skills.rest
handsomelong922
handsomelong922COMMUNITY
Explore all skills by this author →
GitHub

miles-rl-training

Automate enterprise RL training workflows for large MoE models.

Updated Apr 3, 2026
Search Tags:#fp8#rl-training#megatron-lm#moe#int4#speculative-rl#train-inference
One-click install
npx skills add https://github.com/handsomelong922/my-codex-skills --skill miles-rl-training-handsomelong922
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: miles-rl-training
Source: https://github.com/handsomelong922/my-codex-skills/tree/main/skills/miles
Command: npx skills add https://github.com/handsomelong922/my-codex-skills --skill miles-rl-training-handsomelong922

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Enterprise-grade RL training for large MoE models is complex and error-prone; miles provides a production-ready framework and guidance to streamline stability, precision, deployment, and workflow orchestration.

Core Features & Use Cases

  • Low-precision training with FP8 and INT4 QAT for large MoE models.
  • Train-inference alignment and Rollout Routing Replay (R3) to ensure consistency between training and inference.
  • Speculative RL workflows to maximize throughput on Megatron-SGLang pipelines.
  • Production-ready configuration surface that inherits slime API and supports data-center deployment scenarios.

Quick Start

Run miles with your target model checkpoint and a rollout plan to begin enterprise RL training.

Frequently Asked Questions about miles-rl-training

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

FAQPage Schema
How do I set up RL training for large MoE models?▼

Set up RL training for large MoE models by automating workflow orchestration with miles, which coordinates Megatron, SGLang, and online SFT workflows. Run it with your target model checkpoint and rollout plan to begin.

Does Megatron-LM work with FP8 and INT4 precision training?▼

Yes, Megatron-LM supports FP8 and INT4 QAT precision training through miles. The framework automates enterprise-grade configuration to streamline stability and precision across large MoE deployments.

How do I ensure train-inference alignment during RL training?▼

Ensure train-inference alignment during RL training by applying Rollout Routing Replay (R3). This mechanism guarantees consistency between training and inference within the miles framework.

What is speculative RL and how does it maximize throughput?▼

Speculative RL is a workflow approach that maximizes throughput on Megatron-SGLang pipelines. It is supported by miles to optimize data-center-scale deployments for large MoE models.

Can I use slime API for data-center scale deployment configurations?▼

Yes, you can use the slime API for data-center scale deployment configurations. Miles provides a production-ready configuration surface that directly inherits the slime API.

What are the limitations of low-precision RL training for MoE models?▼

Low-precision RL training for MoE models is complex and error-prone, requiring careful stability and precision management. Miles streamlines these constraints by providing a production-ready framework for FP8 and INT4 deployments.

Related Skills

Explore More Software Engineering Skills→
qa-amanqa-aman
19

go-to-market-checklist

Generate a pre-launch readiness checklist from a feature spec and target launch date.

Community
Advanced
Jonah-1Jonah-1

eval

Run inference on neural network checkpoints and generate PSNR, SSIM, and LPIPS metrics.

Community
Intermediate
GregsGreyCodeGregsGreyCode
6

slime-rl-training

Coordinate end-to-end RL training for LLMs with Megatron and SGLang.

Community
Advanced
OpenLAIROpenLAIR
1.0k

miles-rl-training

Configure miles for large MoE RL training with FP8/INT4 and train-inference alignment.

Official
Advanced