[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78X4YS00291F5VHC61QWNEM":3,"skill-same:01M78X4YS00291F5VHC61QWNEM":66},{"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},"01M78X4YS00291F5VHC61QWNEM","inference",null,"Load and smoke-check FlagEmbedding embedders, M3 models, and rerankers for retrieval inference.","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\u002Fflag-embedding\u002Fsub-skills\u002Finference\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: inference\ndescription: \"Load and smoke-check FlagEmbedding embedders, M3 models, and\n  rerankers for retrieval inference.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: MIT\n---\n\n# Inference\n\nUse this sub-skill when the task is to load a FlagEmbedding model for inference,\nencode queries or corpus passages, compute embedding similarity, compute BGE-M3\ndense\u002Fsparse\u002FColBERT scores, rerank query-passage pairs, or smoke-check the\ninference API before a retrieval run.\n\nDo not use this sub-skill for fine-tuning data formats, fine-tuning launchers,\ntraining hyperparameters, or checkpoint production. Route those tasks to the\nsibling fine-tuning sub-skill. Do not use it for benchmark suites, leaderboard\nevaluation, or metric scripts. Route those tasks to the sibling evaluation\nsub-skill. For package installation, import failures, and global optional\ndependency repair, route to the root troubleshooting reference at\n`..\u002F..\u002Freferences\u002Ftroubleshooting.md`.\n\n## Start Here\n\n1. Decide whether the model is an embedder, an M3 embedder, or a reranker.\n2. Prefer `FlagAutoModel.from_finetuned(...)` for embedders and\n   `FlagAutoReranker.from_finetuned(...)` for rerankers when the checkpoint name\n   is in the built-in mapping.\n3. For a custom local checkpoint or an unmapped remote checkpoint, pass\n   `model_class` explicitly and choose the concrete class behavior from\n   [references\u002Fapi-reference.md](references\u002Fapi-reference.md).\n4. For retrieval, use `encode_queries(...)` for queries and\n   `encode_corpus(...)` for passages. Query instructions are applied by\n   `encode_queries`; passages normally should not receive the query instruction.\n5. For reranking, pass a pair as `(\"query\", \"passage\")` or a batch as a list\n   of pairs to `compute_score(...)`; validate the returned list or nested list\n   shape before sorting results.\n6. For a safe local check, run `scripts\u002Fsmoke_inference_api.py --help` first.\n   Running the script with no `--model-name` performs import and signature\n   probing only. Loading or encoding happens only when `--model-name` is passed.\n\n## Public Entry Points\n\nEmbedder auto loader and concrete classes:\n\n- `FlagAutoModel`\n- `FlagModel`\n- `BGEM3FlagModel`\n- `FlagLLMModel`\n- `FlagICLModel`\n- `FlagPseudoMoEModel`\n\nReranker auto loader and concrete classes:\n\n- `FlagAutoReranker`\n- `FlagReranker`\n- `FlagLLMReranker`\n- `LayerWiseFlagLLMReranker`\n- `LightWeightFlagLLMReranker`\n\nEmbedder `model_class` ids:\n\n- `encoder-only-base`: encoder-only dense embedder, normally `FlagModel`.\n- `encoder-only-m3`: BGE-M3 dense\u002Fsparse\u002FColBERT embedder,\n  `BGEM3FlagModel`.\n- `decoder-only-base`: decoder-only LLM embedder, `FlagLLMModel`.\n- `decoder-only-icl`: few-shot ICL LLM embedder, `FlagICLModel`.\n- `decoder-only-pseudo_moe`: pseudo-MoE LLM embedder,\n  `FlagPseudoMoEModel`.\n\nReranker `model_class` ids:\n\n- `encoder-only-base`: encoder cross-encoder reranker, `FlagReranker`.\n- `decoder-only-base`: decoder-only LLM reranker, `FlagLLMReranker`.\n- `decoder-only-layerwise`: layer-selectable LLM reranker,\n  `LayerWiseFlagLLMReranker`.\n- `decoder-only-lightweight`: token-compression LLM reranker,\n  `LightWeightFlagLLMReranker`.\n\n## Loader Choice\n\nUse auto loaders when the checkpoint's effective model name is in the built-in\nmapping. The effective name is the basename of `model_name_or_path`; if the\nbasename starts with `checkpoint-`, the parent directory basename is used. This\nmeans a fine-tuned checkpoint directory such as `...\u002Fmy-bge\u002Fcheckpoint-1000`\nis mapped as `my-bge`, not `checkpoint-1000`.\n\nIf auto mapping raises a model-not-found error, do not keep retrying with the\nsame call. Pick the correct `model_class`, and for embedders also set\n`pooling_method`, `query_instruction_format`, and `trust_remote_code` rather\nthan relying on mapping defaults. Custom decoder-only and remote-code\ncheckpoints are especially likely to need explicit `trust_remote_code=True`\nafter code review.\n\nUse concrete classes directly when the architecture is already known or when\nyou need class-specific options such as M3 return modes, ICL examples,\npseudo-MoE domains, layerwise cutoff layers, or lightweight compression.