[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78VK30Z8DPAXDX96SGS5CP8":3,"skill-same:01M78VK30Z8DPAXDX96SGS5CP8":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},"01M78VK30Z8DPAXDX96SGS5CP8","models-and-serving",null,"Configure DB-GPT 0.8.1 model providers and local backends, validate model TOML, operate model controller\u002Fworker\u002FAPI-server roles, use model-management CLI commands, and diagnose serving failures without mistaking an import or registry record for a live deployment.","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\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: models-and-serving\ndescription: \"Configure DB-GPT 0.8.1 model providers and local backends,\n  validate model TOML, operate model controller\u002Fworker\u002FAPI-server roles, use\n  model-management CLI commands, and diagnose serving failures without mistaking\n  an import or registry record for a live deployment.\"\nmetadata:\n  disco-role: operating\ndisable-model-invocation: true\nlicense: MIT\n---\n\n# Models and serving\n\nUse this route when a DB-GPT task mentions model\u002Fprovider TOML, `models.llms`,\nembeddings or rerankers, OpenAI-compatible endpoints, Ollama, Hugging Face,\nvLLM, llama.cpp, MLX, model workers, controller, model API server, model CLI,\nmodel registration, health\u002Fheartbeat, GPU memory, quantization, or serving logs.\n\nThis route owns model selection and the model-service topology. Route general\ninstallation, profile creation, workspace paths, and generic `start webserver`\nsetup to `setup-and-cli`; route agents\u002FAWEL to `agents-and-awel`; route public\nclient CRUD and sandbox calls to `apis-client-and-sandbox`; route document,\nembedding-pipeline, vector-store, and knowledge workflows to `data-and-rag`.\n\n## Operating contract\n\n- Work against the public DB-GPT 0.8.1 package interface, not a source checkout.\n- Treat a TOML parse, Python import, controller registry row, or `healthy` value\n  observed before a real request as configuration evidence only. Deployment is\n  successful only after the intended service is reachable and a suitable model\n  operation succeeds.\n- Keep provider credentials in environment variables. Never print resolved\n  secrets, copy them into generated files, or put them in shell history.\n- Choose one model role explicitly: `llm`, `text2vec` (embedding), or\n  `reranker`. A chat model cannot silently substitute for an embedding model.\n- CPU is the verified baseline for this route. CUDA\u002Flocal-model, vLLM,\n  bitsandbytes, llama.cpp acceleration, and MLX are optional backend paths;\n  do not claim they work from a CPU import or from host GPU visibility alone.\n\nRead the focused contracts before making changes:\n\n- [model-api-reference.md](references\u002Fmodel-api-reference.md) — configuration\n  dataclasses, model\u002Fprovider fields, worker types, and OpenAI-compatible API\n  surface.\n- [provider-and-backend-matrix.md](references\u002Fprovider-and-backend-matrix.md)\n  — provider credentials, endpoint conventions, embedding pairing, optional\n  dependencies, and backend boundaries.\n- [serving-workflows.md](references\u002Fserving-workflows.md) — proxy\u002Flocal TOML,\n  unified webserver, split controller\u002Fworker\u002FAPI-server deployment, and CLI\n  operations.\n- [backends.md](references\u002Fbackends.md) — local-model, quantization, VRAM, and\n  hardware-specific decision rules.\n- [troubleshooting.md](references\u002Ftroubleshooting.md) — symptom-to-check\n  diagnosis and recovery boundaries.\n\n## Standard workflow\n\n1. **Identify the topology and model roles.** Decide whether the user needs a\n   unified webserver process, a split cluster, a standalone model API server,\n   or only a provider configuration. Record the LLM, embedding, and optional\n   reranker names separately.\n2. **Select the provider\u002Fbackend.** Prefer a remote proxy for a CPU-first smoke\n   test. For a local model, require a real model path or an explicitly allowed\n   model download and select `hf`, `vllm`, `llama.cpp`,\n   `llama.cpp.server`, or `mlx` deliberately. Check the matrix for the\n   provider-specific key and endpoint.\n3. **Write a redacted TOML.** Put model entries under `[models]` using repeated\n   `[[models.llms]]`, `[[models.embeddings]]`, and `[[models.rerankers]]` tables.\n   Use `${env:NAME}` or `${env:NAME:-default}` placeholders. Set\n   `default_llm`, `default_embedding`, or `default_reranker` only when multiple\n   entries make selection ambiguous. Keep web\u002Fdatabase configuration owned by\n   the setup route unless the serving topology requires the related model\n   service section.