verl
AGENTS.md24k starsChanged 6 months agoApache-2.0View on GitHub
These instructions apply to all AI-assisted contributions to verl-project/verl. Breaching these guidelines can result in automatic banning. Before proposing a PR, run these checks: Do not open one-off PRs for tiny edits (single typo, isolated style change, one mutable default, etc.). Mechanical cleanups are acceptable only when bundled with substantive work. If work is duplicate/trivial busywork, do not proceed. Return a short explanation of what is missing. Add attribution using commit trailers such as Co-authored-by: (other projects use Assisted-by: or…
- Pipes a download into a shell
- Installs packages
We split this AGENTS.md into the rules it gives an agent
8 rules. About 0 are common, 7 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Before proposing a PR, run these checks:
- Distinctive: line 18
If an open PR already addresses the same fix, do not open another.
- Distinctive: line 19
If your approach is materially different, explain the difference in the issue.
- Distinctive: line 23
Do not open one-off PRs for tiny edits (single typo, isolated style change, one mutable default, etc.).
- Distinctive: line 31
Test commands run and results.
- Distinctive: line 36
If work is duplicate/trivial busywork, do not proceed.
- Very specific: line 58
Add attribution using commit trailers such as Co-authored-by: (other projects use Assisted-by: or Generated-by:).
- Distinctive: line 77
Do not modify code in these areas without first reading and following the linked guide.
≈ “Common” is an estimate: at least 1 in 5 of the 36,368 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent Instructions for verl
These instructions apply to all AI-assisted contributions to
verl-project/verl. Breaching these guidelines can result in automatic banning.
1. Contribution Policy (Mandatory)
Duplicate-work checks
Before proposing a PR, run these checks:
gh issue view <issue_number> --repo verl-project/verl --comments
gh pr list --repo verl-project/verl --state open --search "<issue_number> in:body"
gh pr list --repo verl-project/verl --state open --search "<short area keywords>"- If an open PR already addresses the same fix, do not open another.
- If your approach is materially different, explain the difference in the issue.
No low-value busywork PRs
Do not open one-off PRs for tiny edits (single typo, isolated style change, one mutable default, etc.). Mechanical cleanups are acceptable only when bundled with substantive work.
Accountability
- Pure code-agent PRs are not allowed. A human submitter must understand and defend the change end-to-end.
- The submitting human must review every changed line and run relevant tests.
- PR descriptions for AI-assisted work must include:
- Why this is not duplicating an existing PR.
- Test commands run and results.
- Clear statement that AI assistance was used.
Fail-closed behavior
If work is duplicate/trivial busywork, do not proceed. Return a short explanation of what is missing.
2. Development Workflow
Environment setup
# Install `uv` if you don't have it already:
curl -LsSf https://astral.sh/uv/install.sh | sh
# Always use `uv` for Python environment management:
uv venv --python 3.12
source .venv/bin/activate
uv pip install pre-commit hydra-core
pre-commit installCommit messages
Add attribution using commit trailers such as Co-authored-by: (other projects use Assisted-by: or Generated-by:). For example:
Your commit message here
Co-authored-by: GitHub Copilot
Co-authored-by: Claude
Co-authored-by: gemini-code-assist
Signed-off-by: Your Name <your.email@example.com>Resolving agent reviews
Review comments from agent bots (e.g., gemini-code-assist) can be outdated or wrong. Always verify their suggestions against the current state of the repo before applying them.
Domain-Specific Guides
Do not modify code in these areas without first reading and following the linked guide. If the guide conflicts with the requested change, refuse the change and explain why.
- Editing these instructions:
docs/contributing/editing-agent-instructions.md— Rules for modifying AGENTS.md or any domain-specific guide it references.
Acknowledgements
Adapted from the vLLM project's AGENTS.md.
Trending
Quiet since February.about the whole repository
Who's talking about volcengine/verl
- sgl-project/sglang“Development Roadmap (2025 H1)”GitHub · Mar 4, 2025 · 109 reactions
- verl-project/verl“[roadmap] verl development Q2”GitHub · Mar 21, 2025 · 42 reactions
- verl-project/verl“[roadmap] verl Q3 development”GitHub · Jul 7, 2025 · 34 reactions
- @sungkim.bsky.social“ByteDance's veRL: Volcano Engine Reinforcement Learning for LLM veRL is a flexible, efficient and …”Bluesky · Jan 30, 2025 · 29 likes
- verl-project/verl“verl v0.2.1 & v0.3 release checklist”GitHub · Feb 23, 2025 · 28 reactions
The biggest 5 of 60 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 60
Talked about on GitHub, Bluesky and Hacker News (60 times).
