[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7920SWBEKNFEM2S7HS5SSTV":3,"skill-same:01M7920SWBEKNFEM2S7HS5SSTV":62},{"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},"01M7920SWBEKNFEM2S7HS5SSTV","conversion",null,"Routes OpenNMT-py checkpoint averaging, release, CTranslate2, external-family conversion, vocabulary or embedding extraction, and LoRA merge workflows.","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\u002Fopennmt-py\u002Fsub-skills\u002Fconversion\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: conversion\ndescription: \"Routes OpenNMT-py checkpoint averaging, release, CTranslate2,\n  external-family conversion, vocabulary or embedding extraction, and LoRA merge\n  workflows.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# Conversion\n\nUse this sub-skill when the task is about OpenNMT-py checkpoint maintenance or model-format conversion rather than training or translation.\n\n## Use this route when\n\n- The request says average checkpoints, strip optimizer state, release a model, convert to CTranslate2, quantize for CT2, or inspect a saved checkpoint.\n- The user needs to convert a LLaMA, Mistral, Mixtral, Phi, T5, MPT, Falcon, RedPajama, XGen, or Hugging Face-style checkpoint into OpenNMT-py format.\n- The user needs to extract source\u002Ftarget vocabularies, inspect or export embeddings, convert pretrained embeddings to tensors, or check vocab availability.\n- The user needs to merge or concatenate LoRA weights into a compatible base OpenNMT-py model.\n\n## Do not use this route when\n\n- The main work is building corpora or tokenizers before training; use the data-preparation route.\n- The main work is choosing architecture, optimizer, LoRA training options, or checkpoint continuation; use the training route.\n- The main work is running inference or serving an already released model; use the inference route.\n\n## Start here\n\n1. Read `references\u002Fcheckpoint-tools.md` to identify the exact checkpoint family, expected files, command\u002FAPI surface, validation checks, and output format.\n2. Before changing a checkpoint, inspect the input with `scripts\u002Fcheck_checkpoint_file.py`:\n\n   ```bash\n   python scripts\u002Fcheck_checkpoint_file.py path\u002Fto\u002Fmodel.pt --json\n   ```\n\n   The helper maps tensors to CPU or fake tensors and reports top-level keys, option fields, vocab sides, tensor counts, and conversion-readiness hints.\n3. Read `references\u002Ftroubleshooting.md` before running weight-heavy conversion, CTranslate2 release, external\u002FHugging Face conversion, or LoRA merge.\n4. Prefer package console commands for installed OpenNMT-py entry points: `onmt_average_models` and `onmt_release_model`. For utilities that were only source-tree tools, use the distilled contracts in `references\u002Fcheckpoint-tools.md` and create a task-local reviewed helper instead of depending on any particular checkout path.\n\n## Primary workflows\n\n- **Average compatible checkpoints**: check all inputs, confirm identical vocabulary and architecture options, then run `onmt_average_models -models ... -output averaged.pt`; add `-fp32` when the release target must not preserve half precision.\n- **Release PyTorch or CTranslate2 models**: use `onmt_release_model --format pytorch` to drop optimizer state, or `--format ctranslate2` with an optional quantization mode to create a CT2 model directory.\n- **Convert external model families**: use the family matrix and option contracts in `references\u002Fcheckpoint-tools.md`; validate that the output checkpoint has `model`, `generator`, `vocab`, and `opt` sections before training or release.\n- **Extract vocabularies or embeddings**: confirm `vocab.src`\u002F`vocab.tgt` first, then use the reference recipes to write vocab text files or embedding tensors without relying on source checkout scripts.\n- **Merge LoRA weights**: verify the base and LoRA checkpoints share architecture\u002Fvocab assumptions, merge for inference or concatenate for resumed training, then inspect the output and release it if needed.\n\n## Safety and validation rules\n\n- Treat `torch.load` checkpoints as trusted pickle files; do not inspect arbitrary untrusted checkpoints without an isolation plan.\n- Conversion and inspection should not place tensors on GPU. Use CPU\u002Ffake-tensor loading for checks, and only select GPU for downstream inference\u002Ftraining if the owning route requires it.\n- Never average checkpoints from different vocabularies, architectures, model tasks, or incompatible LoRA\u002Fbase states.\n- For CT2, validate that the output is a directory with CT2 files such as `model.bin`, `config.json`, and `vocabulary.json`, not a `.pt` checkpoint.\n- Keep real model paths, Hugging Face tokens, cache directories, and machine-specific paths out of skill notes and reusable scripts.\n\n## Bundled files\n\n- `references\u002Fcheckpoint-tools.md` — workflows, command\u002FAPI contracts, external family matrix, vocab\u002Fembedding and LoRA guidance.\n- `references\u002Ftroubleshooting.md` — symptoms, likely causes, and recovery steps for conversion-specific failures.\n- `scripts\u002Fcheck_checkpoint_file.py` — safe checkpoint inventory helper for OpenNMT-py `.pt` metadata and tensor summaries.\n","# Conversion\n\nUse this sub-skill when the task is about OpenNMT-py checkpoint maintenance or model-format conversion rather than training or translation.\n\n## Use this route when\n\n- The request says average checkpoints, strip optimizer state, release a model, convert to CTranslate2, quantize for CT2, or inspect a saved checkpoint.\n- The user needs to convert a LLaMA, Mistral, Mixtral, Phi, T5, MPT, Falcon, RedPajama, XGen, or Hugging Face-style checkpoint into OpenNMT-py format.\n- The user needs to extract source\u002Ftarget vocabularies, inspect or export embeddings, convert pretrained embeddings to tensors, or check vocab availability.