\n4. **Validate without network or model loading.** Run the bundled checker:\n\n   ```bash\n   python scripts\u002Fmodel_config_check.py path\u002Fto\u002Fconfig.toml\n   ```\n\n   It parses TOML, checks model-table shape, provider\u002Frole requirements,\n   duplicate names, placeholder syntax, endpoint URLs, and obvious service-port\n   collisions. It never resolves secrets, calls a provider, contacts a\n   controller, downloads a model, or starts a process. Use `--json` for a\n   machine-readable report and `--allow-missing-embeddings` only when the\n   configuration is intentionally chat-only; RAG\u002Fknowledge workflows must not\n   use that exception.\n5. **Run a safe package-level check.** Use `dbgpt --version`, the relevant\n   static `--help`, and import\u002Fconfig-parser checks. Do not use `dbgpt model\n   start --help` or another dynamic model command as an offline gate: model\n   command discovery can query a controller and return a timeout\u002F502 when no\n   controller is reachable.\n6. **Start only the chosen topology.** Use the exact commands in\n   [serving-workflows.md](references\u002Fserving-workflows.md). Do not start both a\n   unified model manager and an independent worker on the same ports. For\n   split mode, start and verify the controller before workers, then verify\n   workers before the API\u002Fweb layer.\n7. **Verify reachability and model capability.** Check the controller health and\n   registry, then issue a minimal request through the intended API or provider.\n   `dbgpt model list` is a registry\u002Fheartbeat view, not proof that generation or\n   embedding works. An unreachable controller, failed heartbeat, missing model,\n   authentication failure, or failed embedding request is a deployment failure;\n   report it as such with the exact next check.\n8. **Diagnose from the right boundary.** Inspect the command's resolved config\n   path, service logs, bind address, port, controller address, provider base URL,\n   model name, and optional-backend imports in that order. Redact API keys and\n   authorization headers in reports.\n\n## CLI routing facts\n\nThe DB-GPT CLI registers model commands under `dbgpt model` and service commands\nunder `dbgpt start` \u002F `dbgpt stop`:\n\n```text\ndbgpt model [--address CONTROLLER_URL] list|start|stop|restart|chat\ndbgpt start controller --config FILE [--daemon]\ndbgpt start worker --config FILE [--daemon]\ndbgpt start apiserver --config FILE [--daemon]\ndbgpt start webserver [web options]\ndbgpt stop controller|worker|apiserver [--port PORT]\n```\n\n`CONTROLLER_ADDRESS` can supply the controller address used by model-management\ncommands when `--address` is omitted. The dynamic `model start` command obtains\nsupported-model metadata from the controller before constructing some options;\ntherefore a failed remote request during help or invocation is not a successful\nstart. For model-service launch commands, the configuration file is the\nsource of worker\u002Fmodel parameters; do not rely on stale examples that show\nmodel flags which are not present in the installed command tree.\n\n## Validation and boundaries\n\nA valid result for this route includes:\n\n- provider\u002Fmodel\u002Frole entries parse with no secret resolution or network call;\n- an embedding entry exists and is compatible with the intended RAG path;\n- each selected local backend's dependency, path, device, and VRAM assumptions\n  are explicit;\n- service roles have distinct, reachable addresses and non-conflicting ports;\n- controller registry and heartbeat checks are separated from generation,\n  completion, embedding, or reranking checks;\n- optional CUDA\u002FGPU\u002Fprovider\u002Fservice dependencies are labeled unverified when\n  they were not executed.\n\nDo not claim that `dbgpt model list` proves deployment, that `--daemon` proves a\nhealthy child process, that a provider key exists merely because a placeholder\nparsed, or that an A100-visible host proves CUDA\u002Ftorch\u002FvLLM\u002Fbitsandbytes support.\nDo not add API CRUD, agent orchestration, or RAG chunking instructions here.\n","# Models and serving\n\nUse this route when a DB-GPT task mentions model\u002Fprovider TOML, `models.llms`,\nembeddings or rerankers, OpenAI-compatible endpoints, Ollama, Hugging Face,\nvLLM, llama.cpp, MLX, model workers, controller, model API server, model CLI,\nmodel registration, health\u002Fheartbeat, GPU memory, quantization, or serving logs.