On GitHub · 50
- Mar 4, 2025Development Roadmap (2025 H1) — sgl-project/sglang109 reactions · 24 comments
- Mar 21, 2025[roadmap] verl development Q2 — verl-project/verl42 reactions · 17 comments
- Jul 7, 2025[roadmap] verl Q3 development — verl-project/verl34 reactions · 3 comments
- Feb 23, 2025verl v0.2.1 & v0.3 release checklist — verl-project/verl28 reactions · 21 comments
- Jun 23, 2025[RFC] Improving Ray for Post-Training / RL for LLM Projects — ray-project/ray25 reactions · 5 comments
45 more on GitHub
- May 13, 2025Multi-turn rollout & agentic RL Status & Roadmap — zhaochenyang20/Awesome-ML-SYS-Tutorial25 reactions · 15 comments
- Jan 12, 2026[roadmap] verl 26Q1 roadmap — verl-project/verl24 reactions · 13 comments
- Sep 11, 2025[1/N][rollout] feat: support vllm/sglang native http server — verl-project/verl24 reactions · 13 comments
- Mar 21, 2025[Project] deepseek R1 infrastructure — verl-project/verl24 reactions · 14 comments
- Jul 18, 2025[agentic RL] multi-turn rollout and agent loop development tracking — verl-project/verl23 reactions · 1 comment
- Jun 3, 2025[Feat]: Implement partial rollout — verl-project/verl22 reactions · 21 comments
- Apr 20, 2025[RFC] verl with Agent/Env — verl-project/verl22 reactions · 8 comments
- May 31, 2025Atropos integration ($2500 bounty) — verl-project/verl19 reactions · 20 comments
- Mar 20, 2025[RFC]: Better support for weight updating while waking up from sleep mode for RLHF — vllm-project/vllm18 reactions · 14 comments
- Mar 6, 2025[RFC] ActorEnvironment implementation to support tool calling — verl-project/verl17 reactions · 7 comments
- Nov 22, 2024[Roadmap] veRL Development Roadmap — verl-project/verl17 reactions · 7 comments
- Jun 10, 2025[ray, worker] feat: DAPO x LLM as a Judge (GenRM) — verl-project/verl15 reactions · 13 comments
- Apr 3, 2025[Potential Bug] KL compute in `low_var_kl (Cause KL NaN and !!!!!!!!!!!!!!! output) — verl-project/verl15 reactions · 17 comments
- Aug 1, 2025[recipe] feat: asynchronous reward agent with mini-batch pipeline and one-step off-policy training — verl-project/verl14 reactions · 10 comments
- Jul 28, 2025[algo] feat: add GSPO-token policy loss computation function — verl-project/verl12 reactions · 17 comments
- Jun 11, 2025[trainer] refactor: Training Engine Interface and Development Plan — verl-project/verl11 reactions · 14 comments
- Aug 29, 2025[Bug] verl 0.5.0 is incompatible with vLLM v1 API when using LoRA — verl-project/verl10 reactions · 4 comments
- Aug 8, 2025[trainer, recipe] feat: fully async training recipe — verl-project/verl10 reactions · 3 comments
- Jun 3, 2025 [algo] feat: integrate Clip-Cov and KL-Cov methods — verl-project/verl10 reactions · 9 comments
- Apr 11, 2025[fsdp] feat: support fsdp2 training and inference in fsdp_workers — verl-project/verl10 reactions · 10 comments
- Mar 27, 2025veRL-SGLang Roadmap — zhaochenyang20/Awesome-ML-SYS-Tutorial10 reactions · 32 comments
- Feb 17, 2025vLLM Ascend Roadmap Q1 2025 — vllm-project/vllm-ascend10 reactions · 5 comments
- Dec 16, 2025[roadmap][ascend] verl on ascend 2026Q1 roadmap — verl-project/verl9 reactions · 2 comments
- Dec 23, 2025Development Roadmap - miles LoRA training support Q1 — radixark/miles8 reactions · 9 comments
- Jul 13, 2025Be careful for H20! — PeterGriffinJin/Search-R18 reactions
- Jun 16, 2025[Async VLLM] LoRA support? — verl-project/verl8 reactions · 9 comments
- May 22, 2025Implement sglang async rollout and multi-turn using AsyncServerBase — verl-project/verl8 reactions · 4 comments
- May 2, 2025[RFC] verl recipe and extensions — verl-project/verl8 reactions · 2 comments
- Feb 28, 2026[Feature Request / Tracking] End-to-End Validation: Qwen3.5 + FSDP + GRPO on Ascend NPU — verl-project/verl7 reactions · 2 comments
