[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7972Y5JB29RKJSBC0G091MF":3,"skill-same:01M7972Y5JB29RKJSBC0G091MF":67},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":27,"commitSha":28,"skillDirectory":29,"forksCount":30,"repositoryHtmlUrl":31,"repositoryHomepageUrl":6,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M7972Y5JB29RKJSBC0G091MF","verl",null,"Use verl for LLM post-training workflows: setup, data and rewards, PPO\u002FGRPO\u002FSFT configs, rollout tools, checkpoints, profiling, and repository maintenance.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fverl\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: verl\ndescription: \"Use verl for LLM post-training workflows: setup, data and rewards,\n  PPO\u002FGRPO\u002FSFT configs, rollout tools, checkpoints, profiling, and repository\n  maintenance.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# verl\n\nUse this repo skill when a user asks how to install, configure, run, debug, or maintain `verl`, the Volcano Engine Reinforcement Learning library for LLM post-training.\n\n## Start Here\n\n- Read `references\u002Frepo-provenance.md` before relying on this skill for a different checkout or after package metadata, config files, examples, or docs change.\n- Read `references\u002Ftroubleshooting.md` for cross-cutting failures that span setup, Hydra configs, data schema, rollout backends, checkpoint operations, and repo maintenance.\n- Run `scripts\u002Fcheck_skill_integrity.py` when editing or importing this skill to check frontmatter, links, private-path leaks, and generated cache files.\n\n## Route By User Task\n\n- **Install or verify an environment**: use `sub-skills\u002Fsetup-and-backends\u002F` for base install choices, Docker-vs-custom setup, Python\u002FCUDA requirements, optional extras, vLLM\u002FSGLang\u002FMegatron\u002FTensorRT-LLM\u002FNPU\u002FROCm caveats, and safe import\u002Fbackend checks.\n- **Prepare or validate data and rewards**: use `sub-skills\u002Fdata-and-rewards\u002F` for parquet row schemas, chat prompt fields, `reward_model.ground_truth`, custom reward modules, and the bundled parquet validator.\n- **Build training commands and Hydra overrides**: use `sub-skills\u002Ftraining-and-configs\u002F` for PPO, GRPO, SFT, on-policy distillation, backend strategy switches, batch\u002Fmicro-batch sizing, generated config references, and the dry-run PPO command builder.\n- **Configure rollout, agent loops, and tools**: use `sub-skills\u002Frollout-and-tools\u002F` for rollout engines, async generation server flows, multi-turn agent loops, `@function_tool`, native `BaseTool`, tokenization sanity checks, traces, and tool schema validation.\n- **Operate checkpoints and model artifacts**: use `sub-skills\u002Fcheckpoints-and-model-ops\u002F` for checkpoint layout inspection, `verl.model_merger`, FSDP\u002FMegatron\u002FHuggingFace export decisions, LoRA merge implications, profiling, and performance diagnostics.\n- **Maintain the repository**: use `sub-skills\u002Frepo-development\u002F` for contribution policy, duplicate-work checks, AGENTS.md editing rules, generated trainer config maintenance, focused test selection, and non-executing test suggestions.\n\n## Common Decision Points\n\n- If the user says “training fails,” first classify the failure: installation\u002Fimport (`setup-and-backends`), parquet\u002Freward schema (`data-and-rewards`), Hydra override\u002Fresource sizing (`training-and-configs`), rollout backend\u002Ftooling (`rollout-and-tools`), or checkpoint\u002Fexport (`checkpoints-and-model-ops`).\n- If the user asks for a command, prefer the nearest bundled command builder or validator; never require the original source checkout examples to remain available.\n- If the user asks for GPU execution, distinguish a base Python import check from real accelerator runtime validation. vLLM, SGLang, Megatron, flash-attn, TensorRT-LLM, ROCm, and NPU stacks need backend-specific environments.\n- If the user is editing the repo, apply `repo-development` before changing files so duplicate-work, AGENTS.md scope, generated config, and focused-test rules are not missed.\n\n## Bundled Helpers\n\n- `sub-skills\u002Fsetup-and-backends\u002Fscripts\u002Fcheck_verl_environment.py` checks active Python package metadata, imports, optional CUDA state, and dependency consistency.\n- `sub-skills\u002Fdata-and-rewards\u002Fscripts\u002Fvalidate_verl_parquet.py` validates verl parquet rows without importing `verl`.\n- `sub-skills\u002Ftraining-and-configs\u002Fscripts\u002Fbuild_ppo_command.py` prints a dry-run `python -m verl.trainer.main_ppo` command from common arguments.