\n\nThis route owns model selection and the model-service topology. Route general\ninstallation, profile creation, workspace paths, and generic `start webserver`\nsetup to `setup-and-cli`; route agents\u002FAWEL to `agents-and-awel`; route public\nclient CRUD and sandbox calls to `apis-client-and-sandbox`; route document,\nembedding-pipeline, vector-store, and knowledge workflows to `data-and-rag`.\n\n## Operating contract\n\n- Work against the public DB-GPT 0.8.1 package interface, not a source checkout.\n- Treat a TOML parse, Python import, controller registry row, or `healthy` value\n  observed before a real request as configuration evidence only. Deployment is\n  successful only after the intended service is reachable and a suitable model\n  operation succeeds.\n- Keep provider credentials in environment variables. Never print resolved\n  secrets, copy them into generated files, or put them in shell history.\n- Choose one model role explicitly: `llm`, `text2vec` (embedding), or\n  `reranker`. A chat model cannot silently substitute for an embedding model.\n- CPU is the verified baseline for this route. CUDA\u002Flocal-model, vLLM,\n  bitsandbytes, llama.cpp acceleration, and MLX are optional backend paths;\n  do not claim they work from a CPU import or from host GPU visibility alone.\n\nRead the focused contracts before making changes:\n\n- [model-api-reference.md](references\u002Fmodel-api-reference.md) — configuration\n  dataclasses, model\u002Fprovider fields, worker types, and OpenAI-compatible API\n  surface.\n- [provider-and-backend-matrix.md](references\u002Fprovider-and-backend-matrix.md)\n  — provider credentials, endpoint conventions, embedding pairing, optional\n  dependencies, and backend boundaries.\n- [serving-workflows.md](references\u002Fserving-workflows.md) — proxy\u002Flocal TOML,\n  unified webserver, split controller\u002Fworker\u002FAPI-server deployment, and CLI\n  operations.\n- [backends.md](references\u002Fbackends.md) — local-model, quantization, VRAM, and\n  hardware-specific decision rules.\n- [troubleshooting.md](references\u002Ftroubleshooting.md) — symptom-to-check\n  diagnosis and recovery boundaries.\n\n## Standard workflow\n\n1. **Identify the topology and model roles.** Decide whether the user needs a\n   unified webserver process, a split cluster, a standalone model API server,\n   or only a provider configuration. Record the LLM, embedding, and optional\n   reranker names separately.\n2. **Select the provider\u002Fbackend.** Prefer a remote proxy for a CPU-first smoke\n   test. For a local model, require a real model path or an explicitly allowed\n   model download and select `hf`, `vllm`, `llama.cpp`,\n   `llama.cpp.server`, or `mlx` deliberately. Check the matrix for the\n   provider-specific key and endpoint.\n3. **Write a redacted TOML.** Put model entries under `[models]` using repeated\n   `[[models.llms]]`, `[[models.embeddings]]`, and `[[models.rerankers]]` tables.\n   Use `${env:NAME}` or `${env:NAME:-default}` placeholders. Set\n   `default_llm`, `default_embedding`, or `default_reranker` only when multiple\n   entries make selection ambiguous. Keep web\u002Fdatabase configuration owned by\n   the setup route unless the serving topology requires the related model\n   service section.\n4. **Validate without network or model loading.** Run the bundled checker:\n\n   ```bash\n   python scripts\u002Fmodel_config_check.py path\u002Fto\u002Fconfig.toml\n   ```\n\n   It parses TOML, checks model-table shape, provider\u002Frole requirements,\n   duplicate names, placeholder syntax, endpoint URLs, and obvious service-port\n   collisions. It never resolves secrets, calls a provider, contacts a\n   controller, downloads a model, or starts a process. Use `--json` for a\n   machine-readable report and `--allow-missing-embeddings` only when the\n   configuration is intentionally chat-only; RAG\u002Fknowledge workflows must not\n   use that exception.\n5. **Run a safe package-level check.** Use `dbgpt --version`, the relevant\n   static `--help`, and import\u002Fconfig-parser checks. Do not use `dbgpt model\n   start --help` or another dynamic model command as an offline gate: model\n   command discovery can query a controller and return a timeout\u002F502 when no\n   controller is reachable.\n6. **Start only the chosen topology.** Use the exact commands in\n   [serving-workflows.md](references\u002Fserving-workflows.md). Do not start both a\n   unified model manager and an independent worker on the same ports. For\n   split mode, start and verify the controller before workers, then verify\n   workers before the API\u002Fweb layer.