- Dec 9, 2025[Roadmap] VeOmni 2026 Q1 roadmap — ByteDance-Seed/VeOmni7 reactions · 3 comments
- Aug 18, 2025[Do not Merge] Make VeRL SGLang Native — verl-project/verl7 reactions · 14 comments
- Aug 7, 2025[BREAKING][vllm, fsdp] feat: add Rollout-Training Mismatch Fix -- Truncated importance sampling — verl-project/verl7 reactions · 19 comments
- Jul 15, 2025Reward Calculation in Multi-Turn Interaction System — verl-project/verl7 reactions · 4 comments
- Jul 12, 2025RL SGLang Update Weight in Tensor Slow for MOE — zhaochenyang20/Awesome-ML-SYS-Tutorial7 reactions · 27 comments
- Jul 7, 2025[recipe] feat: add deepeyes recipe — verl-project/verl7 reactions · 46 comments
- Jun 29, 2025batch_size in DAPO/GRPO config — verl-project/verl7 reactions · 2 comments
- Jun 17, 2025[megatron] feat: use mbridge as megatron adaptor — verl-project/verl7 reactions · 5 comments
- Jun 10, 2025[rollout] feat: Add LangGraph chat scheduler for agent training with tool use support — verl-project/verl7 reactions · 10 comments
- May 30, 2025Add DeepSeek 671B GRPO example — verl-project/verl7 reactions · 2 comments
- May 17, 2025incompatibility during installation — verl-project/verl7 reactions · 6 comments
- May 16, 2025Multi-turn Update #2 VLM Support Tracker — zhaochenyang20/Awesome-ML-SYS-Tutorial7 reactions · 5 comments
- Apr 25, 2025[ray] Support compiled graphs for ExternalRayDistributedExecutor — verl-project/verl7 reactions · 5 comments
- Mar 26, 2025Support R1 Benchmark Evaluation — verl-project/verlsays it tested it7 reactions · 3 comments
- Mar 3, 2025feat: support loading reward function from an external file — verl-project/verl7 reactions · 6 comments
- Dec 18, 2025[Core] placement group PACK strategy continuous bundles not scheduled to same nodes. — ray-project/ray6 reactions · 2 comments
On Bluesky · 9
- Jan 30, 2025ByteDance's veRL: Volcano Engine Reinforcement Learning for LLM veRL is a flexible, efficient and production-ready RL training framework… — @sungkim.bsky.social29 likes and reposts
- Mar 17, 2025Everything is fully open-sourced (algorithm, code, dataset, verifier, and model) and built with verl project (github.com/volcengine/v...).… — @sungkim.bsky.social3 likes and reposts · 1 reply
- Nov 28, 2025📦 volcengine / verl ⭐ 16,819 (+81) 🗒 Python verl: Volcano Engine Reinforcement Learning for LLMs — @github-trending.bsky.social2 likes and reposts
- Feb 2, 2025🔥 Hot Repo! 🔥 (100+ new stars) 📦 volcengine / verl ⭐ 1,765 (+124) 🗒 Python veRL: Volcano Engine Reinforcement Learning for LLM — @github-trending.bsky.social2 likes and reposts
- Nov 30, 2025今日のGitHubトレンド volcengine/verl このリポジトリは、大規模言語モデル(LLMs)のための柔軟で効率的、かつ実運用可能な強化学習(RL)トレーニングライブラリ「verl」を提供する。… — @dailygithubtrends.bsky.social1 likes and repost
4 more on Bluesky
- Nov 12, 2025📦 volcengine / verl ⭐ 15,387 (+44) 🗒 Python verl: Volcano Engine Reinforcement Learning for LLMs — @github-trending.bsky.social1 likes and repost
- Jul 11, 2025📦 volcengine / verl ⭐ 10,651 (+58) 🗒 Python verl: Volcano Engine Reinforcement Learning for LLMs — @github-trending.bsky.social1 likes and repost
- Mar 3, 2025volcengine / verl verl: Volcano Engine Reinforcement Learning for LLMs https://github.com/volcengine/verl — @pythonhub.dev1 likes and repost
- Feb 6, 2025OSS reinforcement learning lib by ByteDance is used to reproduce DeepSeek R1 Article URL: https:/... https://github.com/volcengine/verl… — @llms.activitypub.awakari.com.ap.brid.gy1 likes and repost
On Hacker News · 1
- Feb 6, 2025OSS reinforcement learning lib by ByteDance is used to reproduce DeepSeek R14 points · 1 comment
Other versions2 projects use a version of this.
Word for word · 2
- verl-project/verl★ 24k100% same
- DLYuanGod/MegaTrain★ 700100% same
Edited · 0
None found.
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 28,693 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