\n- `sub-skills\u002Frollout-and-tools\u002Fscripts\u002Fvalidate_function_tool.py` preflights trusted function-tool Python files for type hints and Google-style `Args:` docs.\n- `sub-skills\u002Fcheckpoints-and-model-ops\u002Fscripts\u002Finspect_checkpoint_path.py` classifies checkpoint\u002Frun\u002Frole directories without loading tensors.\n- `sub-skills\u002Frepo-development\u002Fscripts\u002Fselect_verl_tests.py` maps changed paths to suggested tests and pre-commit hooks without running them.\n\n## Safety Rules\n\n- Do not run training, dataset downloads, Docker builds, GPU\u002FNPU jobs, checkpoint merges, or source-mutating maintenance scripts unless the user explicitly asks and the environment is appropriate.\n- Treat original repo docs, examples, scripts, and tests as evidence only. Runtime guidance in this skill is self-contained and should use bundled references\u002Fscripts.\n- Do not leak local checkout paths, private Python prefixes, cache paths, credentials, or machine-specific install commands into public skill content.\n- For native verification, prefer CPU-safe tests and help\u002Fparser checks first; record GPU\u002FNPU\u002Fe2e examples as skipped unless the user authorizes heavy hardware runs.\n","# verl\n\nUse this repo skill when a user asks how to install, configure, run, debug, or maintain `verl`, the Volcano Engine Reinforcement Learning library for LLM post-training.\n\n## Start Here\n\n- Read `references\u002Frepo-provenance.md` before relying on this skill for a different checkout or after package metadata, config files, examples, or docs change.\n- Read `references\u002Ftroubleshooting.md` for cross-cutting failures that span setup, Hydra configs, data schema, rollout backends, checkpoint operations, and repo maintenance.\n- Run `scripts\u002Fcheck_skill_integrity.py` when editing or importing this skill to check frontmatter, links, private-path leaks, and generated cache files.\n\n## Route By User Task\n\n- **Install or verify an environment**: use `sub-skills\u002Fsetup-and-backends\u002F` for base install choices, Docker-vs-custom setup, Python\u002FCUDA requirements, optional extras, vLLM\u002FSGLang\u002FMegatron\u002FTensorRT-LLM\u002FNPU\u002FROCm caveats, and safe import\u002Fbackend checks.\n- **Prepare or validate data and rewards**: use `sub-skills\u002Fdata-and-rewards\u002F` for parquet row schemas, chat prompt fields, `reward_model.ground_truth`, custom reward modules, and the bundled parquet validator.\n- **Build training commands and Hydra overrides**: use `sub-skills\u002Ftraining-and-configs\u002F` for PPO, GRPO, SFT, on-policy distillation, backend strategy switches, batch\u002Fmicro-batch sizing, generated config references, and the dry-run PPO command builder.\n- **Configure rollout, agent loops, and tools**: use `sub-skills\u002Frollout-and-tools\u002F` for rollout engines, async generation server flows, multi-turn agent loops, `@function_tool`, native `BaseTool`, tokenization sanity checks, traces, and tool schema validation.\n- **Operate checkpoints and model artifacts**: use `sub-skills\u002Fcheckpoints-and-model-ops\u002F` for checkpoint layout inspection, `verl.model_merger`, FSDP\u002FMegatron\u002FHuggingFace export decisions, LoRA merge implications, profiling, and performance diagnostics.\n- **Maintain the repository**: use `sub-skills\u002Frepo-development\u002F` for contribution policy, duplicate-work checks, AGENTS.md editing rules, generated trainer config maintenance, focused test selection, and non-executing test suggestions.\n\n## Common Decision Points\n\n- If the user says “training fails,” first classify the failure: installation\u002Fimport (`setup-and-backends`), parquet\u002Freward schema (`data-and-rewards`), Hydra override\u002Fresource sizing (`training-and-configs`), rollout backend\u002Ftooling (`rollout-and-tools`), or checkpoint\u002Fexport (`checkpoints-and-model-ops`).\n- If the user asks for a command, prefer the nearest bundled command builder or validator; never require the original source checkout examples to remain available.\n- If the user asks for GPU execution, distinguish a base Python import check from real accelerator runtime validation. vLLM, SGLang, Megatron, flash-attn, TensorRT-LLM, ROCm, and NPU stacks need backend-specific environments.\n- If the user is editing the repo, apply `repo-development` before changing files so duplicate-work, AGENTS.md scope, generated config, and focused-test rules are not missed.\n\n## Bundled Helpers\n\n- `sub-skills\u002Fsetup-and-backends\u002Fscripts\u002Fcheck_verl_environment.py` checks active Python package metadata, imports, optional CUDA state, and dependency consistency.