[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M792CXX2CYBMBF94B0CEP41P":3,"skill-same:01M792CXX2CYBMBF94B0CEP41P":75},{"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},"01M792CXX2CYBMBF94B0CEP41P","detection",null,"Use for PaddleViT object detection workflows with DETR, Swin, or PVTv2: validate COCO data, select configs, build\u002Ftrain\u002Fevaluate models, reason about transforms, losses, post-processing, and run safe utility smokes. Excludes segmentation and generic export; cross-link deployment-and-operations for shared runtime concerns.","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\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: detection\ndescription: \"Use for PaddleViT object detection workflows with DETR, Swin, or\n  PVTv2: validate COCO data, select configs, build\u002Ftrain\u002Fevaluate models, reason\n  about transforms, losses, post-processing, and run safe utility smokes.\n  Excludes segmentation and generic export; cross-link deployment-and-operations\n  for shared runtime concerns.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# PaddleViT object detection\n\nUse this route for standalone PaddleViT object detection under `object_detection\u002F`:\nDETR, Swin, and PVTv2 with COCO boxes. It covers family\u002Fconfig selection,\ndata preflight, transforms and targets, safe model-contract checks, checkpoint\ntriage, training\u002Fevaluation boundaries, and backend-aware diagnosis. It does\nnot own semantic segmentation, generic export\u002Finference, quantization, or\nweight porting; route those requests to\n[deployment-and-operations](..\u002Fdeployment-and-operations\u002FSKILL.md).\n\nThe bundled helpers are self-contained and do not import the original source\ncheckout, download data or weights, or modify a user-owned dataset. A source\ncheckout is optional and is needed only for a source-model build or native\ncandidate test. Details that are useful during a focused investigation live in\n[model overview](references\u002Fmodel-overview.md),\n[data formats](references\u002Fdata-formats.md),\n[workflows](references\u002Fworkflows.md), and\n[troubleshooting](references\u002Ftroubleshooting.md).\n\n## Route the request\n\n1. Identify the family before changing a config or launching a command:\n   - **DETR** for query-based, end-to-end detection with transformer decoder\n     queries, Hungarian matching, and direct box post-processing.\n   - **Swin** or **PVTv2** for hierarchical backbones feeding FPN, RPN, and\n     RoI-style heads with anchors, proposals, and NMS.\n2. If the request says only “PaddleViT detection,” ask which family and\n   whether the goal is a utility check, source build, evaluation, or training.\n   Do not mix family directories in one Python process: common modules such as\n   `config`, `coco`, `box_ops`, and `utils` can resolve incorrectly.\n3. Keep segmentation separate, even if a request mentions masks or a DETR\n   import. For export, predictor, AMP, distributed, or shared runtime issues,\n   use [deployment-and-operations](..\u002Fdeployment-and-operations\u002FSKILL.md).\n\n## Preflight before a source run\n\nValidate the **COCO root**, not a split directory. The expected root contains\n`annotations\u002Finstances_{train,val}2017.json` and the matching\n`train2017\u002F` or `val2017\u002F` image directory. Run the read-only helper from the\nskill root (replace placeholders with caller-owned values):\n\n```bash\npython \u003Cskill-root>\u002Fscripts\u002Fcheck_coco_layout.py \u003Ccoco-root> --split val\n```\n\nAdd `--check-images` to decode referenced images and compare dimensions, or\n`--check-api` to parse the annotation through `pycocotools`. Use `--json` for a\nmachine-readable report and `--demo` for a temporary one-image validator\nfixture. A failed check is a stop condition: repair the dataset outside this\nskill and rerun; do not download or silently substitute another dataset.\n\nBefore importing a source model, run the bounded synthetic contract smoke:\n\n```bash\npython \u003Cskill-root>\u002Fscripts\u002Fdetection_model_smoke.py --model all --device cpu\n```\n\nThis checks tiny DETR and four-level anchor-family shapes and finite values. It\nis not a source-model build, checkpoint test, COCO test, mAP result, or\nbenchmark reproduction. Use `--device gpu:0` only when a GPU claim is required\nand the backend has already passed its environment probe.\n\n## Select a source root and config\n\nThe three projects are standalone. For a source-backed run, enter exactly one\nfamily directory and expose only that directory (and its intended parent) to\nimports:\n\n```bash\ncd \u003Csource-root>\u002Fobject_detection\u002FDETR   # or Swin or PVTv2\nexport PYTHONPATH=\"$PWD:$PWD\u002F..:${PYTHONPATH:-}\"\n```\n\nResolve configuration in this order: Python defaults, recursive YAML `BASE`\nfiles, then CLI overrides. Print and inspect the effective config before model\nconstruction. Check backbone output channels against FPN inputs, pyramid\nstrides and anchor levels, class\u002Fcategory conventions, image divisibility, and\nDETR embedding-dimension\u002Fhead divisibility. Swin\u002FPVTv2 configs preserve the\nhistorical `ROI.NUM_ClASSES` spelling; do not replace it with a guessed key.