\n\n## Core Parameters\n\nUse [references\u002Fapi-reference.md](references\u002Fapi-reference.md) for verified\nsignatures and parameter defaults. The parameters most likely to affect\ninference behavior are:\n\n- `normalize_embeddings`: normalize dense embeddings before similarity.\n- `use_fp16` and `use_bf16`: inference dtype. Use CPU with full precision for\n  conservative smoke checks.\n- `query_instruction_for_retrieval` and `query_instruction_format`: query-side\n  retrieval instructions for embedders.\n- `devices`: `None` auto-selects available accelerators then CPU; a string such\n  as `\"cpu\"` or `\"cuda:0\"` pins one device; a list enables multi-process\n  inference across target devices.\n- `pooling_method`: use `cls` or `mean` for compatible encoder-only models;\n  decoder-only embedders require `last_token`.\n- `trust_remote_code`: keep `False` unless the checkpoint requires custom model\n  code and that code has been reviewed.\n- `truncate_dim`: truncate output vectors for Matryoshka-style embeddings.\n- `batch_size`, `query_max_length`, and `passage_max_length`: embedder\n  throughput and truncation controls.\n- Reranker `query_max_length`, `max_length`, and `normalize`: pair scoring\n  truncation and optional sigmoid normalization.\n- Layerwise\u002Flightweight reranker `cutoff_layers`, `compress_ratio`, and\n  `compress_layers`: score layer selection and token compression controls.\n\n## Expected Outputs\n\nBase embedders return a NumPy array by default. A single input returns a 1-D\nvector; a list returns a 2-D array shaped like `(num_texts, embedding_dim)`.\nWith `convert_to_numpy=False`, expect a torch tensor instead.\n\n`BGEM3FlagModel` returns a dictionary with `dense_vecs`, `lexical_weights`, and\n`colbert_vecs`. Keys whose return flag is false are present with `None` values.\nUse `dense_vecs @ dense_vecs.T` for dense similarity, use\n`compute_lexical_matching_score(...)` for sparse lexical weights, and use\n`compute_score(...)` for per-pair M3 mode scores.\n\nStandard rerankers return a list of float scores for a batch. Layerwise and\nlightweight rerankers return one score list when a single cutoff layer is used,\nor a list of score lists when multiple cutoff layers are requested. Do not sort\ndocuments until this shape is validated.\n\n## Workflows\n\nUse [references\u002Fworkflows.md](references\u002Fworkflows.md) for these recipes:\n\n- Auto and concrete embedder loading.\n- Query\u002Fcorpus encoding and similarity computation.\n- M3 dense, sparse, ColBERT, and `compute_score` workflows.\n- Decoder-only, ICL, and pseudo-MoE embedder options.\n- Encoder, decoder, layerwise, and lightweight reranker workflows.\n- Batch sizing, CPU fallback, and multi-device inference.\n\nUse [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md) when:\n\n- Auto mapping cannot find a model.\n- Remote-code or local custom-code loading fails.\n- Network\u002Fcache behavior is surprising.\n- CPU\u002FGPU fallback, dtype, or OOM behavior changes results or fails.\n- Output shapes or score modes are misused.\n\n## Smoke Checks\n\nRun the bundled helper before writing a larger pipeline:\n\n```bash\npython sub-skills\u002Finference\u002Fscripts\u002Fsmoke_inference_api.py\n```\n\nThe default command imports public inference symbols and prints signatures only.\nIt does not instantiate a model and should not trigger downloads. From the\n`flag-embedding` skill directory, load and encode only when an explicit local\npath or model id is supplied:\n\n```bash\npython sub-skills\u002Finference\u002Fscripts\u002Fsmoke_inference_api.py \\\n  --model-kind embedder \\\n  --model-name MODEL_NAME_OR_PATH \\\n  --devices cpu \\\n  --batch-size 2\n```\n\nFor a custom local checkpoint, add `--model-class`, `--pooling-method`, and\n`--query-instruction-format` as needed. For a reranker, use\n`--model-kind reranker`; for BGE-M3-specific probes, use `--model-kind m3`.\n","# Inference\n\nUse this sub-skill when the task is to load a FlagEmbedding model for inference,\nencode queries or corpus passages, compute embedding similarity, compute BGE-M3\ndense\u002Fsparse\u002FColBERT scores, rerank query-passage pairs, or smoke-check the\ninference API before a retrieval run.\n\nDo not use this sub-skill for fine-tuning data formats, fine-tuning launchers,\ntraining hyperparameters, or checkpoint production. Route those tasks to the\nsibling fine-tuning sub-skill. Do not use it for benchmark suites, leaderboard\nevaluation, or metric scripts. Route those tasks to the sibling evaluation\nsub-skill. For package installation, import failures, and global optional\ndependency repair, route to the root troubleshooting reference at\n`..