[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M796JKHSZDMF6VFGY52K2RF3":3,"skill-same:01M796JKHSZDMF6VFGY52K2RF3":74},{"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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":31,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":32,"install":33,"resources":45},"01M796JKHSZDMF6VFGY52K2RF3","post-training-rlhf",null,"Guide the train-llm-from-scratch post-training alignment pipeline: SFT, reward modeling, DPO\u002FORPO\u002FKTO, PPO, GRPO\u002FRLVR, rollout\u002Flog-prob mechanics, metrics, and troubleshooting.","VectorSpaceLab\u002FAREX-Skill","VectorSpaceLab","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F180637102?v=4","01M7538C5WPSJGWF7KHR4EFZ9F","01M7538C6R2GFP53NPXTCZNBRT",351,"MIT","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: post-training-rlhf\ndescription: \"Guide the train-llm-from-scratch post-training alignment pipeline:\n  SFT, reward modeling, DPO\u002FORPO\u002FKTO, PPO, GRPO\u002FRLVR, rollout\u002Flog-prob\n  mechanics, metrics, and troubleshooting.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# Post-Training RLHF Router\n\nUse this sub-skill when the task is about turning a pretrained base checkpoint into aligned instruction\u002Freasoning checkpoints with this repo's from-scratch post-training stack.\n\n## Use this for\n\n- Running or debugging SFT, reward-model training, DPO\u002FORPO\u002FKTO, PPO, and GRPO\u002FRLVR stages.\n- Understanding masked SFT loss, Bradley-Terry reward modeling, preference objectives, PPO GAE\u002Fclipping\u002FKL, GRPO group-relative advantages, rollout generation, log-prob recomputation, and verifier rewards.\n- Planning checkpoint\u002Fdata dependencies across `base_pretrained.pt -> sft.pt -> reward.pt \u002F dpo.pt \u002F ppo.pt \u002F grpo.pt`.\n- Interpreting stage JSONL metrics and deciding safe recovery actions.\n- Building dry-run stage commands before launching expensive training.\n\n## Route elsewhere\n\n- Prepare or validate raw Pile\u002FSFT\u002Fpreference\u002FRL prompt files: `..\u002Fdata-preparation\u002FSKILL.md`.\n- Pretrain the base Transformer or inspect model architecture: `..\u002Fmodel-pretraining\u002FSKILL.md`.\n- Run GSM8K table evaluation, parse answers for evaluation, or chat with a checkpoint: `..\u002Fevaluation-chat\u002FSKILL.md`.\n- Edit config files through forms, use the Streamlit UI, or explain config merge precedence: `..\u002Fconfiguration-ui\u002FSKILL.md`.\n\n## Fast operating path\n\n1. Identify the requested stage and verify its upstream checkpoint\u002Fdata dependencies with [references\u002Fstage-workflows.md](references\u002Fstage-workflows.md).\n2. Use [scripts\u002Fbuild_stage_command.py](scripts\u002Fbuild_stage_command.py) or [scripts\u002Fplan_posttraining_pipeline.py](scripts\u002Fplan_posttraining_pipeline.py) to print a dry-run command. Inspect it before running anything expensive.\n3. For algorithm-specific questions, use [references\u002Falgorithm-reference.md](references\u002Falgorithm-reference.md) to map metrics and function names to the underlying loss\u002Freward\u002Flog-prob mechanics.\n4. For failures, start with [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md), then route to data\u002Fmodel\u002Feval\u002FUI sub-skills when the root cause is outside post-training.\n5. Inspect JSONL metrics with [scripts\u002Finspect_metrics_jsonl.py](scripts\u002Finspect_metrics_jsonl.py) before changing hyperparameters; prefer the smallest recovery that matches the observed metric failure.\n\n## Safety and scope notes\n\n- The bundled scripts are dry-run or read-only helpers. They do not download data, train, delete checkpoints, or modify configs.\n- Training commands assume the user is in their own repo checkout or compatible workspace and has already prepared data\u002Fcheckpoints.\n- Keep log-prob math in fp32 when comparing policy\u002Freference\u002Fold log-probs; do not \"optimize\" this away under bf16.\n- Treat Weights & Biases as optional. JSONL metrics are the source of truth.\n","# Post-Training RLHF Router\n\nUse this sub-skill when the task is about turning a pretrained base checkpoint into aligned instruction\u002Freasoning checkpoints with this repo's from-scratch post-training stack.\n\n## Use this for\n\n- Running or debugging SFT, reward-model training, DPO\u002FORPO\u002FKTO, PPO, and GRPO\u002FRLVR stages.\n- Understanding masked SFT loss, Bradley-Terry reward modeling, preference objectives, PPO GAE\u002Fclipping\u002FKL, GRPO group-relative advantages, rollout generation, log-prob recomputation, and verifier rewards.\n- Planning checkpoint\u002Fdata dependencies across `base_pretrained.pt -> sft.pt -> reward.pt \u002F dpo.pt \u002F ppo.pt \u002F grpo.pt`.\n- Interpreting stage JSONL metrics and deciding safe recovery actions.\n- Building dry-run stage commands before launching expensive training.\n\n## Route elsewhere\n\n- Prepare or validate raw Pile\u002FSFT\u002Fpreference\u002FRL prompt files: `..\u002Fdata-preparation\u002FSKILL.md`.