\n- `sub-skills\u002Fdata-and-rewards\u002Fscripts\u002Fvalidate_verl_parquet.py` validates verl parquet rows without importing `verl`.\n- `sub-skills\u002Ftraining-and-configs\u002Fscripts\u002Fbuild_ppo_command.py` prints a dry-run `python -m verl.trainer.main_ppo` command from common arguments.\n- `sub-skills\u002Frollout-and-tools\u002Fscripts\u002Fvalidate_function_tool.py` preflights trusted function-tool Python files for type hints and Google-style `Args:` docs.\n- `sub-skills\u002Fcheckpoints-and-model-ops\u002Fscripts\u002Finspect_checkpoint_path.py` classifies checkpoint\u002Frun\u002Frole directories without loading tensors.\n- `sub-skills\u002Frepo-development\u002Fscripts\u002Fselect_verl_tests.py` maps changed paths to suggested tests and pre-commit hooks without running them.\n\n## Safety Rules\n\n- Do not run training, dataset downloads, Docker builds, GPU\u002FNPU jobs, checkpoint merges, or source-mutating maintenance scripts unless the user explicitly asks and the environment is appropriate.\n- Treat original repo docs, examples, scripts, and tests as evidence only. Runtime guidance in this skill is self-contained and should use bundled references\u002Fscripts.\n- Do not leak local checkout paths, private Python prefixes, cache paths, credentials, or machine-specific install commands into public skill content.\n- For native verification, prefer CPU-safe tests and help\u002Fparser checks first; record GPU\u002FNPU\u002Fe2e examples as skipped unless the user authorizes heavy hardware runs.\n",{"name":5,"license":23,"metadata":24,"description":7,"disable-model-invocation":26},"Apache 2.0",{"disco-role":25},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fverl",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":26},[36,39,42],{"tool":37,"command":38},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill verl",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill verl",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill verl","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fverl\u002FSKILL.md",[47,53,57,61],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Frepo-provenance.md","REFERENCE","参考文档",2461,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fverl\u002Freferences\u002Frepo-provenance.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Frepo-routing-metadata.json",564,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fverl\u002Freferences\u002Frepo-routing-metadata.json",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Ftroubleshooting.md",3336,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fverl\u002Freferences\u002Ftroubleshooting.md",{"path":62,"resourceType":63,"resourceTypeDesc":64,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Fcheck_skill_integrity.py","SCRIPT","脚本",4783,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fverl\u002Fscripts\u002Fcheck_skill_integrity.py",[68,75,82,89,96,103],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":18},"01M799NYSH1R3RVAHSM2D88Y84","passnet-triton-opt","Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block\u002Fgrid\u002Fwarp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":18},"01M799NS5APEC0N1FZ7XZVW6BN","passnet-solve","END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-solve\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":18},"01M799NNY9BKSM21V5R7QVTBJX","passnet-skill","PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under .\u002Fpass_dir\u002F, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-skill\u002FSKILL.md",[],[],{"skillNo":90,"name":91,"nameZh":6,"description":92,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":93,"categories":94,"keywords":95,"publishedAt":18},"01M799NHH6NP7Z1RK77KGXAN6S","passnet-pattern-fusion","Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()\u002Freplacement_args()\u002Freplacement_func() files, when a pass \"failed to match\", or when deciding how to fuse multiple kernels\u002Fops.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-pattern-fusion\u002FSKILL.md",[],[],{"skillNo":97,"name":98,"nameZh":6,"description":99,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":100,"categories":101,"keywords":102,"publishedAt":18},"01M799ND2J4WNC2PGAZ0A06NDF","passnet-orchestrate","Round\u002Fbatch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep\u002Frevert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-orchestrate\u002FSKILL.md",[],[],{"skillNo":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":109,"publishedAt":18},"01M799N8SBTMY02ZDG2AZF5ZWE","passnet-feedback","Fast, deterministic feedback for PassNet work: pre-flight pattern\u002Fmatch verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-feedback\u002FSKILL.md",[],[]]