\u002F..\u002Freferences\u002Ftroubleshooting.md`.\n\n## Start Here\n\n1. Decide whether the model is an embedder, an M3 embedder, or a reranker.\n2. Prefer `FlagAutoModel.from_finetuned(...)` for embedders and\n   `FlagAutoReranker.from_finetuned(...)` for rerankers when the checkpoint name\n   is in the built-in mapping.\n3. For a custom local checkpoint or an unmapped remote checkpoint, pass\n   `model_class` explicitly and choose the concrete class behavior from\n   [references\u002Fapi-reference.md](references\u002Fapi-reference.md).\n4. For retrieval, use `encode_queries(...)` for queries and\n   `encode_corpus(...)` for passages. Query instructions are applied by\n   `encode_queries`; passages normally should not receive the query instruction.\n5. For reranking, pass a pair as `(\"query\", \"passage\")` or a batch as a list\n   of pairs to `compute_score(...)`; validate the returned list or nested list\n   shape before sorting results.\n6. For a safe local check, run `scripts\u002Fsmoke_inference_api.py --help` first.\n   Running the script with no `--model-name` performs import and signature\n   probing only. Loading or encoding happens only when `--model-name` is passed.\n\n## Public Entry Points\n\nEmbedder auto loader and concrete classes:\n\n- `FlagAutoModel`\n- `FlagModel`\n- `BGEM3FlagModel`\n- `FlagLLMModel`\n- `FlagICLModel`\n- `FlagPseudoMoEModel`\n\nReranker auto loader and concrete classes:\n\n- `FlagAutoReranker`\n- `FlagReranker`\n- `FlagLLMReranker`\n- `LayerWiseFlagLLMReranker`\n- `LightWeightFlagLLMReranker`\n\nEmbedder `model_class` ids:\n\n- `encoder-only-base`: encoder-only dense embedder, normally `FlagModel`.\n- `encoder-only-m3`: BGE-M3 dense\u002Fsparse\u002FColBERT embedder,\n  `BGEM3FlagModel`.\n- `decoder-only-base`: decoder-only LLM embedder, `FlagLLMModel`.\n- `decoder-only-icl`: few-shot ICL LLM embedder, `FlagICLModel`.\n- `decoder-only-pseudo_moe`: pseudo-MoE LLM embedder,\n  `FlagPseudoMoEModel`.\n\nReranker `model_class` ids:\n\n- `encoder-only-base`: encoder cross-encoder reranker, `FlagReranker`.\n- `decoder-only-base`: decoder-only LLM reranker, `FlagLLMReranker`.\n- `decoder-only-layerwise`: layer-selectable LLM reranker,\n  `LayerWiseFlagLLMReranker`.\n- `decoder-only-lightweight`: token-compression LLM reranker,\n  `LightWeightFlagLLMReranker`.\n\n## Loader Choice\n\nUse auto loaders when the checkpoint's effective model name is in the built-in\nmapping. The effective name is the basename of `model_name_or_path`; if the\nbasename starts with `checkpoint-`, the parent directory basename is used. This\nmeans a fine-tuned checkpoint directory such as `...\u002Fmy-bge\u002Fcheckpoint-1000`\nis mapped as `my-bge`, not `checkpoint-1000`.\n\nIf auto mapping raises a model-not-found error, do not keep retrying with the\nsame call. Pick the correct `model_class`, and for embedders also set\n`pooling_method`, `query_instruction_format`, and `trust_remote_code` rather\nthan relying on mapping defaults. Custom decoder-only and remote-code\ncheckpoints are especially likely to need explicit `trust_remote_code=True`\nafter code review.\n\nUse concrete classes directly when the architecture is already known or when\nyou need class-specific options such as M3 return modes, ICL examples,\npseudo-MoE domains, layerwise cutoff layers, or lightweight compression.\n\n## Core Parameters\n\nUse [references\u002Fapi-reference.md](references\u002Fapi-reference.md) for verified\nsignatures and parameter defaults. The parameters most likely to affect\ninference behavior are:\n\n- `normalize_embeddings`: normalize dense embeddings before similarity.\n- `use_fp16` and `use_bf16`: inference dtype. Use CPU with full precision for\n  conservative smoke checks.\n- `query_instruction_for_retrieval` and `query_instruction_format`: query-side\n  retrieval instructions for embedders.\n- `devices`: `None` auto-selects available accelerators then CPU; a string such\n  as `\"cpu\"` or `\"cuda:0\"` pins one device; a list enables multi-process\n  inference across target devices.\n- `pooling_method`: use `cls` or `mean` for compatible encoder-only models;\n  decoder-only embedders require `last_token`.\n- `trust_remote_code`: keep `False` unless the checkpoint requires custom model\n  code and that code has been reviewed.\n- `truncate_dim`: truncate output vectors for Matryoshka-style embeddings.\n- `batch_size`, `query_max_length`, and `passage_max_length`: embedder\n  throughput and truncation controls.\n- Reranker `query_max_length`, `max_length`, and `normalize`: pair scoring\n  truncation and optional sigmoid normalization.