\n- Pretrain the base Transformer or inspect model architecture: `..\u002Fmodel-pretraining\u002FSKILL.md`.\n- Run GSM8K table evaluation, parse answers for evaluation, or chat with a checkpoint: `..\u002Fevaluation-chat\u002FSKILL.md`.\n- Edit config files through forms, use the Streamlit UI, or explain config merge precedence: `..\u002Fconfiguration-ui\u002FSKILL.md`.\n\n## Fast operating path\n\n1. Identify the requested stage and verify its upstream checkpoint\u002Fdata dependencies with [references\u002Fstage-workflows.md](references\u002Fstage-workflows.md).\n2. Use [scripts\u002Fbuild_stage_command.py](scripts\u002Fbuild_stage_command.py) or [scripts\u002Fplan_posttraining_pipeline.py](scripts\u002Fplan_posttraining_pipeline.py) to print a dry-run command. Inspect it before running anything expensive.\n3. For algorithm-specific questions, use [references\u002Falgorithm-reference.md](references\u002Falgorithm-reference.md) to map metrics and function names to the underlying loss\u002Freward\u002Flog-prob mechanics.\n4. For failures, start with [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md), then route to data\u002Fmodel\u002Feval\u002FUI sub-skills when the root cause is outside post-training.\n5. Inspect JSONL metrics with [scripts\u002Finspect_metrics_jsonl.py](scripts\u002Finspect_metrics_jsonl.py) before changing hyperparameters; prefer the smallest recovery that matches the observed metric failure.\n\n## Safety and scope notes\n\n- The bundled scripts are dry-run or read-only helpers. They do not download data, train, delete checkpoints, or modify configs.\n- Training commands assume the user is in their own repo checkout or compatible workspace and has already prepared data\u002Fcheckpoints.\n- Keep log-prob math in fp32 when comparing policy\u002Freference\u002Fold log-probs; do not \"optimize\" this away under bf16.\n- Treat Weights & Biases as optional. JSONL metrics are the source of truth.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf",27,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab","2026-09-03 10:25:21",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":25},[35,38,41],{"tool":36,"command":37},"npx","npx skills add VectorSpaceLab\u002FAREX-Skill --skill post-training-rlhf",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill post-training-rlhf",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill post-training-rlhf","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002FSKILL.md",[46,52,56,60,66,70],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Falgorithm-reference.md","REFERENCE","参考文档",9460,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002Freferences\u002Falgorithm-reference.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fstage-workflows.md",9570,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002Freferences\u002Fstage-workflows.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Ftroubleshooting.md",8644,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002Freferences\u002Ftroubleshooting.md",{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"scripts\u002Fbuild_stage_command.py","SCRIPT","脚本",6678,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002Fscripts\u002Fbuild_stage_command.py",{"path":67,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":68,"sourceHtmlUrl":69},"scripts\u002Finspect_metrics_jsonl.py",5027,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002Fscripts\u002Finspect_metrics_jsonl.py",{"path":71,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":72,"sourceHtmlUrl":73},"scripts\u002Fplan_posttraining_pipeline.py",7289,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Ftrain-llm-from-scratch\u002Fsub-skills\u002Fpost-training-rlhf\u002Fscripts\u002Fplan_posttraining_pipeline.py",[75,83,90,97,104,111],{"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":79,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"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.","Apache-2.0","https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Ftask-oriented\u002FPassNet\u002Fskills\u002Fpassnet-triton-opt\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":79,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"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":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":79,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"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":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":79,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":103,"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":105,"name":106,"nameZh":6,"description":107,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":79,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"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":112,"name":113,"nameZh":6,"description":114,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":79,"licenseFileUrl":6,"sourceHtmlUrl":115,"categories":116,"keywords":117,"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",[],[]]