\nThe family-specific config and output\u002Ftarget details are in\n[model overview](references\u002Fmodel-overview.md) and\n[data formats](references\u002Fdata-formats.md).\n\n## CLI flag contract\n\nThe family `main_single_gpu.py` and `main_multi_gpu.py` scripts use these\nshort, single-dash options:\n\n```text\n-cfg PATH             YAML config (BASE files are recursively merged)\n-dataset coco         dataset selector\n-data_path PATH       COCO root, not train2017\u002Fval2017\n-batch_size N         per-process\u002Fper-GPU batch size\n-eval                 evaluation-only mode\n-pretrained PREFIX    source appends .pdparams\n-resume PREFIX        source expects .pdparams and .pdopt\n-last_epoch N         resume epoch metadata\n-ngpus N              configured multi-GPU worker count\n```\n\n`-cfg=...` and `-cfg ...` are both acceptable. Pass checkpoint **prefixes**\nwithout `.pdparams` when following the source loader. Inspect launcher shell\nfiles rather than executing them blindly: paths, visible devices, output\nprefixes, and training duration are caller-owned. A representative command\nshape is:\n\n```bash\ncd \u003Csource-root>\u002Fobject_detection\u002FDETR\nCUDA_VISIBLE_DEVICES=0 python main_single_gpu.py \\\n  -cfg=.\u002Fconfigs\u002Fdetr_resnet50.yaml -dataset=coco \\\n  -batch_size=1 -data_path=\u003Ccoco-root> -eval \\\n  -pretrained=\u003Ccheckpoint-prefix>\n```\n\nTreat real training\u002Fevaluation as expensive and data\u002Fcheckpoint\u002FGPU dependent;\nuse the [workflow reference](references\u002Fworkflows.md) for gated sequencing.\n\n## Family contracts to preserve\n\n- **DETR:** expect logits shaped `[B,Q,C+1]`, normalized center-size boxes\n  `[B,Q,4]`, optional auxiliary decoder outputs, and losses including\n  classification, L1, and GIoU. Post-processing needs target sizes in\n  `[height,width]` order and emits absolute `xyxy` boxes, scores, and labels.\n- **Swin\u002FPVTv2:** expect hierarchical features, FPN\u002FRPN\u002FRoI losses during\n  training, and post-NMS rows shaped `[label, score, xmin, ymin, xmax, ymax]`\n  during evaluation. Their target path uses absolute boxes and contiguous\n  classes; the two families share the broad head\u002Fneck contract but not every\n  backbone channel\u002Fconfig value.\n- **All families:** COCO boxes begin as `[x,y,width,height]`; transformations\n  must update geometry and area. A valid synthetic training fixture needs at\n  least one non-empty target. Preserve original image IDs for COCO results.\n  Full transform and output schemas are in the linked references.\n\n## Verification tiers\n\nChoose the cheapest tier that answers the question and record command, device,\nPaddle version, config, and result:\n\n1. `check_coco_layout.py` for root layout, JSON arrays, IDs, boxes, files, and\n   optional Pillow\u002FCOCO-API checks.\n2. `detection_model_smoke.py` for standalone tiny shape and finite-value\n   contracts; a non-32-divisible size intentionally reports the padding caveat.\n3. Native DETR box\u002Ftransform\u002Fmodel tests when a source checkout and dependencies\n   are available. They are evidence candidates, not runtime prerequisites;\n   some are skipped or require an untracked fixture.\n4. A one-batch source build\u002Fforward using one family root and a tiny local\n   fixture, only after config\u002Fimport checks pass.\n5. Real COCO evaluation or training only with explicit data, checkpoint, GPU,\n   and time approval. Never call a smoke or random-weight forward mAP evidence.\n\nCPU can validate parsing, layout, box utilities, and tiny diagnostics. It does\nnot establish CUDA, AMP, distributed, or full detector claims. For those\nboundaries use [deployment-and-operations](..\u002Fdeployment-and-operations\u002FSKILL.md).\n\n## Stop and recover\n\n- **Missing files\u002FAPI:** `-data_path` must be the root; verify exact split\n  filenames. If `pycocotools` is absent, omit `--check-api` but do not claim\n  COCO evaluation is verified.\n- **Import\u002FYAML errors:** start a fresh process, isolate one family root,\n  resolve `BASE` relative to its YAML, compare keys to that family's config,\n  and inspect the final config. Do not mix module paths or claim a build that\n  never constructed a model.\n- **Shape\u002Fcheckpoint errors:** verify family, backbone\u002FFPN channels, attention\n  divisibility, query\u002Fclass counts, head spelling, and checkpoint provenance.\n  A prefix normally maps to `.pdparams`; resume also needs `.pdopt`. Do not\n  reshape incompatible state dictionaries.\n- **Bad boxes\u002Fempty batches:** inspect crowd flags, category mapping, positive\n  area, coordinate format, resize\u002Fflip\u002Fcrop order, and image IDs. Keep one\n  valid target in a fixture; an all-empty batch is not a valid smoke.\n- **NaN\u002FInf or no detections:** run CPU geometry checks, verify normalized DETR\n  boxes or absolute Swin\u002FPVTv2 boxes, then inspect thresholds, NMS, and loaded\n  keys. Random weights cannot support an accuracy claim.