verl-e2e-testing

Validate Megatron-Bridge provider changes with an external verl training loop smoke.

Updated May 23, 2026
One-click install
npx skills add https://github.com/yo-steven/skills-exploration-20260522 --skill verl-e2e-testing
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: verl-e2e-testing
Source: https://github.com/yo-steven/skills-exploration-20260522/tree/main/skills/Megatron-Bridge/verl-e2e-testing
Command: npx skills add https://github.com/yo-steven/skills-exploration-20260522 --skill verl-e2e-testing

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It reduces the risk of breaking Megatron-Bridge model/provider integrations by verifying the provider end-to-end inside verl, including the external RL/training-loop wiring that Bridge-only tests may miss.

Core Features & Use Cases

  • Runs a minimal verl Megatron PPO e2e: Executes a small Megatron backend job from a Bridge checkout to confirm the integration works end-to-end.
  • Covers key compatibility surfaces: Validates provider configuration, HF import via Bridge, PEFT/LoRA wrapping, Megatron DDP wrapping, optimizer setup, and rollout/ref/critic wiring.
  • Supports staged coverage: Moves from Level 0 smoke to optional save/resume, parallelism stress, optional Megatron-FSDP, architecture-specific checks, or a learning-signal run.

Quick Start

Start a Level 0 PPO e2e by exporting USE_MBRIDGE=True, VANILLA_MBRIDGE=False, VALUE_VANILLA_MBRIDGE=False, LORA_RANK=4, TOTAL_TRAIN_STEPS=1, and then running the verl wrapper script tests/special_e2e/run_ppo_trainer_megatron.sh with ENGINE=vllm.

Frequently Asked Questions about verl-e2e-testing

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

FAQPage Schema
How do I validate Megatron-Bridge provider changes in verl?▼

You validate Megatron-Bridge provider changes by running a verl PPO end-to-end smoke test from your local Bridge checkout, capturing logs to confirm the integration works before delivering.

What is covered in a verl e2e testing smoke run for Megatron?▼

A verl e2e testing smoke run covers provider configuration, HF import via Bridge, PEFT/LoRA wrapping, Megatron DDP wrapping, optimizer setup, and rollout/ref/critic wiring.

How do I start a Level 0 PPO e2e test for Megatron-Bridge?▼

To start a Level 0 PPO e2e test, export USE_MBRIDGE=True, set VANILLA_MBRIDGE=False, configure LORA_RANK=4 and TOTAL_TRAIN_STEPS=1, then run the verl wrapper script with ENGINE=vllm.

Does verl e2e testing support save/resume and parallelism stress for Bridge providers?▼

Yes, verl e2e testing supports staged coverage that moves from Level 0 smoke to optional save/resume, parallelism stress, and Megatron-FSDP variants during compatibility checks.

Do I need to use PYTHONPATH to run verl Megatron compatibility checks?▼

Yes, you must set PYTHONPATH to use your local Bridge checkout when running verl Megatron compatibility checks to ensure the external training loop wiring validates correctly.

Why does my Bridge-only test miss integration issues that verl e2e testing catches?▼

Bridge-only tests miss integration issues because they skip the external RL/training-loop wiring, whereas verl e2e testing verifies the provider end-to-end inside the actual training loop.