffiecsu
brittaniebuffiecsuCOMMUNITY
Explore all skills by this author →
GitHub

slime-rl-training

Train large language models with reinforcement learning using Megatron-LM and SGLang.

1|Updated Feb 17, 2026
Search Tags:#reinforcement learning#LLM training#model scaling#large language models#SGLang#Megatron-LM
One-click install
npx skills add https://github.com/brittaniebuffiecsu/zerogravityclaw --skill slime-rl-training-brittaniebuffiecsu
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: slime-rl-training
Source: https://github.com/brittaniebuffiecsu/zerogravityclaw/tree/main/src/hermes-core/optional-skills/mlops/slime
Command: npx skills add https://github.com/brittaniebuffiecsu/zerogravityclaw --skill slime-rl-training-brittaniebuffiecsu

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires sglang-router>=0.2.3, ray, torch>=2.0.0, transformers>=4.40.0, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill provides a comprehensive framework for large language model (LLM) post-training with reinforcement learning (RL) using the slime tool, addressing challenges such as RL scaling and custom data generation workflows.

Core Features & Use Cases

  • Megatron-LM Integration: Offers seamless integration with Megatron-LM for training.
  • Custom Data Generation: Supports custom data generation workflows for flexible prompt management and sample storage.
  • High-Throughput Rollout Generation: Leverages SGLang for high-throughput rollout generation.
  • Model Support: Supports training of GLM, Qwen3, DeepSeek V3/R1, and Llama 3 models.
  • Use Case: Ideal for research-grade framework users looking to implement RL scaling for large models with tight integration of Megatron-LM and SGLang.

Quick Start

Train a GLM model using slime by executing the following command in your terminal:

python train.py \
  --actor-num-nodes 1 \
  --actor-num-gpus-per-node 4 \
  --rollout-num-gpus 4 \
  --advantage-estimator grpo \
  --use-kl-loss --kl-loss-coef 0.001 \
  --rollout-batch-size 32 \
  --n-samples-per-prompt 8 \
  --global-batch-size 256 \
  --num-rollout 3000 \
  --prompt-data /path/to/data.jsonl

Frequently Asked Questions about slime-rl-training

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

FAQPage Schema
How do I scale reinforcement learning training for large language models?▼

Scale reinforcement learning training for large language models using the slime framework, which provides tight integration with Megatron-LM and SGLang to handle high-throughput rollout generation and efficient model scaling.

Can I train Llama 3 and DeepSeek models with reinforcement learning using SGLang?▼

Yes, you can train Llama 3 and DeepSeek V3/R1 models with reinforcement learning using SGLang. The framework supports these alongside GLM and Qwen3 for high-throughput rollout generation.

How do I generate custom data for reinforcement learning rollout workflows?▼

Generate custom data for reinforcement learning rollout workflows using the framework's custom data generation features, which support flexible prompt management and sample storage for training large language models.

Does this reinforcement learning framework require Megatron-LM and SGLang integration?▼

Yes, the framework relies on tight integration with Megatron-LM for training and SGLang for high-throughput rollout generation to effectively scale reinforcement learning for large language models.

What is the best way to configure GRPO advantage estimation for LLM training?▼

Configure GRPO advantage estimation for LLM training by passing the advantage estimator argument in the training command, alongside parameters for KL loss coefficients and rollout batch sizes to optimize the reinforcement learning process.

What dependencies are needed to run reinforcement learning training for large language models?▼

Dependencies required to run reinforcement learning training include sglang-router, ray, torch, and transformers, ensuring the environment supports large-scale model scaling and rollout generation.

Related Skills

Explore More Software Engineering Skills→
ricablericable

fine-tuning-with-trl

Fine-tune large language models with TRL for supervised and preference alignment.

Community
Advanced
ragnarokhaaragnarokhaa
4

slime-rl-training

Integrate Megatron-LM and SGLang for LLM post-training reinforcement learning.

Community
Advanced
InverterNetworkInverterNetwork

slime-rl-training

Orchestrate Megatron-LM training and SGLang rollout generation for large-scale RL post-training.

Official
Advanced
hhhi21ghhhi21g

slime-rl-training

Scale RL post-training for LLMs with Megatron-LM and SGLang.

Community
Advanced