\n7. **Verify reachability and model capability.** Check the controller health and\n   registry, then issue a minimal request through the intended API or provider.\n   `dbgpt model list` is a registry\u002Fheartbeat view, not proof that generation or\n   embedding works. An unreachable controller, failed heartbeat, missing model,\n   authentication failure, or failed embedding request is a deployment failure;\n   report it as such with the exact next check.\n8. **Diagnose from the right boundary.** Inspect the command's resolved config\n   path, service logs, bind address, port, controller address, provider base URL,\n   model name, and optional-backend imports in that order. Redact API keys and\n   authorization headers in reports.\n\n## CLI routing facts\n\nThe DB-GPT CLI registers model commands under `dbgpt model` and service commands\nunder `dbgpt start` \u002F `dbgpt stop`:\n\n```text\ndbgpt model [--address CONTROLLER_URL] list|start|stop|restart|chat\ndbgpt start controller --config FILE [--daemon]\ndbgpt start worker --config FILE [--daemon]\ndbgpt start apiserver --config FILE [--daemon]\ndbgpt start webserver [web options]\ndbgpt stop controller|worker|apiserver [--port PORT]\n```\n\n`CONTROLLER_ADDRESS` can supply the controller address used by model-management\ncommands when `--address` is omitted. The dynamic `model start` command obtains\nsupported-model metadata from the controller before constructing some options;\ntherefore a failed remote request during help or invocation is not a successful\nstart. For model-service launch commands, the configuration file is the\nsource of worker\u002Fmodel parameters; do not rely on stale examples that show\nmodel flags which are not present in the installed command tree.\n\n## Validation and boundaries\n\nA valid result for this route includes:\n\n- provider\u002Fmodel\u002Frole entries parse with no secret resolution or network call;\n- an embedding entry exists and is compatible with the intended RAG path;\n- each selected local backend's dependency, path, device, and VRAM assumptions\n  are explicit;\n- service roles have distinct, reachable addresses and non-conflicting ports;\n- controller registry and heartbeat checks are separated from generation,\n  completion, embedding, or reranking checks;\n- optional CUDA\u002FGPU\u002Fprovider\u002Fservice dependencies are labeled unverified when\n  they were not executed.\n\nDo not claim that `dbgpt model list` proves deployment, that `--daemon` proves a\nhealthy child process, that a provider key exists merely because a placeholder\nparsed, or that an A100-visible host proves CUDA\u002Ftorch\u002FvLLM\u002Fbitsandbytes support.\nDo not add API CRUD, agent orchestration, or RAG chunking instructions here.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving",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 models-and-serving",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill models-and-serving",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill models-and-serving","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002FSKILL.md",[46,52,56,60,64,68],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fbackends.md","REFERENCE","参考文档",4932,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002Freferences\u002Fbackends.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fmodel-api-reference.md",7748,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002Freferences\u002Fmodel-api-reference.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fprovider-and-backend-matrix.md",6399,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002Freferences\u002Fprovider-and-backend-matrix.md",{"path":61,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":62,"sourceHtmlUrl":63},"references\u002Fserving-workflows.md",7531,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002Freferences\u002Fserving-workflows.md",{"path":65,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":66,"sourceHtmlUrl":67},"references\u002Ftroubleshooting.md",7117,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002Freferences\u002Ftroubleshooting.md",{"path":69,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":72,"sourceHtmlUrl":73},"scripts\u002Fmodel_config_check.py","SCRIPT","脚本",10341,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdb-gpt\u002Fsub-skills\u002Fmodels-and-serving\u002Fscripts\u002Fmodel_config_check.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",[],[]]