\n- Layerwise\u002Flightweight reranker `cutoff_layers`, `compress_ratio`, and\n  `compress_layers`: score layer selection and token compression controls.\n\n## Expected Outputs\n\nBase embedders return a NumPy array by default. A single input returns a 1-D\nvector; a list returns a 2-D array shaped like `(num_texts, embedding_dim)`.\nWith `convert_to_numpy=False`, expect a torch tensor instead.\n\n`BGEM3FlagModel` returns a dictionary with `dense_vecs`, `lexical_weights`, and\n`colbert_vecs`. Keys whose return flag is false are present with `None` values.\nUse `dense_vecs @ dense_vecs.T` for dense similarity, use\n`compute_lexical_matching_score(...)` for sparse lexical weights, and use\n`compute_score(...)` for per-pair M3 mode scores.\n\nStandard rerankers return a list of float scores for a batch. Layerwise and\nlightweight rerankers return one score list when a single cutoff layer is used,\nor a list of score lists when multiple cutoff layers are requested. Do not sort\ndocuments until this shape is validated.\n\n## Workflows\n\nUse [references\u002Fworkflows.md](references\u002Fworkflows.md) for these recipes:\n\n- Auto and concrete embedder loading.\n- Query\u002Fcorpus encoding and similarity computation.\n- M3 dense, sparse, ColBERT, and `compute_score` workflows.\n- Decoder-only, ICL, and pseudo-MoE embedder options.\n- Encoder, decoder, layerwise, and lightweight reranker workflows.\n- Batch sizing, CPU fallback, and multi-device inference.\n\nUse [references\u002Ftroubleshooting.md](references\u002Ftroubleshooting.md) when:\n\n- Auto mapping cannot find a model.\n- Remote-code or local custom-code loading fails.\n- Network\u002Fcache behavior is surprising.\n- CPU\u002FGPU fallback, dtype, or OOM behavior changes results or fails.\n- Output shapes or score modes are misused.\n\n## Smoke Checks\n\nRun the bundled helper before writing a larger pipeline:\n\n```bash\npython sub-skills\u002Finference\u002Fscripts\u002Fsmoke_inference_api.py\n```\n\nThe default command imports public inference symbols and prints signatures only.\nIt does not instantiate a model and should not trigger downloads. From the\n`flag-embedding` skill directory, load and encode only when an explicit local\npath or model id is supplied:\n\n```bash\npython sub-skills\u002Finference\u002Fscripts\u002Fsmoke_inference_api.py \\\n  --model-kind embedder \\\n  --model-name MODEL_NAME_OR_PATH \\\n  --devices cpu \\\n  --batch-size 2\n```\n\nFor a custom local checkpoint, add `--model-class`, `--pooling-method`, and\n`--query-instruction-format` as needed. For a reranker, use\n`--model-kind reranker`; for BGE-M3-specific probes, use `--model-kind m3`.\n",{"name":5,"license":14,"metadata":23,"description":7,"disable-model-invocation":25},{"disco-role":24},"operating",true,"main","ac3fe1afa80fb9a09775ecfb2b6cc3ba850a2db6","skills\u002Frepositories\u002Frepo-skills\u002Fflag-embedding\u002Fsub-skills\u002Finference",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 inference",{"tool":39,"command":40},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill inference",{"tool":42,"command":43},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill inference","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fflag-embedding\u002Fsub-skills\u002Finference\u002FSKILL.md",[46,52,56,60],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fapi-reference.md","REFERENCE","参考文档",10414,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fflag-embedding\u002Fsub-skills\u002Finference\u002Freferences\u002Fapi-reference.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Ftroubleshooting.md",6788,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fflag-embedding\u002Fsub-skills\u002Finference\u002Freferences\u002Ftroubleshooting.md",{"path":57,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Fworkflows.md",13793,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fflag-embedding\u002Fsub-skills\u002Finference\u002Freferences\u002Fworkflows.md",{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"scripts\u002Fsmoke_inference_api.py","SCRIPT","脚本",14471,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fflag-embedding\u002Fsub-skills\u002Finference\u002Fscripts\u002Fsmoke_inference_api.py",[67,75,82,89,96,103],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":71,"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.","Apache-2.0","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":71,"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":71,"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":71,"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":71,"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":71,"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",[],[]]