\n- The user needs to merge or concatenate LoRA weights into a compatible base OpenNMT-py model.\n\n## Do not use this route when\n\n- The main work is building corpora or tokenizers before training; use the data-preparation route.\n- The main work is choosing architecture, optimizer, LoRA training options, or checkpoint continuation; use the training route.\n- The main work is running inference or serving an already released model; use the inference route.\n\n## Start here\n\n1. Read `references\u002Fcheckpoint-tools.md` to identify the exact checkpoint family, expected files, command\u002FAPI surface, validation checks, and output format.\n2. Before changing a checkpoint, inspect the input with `scripts\u002Fcheck_checkpoint_file.py`:\n\n   ```bash\n   python scripts\u002Fcheck_checkpoint_file.py path\u002Fto\u002Fmodel.pt --json\n   ```\n\n   The helper maps tensors to CPU or fake tensors and reports top-level keys, option fields, vocab sides, tensor counts, and conversion-readiness hints.\n3. Read `references\u002Ftroubleshooting.md` before running weight-heavy conversion, CTranslate2 release, external\u002FHugging Face conversion, or LoRA merge.\n4. Prefer package console commands for installed OpenNMT-py entry points: `onmt_average_models` and `onmt_release_model`. For utilities that were only source-tree tools, use the distilled contracts in `references\u002Fcheckpoint-tools.md` and create a task-local reviewed helper instead of depending on any particular checkout path.\n\n## Primary workflows\n\n- **Average compatible checkpoints**: check all inputs, confirm identical vocabulary and architecture options, then run `onmt_average_models -models ... -output averaged.pt`; add `-fp32` when the release target must not preserve half precision.\n- **Release PyTorch or CTranslate2 models**: use `onmt_release_model --format pytorch` to drop optimizer state, or `--format ctranslate2` with an optional quantization mode to create a CT2 model directory.\n- **Convert external model families**: use the family matrix and option contracts in `references\u002Fcheckpoint-tools.md`; validate that the output checkpoint has `model`, `generator`, `vocab`, and `opt` sections before training or release.\n- **Extract vocabularies or embeddings**: confirm `vocab.src`\u002F`vocab.tgt` first, then use the reference recipes to write vocab text files or embedding tensors without relying on source checkout scripts.\n- **Merge LoRA weights**: verify the base and LoRA checkpoints share architecture\u002Fvocab assumptions, merge for inference or concatenate for resumed training, then inspect the output and release it if needed.\n\n## Safety and validation rules\n\n- Treat `torch.load` checkpoints as trusted pickle files; do not inspect arbitrary untrusted checkpoints without an isolation plan.\n- Conversion and inspection should not place tensors on GPU. Use CPU\u002Ffake-tensor loading for checks, and only select GPU for downstream inference\u002Ftraining if the owning route requires it.\n- Never average checkpoints from different vocabularies, architectures, model tasks, or incompatible LoRA\u002Fbase states.\n- For CT2, validate that the output is a directory with CT2 files such as `model.bin`, `config.json`, and `vocabulary.json`, not a `.pt` checkpoint.\n- Keep real model paths, Hugging Face tokens, cache directories, and machine-specific paths out of skill notes and reusable scripts.\n\n## Bundled files\n\n- `references\u002Fcheckpoint-tools.md` — workflows, command\u002FAPI contracts, external family matrix, vocab\u002Fembedding and LoRA guidance.\n- `references\u002Ftroubleshooting.md` — symptoms, likely causes, and recovery steps for conversion-specific failures.\n- `scripts\u002Fcheck_checkpoint_file.py` — safe checkpoint inventory helper for OpenNMT-py `.pt` metadata and tensor summaries.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fopennmt-py\u002Fsub-skills\u002Fconversion",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 conversion",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill conversion",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill conversion","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fopennmt-py\u002Fsub-skills\u002Fconversion\u002FSKILL.md",[46,52,56],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fcheckpoint-tools.md","REFERENCE","参考文档",13361,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fopennmt-py\u002Fsub-skills\u002Fconversion\u002Freferences\u002Fcheckpoint-tools.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Ftroubleshooting.md",10553,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fopennmt-py\u002Fsub-skills\u002Fconversion\u002Freferences\u002Ftroubleshooting.md",{"path":57,"resourceType":58,"resourceTypeDesc":59,"sizeBytes":60,"sourceHtmlUrl":61},"scripts\u002Fcheck_checkpoint_file.py","SCRIPT","脚本",18721,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fopennmt-py\u002Fsub-skills\u002Fconversion\u002Fscripts\u002Fcheck_checkpoint_file.py",[63,71,78,85,92,99],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"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":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"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":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"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":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"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":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"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":100,"name":101,"nameZh":6,"description":102,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":67,"licenseFileUrl":6,"sourceHtmlUrl":103,"categories":104,"keywords":105,"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",[],[]]