\n- **CUDA OOM\u002Funavailable:** lower per-GPU batch size or input scale for OOM.\n  If CUDA is unavailable, report CPU-only partial evidence; never substitute it\n  for a required GPU result. For multi-GPU, check visible devices, `-ngpus`,\n  per-process batch semantics, NCCL, and rank gathering before retrying.\n\nSee [troubleshooting](references\u002Ftroubleshooting.md) for expanded recovery\nbranches. Do not run network, full-native, long-training, or multi-GPU\nworkflows as part of this route unless explicitly authorized.\n","# PaddleViT object detection\n\nUse this route for standalone PaddleViT object detection under `object_detection\u002F`:\nDETR, Swin, and PVTv2 with COCO boxes. It covers family\u002Fconfig selection,\ndata preflight, transforms and targets, safe model-contract checks, checkpoint\ntriage, training\u002Fevaluation boundaries, and backend-aware diagnosis. It does\nnot own semantic segmentation, generic export\u002Finference, quantization, or\nweight porting; route those requests to\n[deployment-and-operations](..\u002Fdeployment-and-operations\u002FSKILL.md).\n\nThe bundled helpers are self-contained and do not import the original source\ncheckout, download data or weights, or modify a user-owned dataset. A source\ncheckout is optional and is needed only for a source-model build or native\ncandidate test. Details that are useful during a focused investigation live in\n[model overview](references\u002Fmodel-overview.md),\n[data formats](references\u002Fdata-formats.md),\n[workflows](references\u002Fworkflows.md), and\n[troubleshooting](references\u002Ftroubleshooting.md).\n\n## Route the request\n\n1. Identify the family before changing a config or launching a command:\n   - **DETR** for query-based, end-to-end detection with transformer decoder\n     queries, Hungarian matching, and direct box post-processing.\n   - **Swin** or **PVTv2** for hierarchical backbones feeding FPN, RPN, and\n     RoI-style heads with anchors, proposals, and NMS.\n2. If the request says only “PaddleViT detection,” ask which family and\n   whether the goal is a utility check, source build, evaluation, or training.\n   Do not mix family directories in one Python process: common modules such as\n   `config`, `coco`, `box_ops`, and `utils` can resolve incorrectly.\n3. Keep segmentation separate, even if a request mentions masks or a DETR\n   import. For export, predictor, AMP, distributed, or shared runtime issues,\n   use [deployment-and-operations](..\u002Fdeployment-and-operations\u002FSKILL.md).\n\n## Preflight before a source run\n\nValidate the **COCO root**, not a split directory. The expected root contains\n`annotations\u002Finstances_{train,val}2017.json` and the matching\n`train2017\u002F` or `val2017\u002F` image directory. Run the read-only helper from the\nskill root (replace placeholders with caller-owned values):\n\n```bash\npython \u003Cskill-root>\u002Fscripts\u002Fcheck_coco_layout.py \u003Ccoco-root> --split val\n```\n\nAdd `--check-images` to decode referenced images and compare dimensions, or\n`--check-api` to parse the annotation through `pycocotools`. Use `--json` for a\nmachine-readable report and `--demo` for a temporary one-image validator\nfixture. A failed check is a stop condition: repair the dataset outside this\nskill and rerun; do not download or silently substitute another dataset.\n\nBefore importing a source model, run the bounded synthetic contract smoke:\n\n```bash\npython \u003Cskill-root>\u002Fscripts\u002Fdetection_model_smoke.py --model all --device cpu\n```\n\nThis checks tiny DETR and four-level anchor-family shapes and finite values. It\nis not a source-model build, checkpoint test, COCO test, mAP result, or\nbenchmark reproduction. Use `--device gpu:0` only when a GPU claim is required\nand the backend has already passed its environment probe.\n\n## Select a source root and config\n\nThe three projects are standalone. For a source-backed run, enter exactly one\nfamily directory and expose only that directory (and its intended parent) to\nimports:\n\n```bash\ncd \u003Csource-root>\u002Fobject_detection\u002FDETR   # or Swin or PVTv2\nexport PYTHONPATH=\"$PWD:$PWD\u002F..:${PYTHONPATH:-}\"\n```\n\nResolve configuration in this order: Python defaults, recursive YAML `BASE`\nfiles, then CLI overrides. Print and inspect the effective config before model\nconstruction. Check backbone output channels against FPN inputs, pyramid\nstrides and anchor levels, class\u002Fcategory conventions, image divisibility, and\nDETR embedding-dimension\u002Fhead divisibility. Swin\u002FPVTv2 configs preserve the\nhistorical `ROI.NUM_ClASSES` spelling; do not replace it with a guessed key.\nThe family-specific config and output\u002Ftarget details are in\n[model overview](references\u002Fmodel-overview.md) and\n[data formats](references\u002Fdata-formats.md).\n\n## CLI flag contract\n\nThe family `main_single_gpu.py` and `main_multi_gpu.py` scripts use these\nshort, single-dash options:\n\n```text\n-cfg PATH             YAML config (BASE files are recursively merged)\n-dataset coco         dataset selector\n-data_path PATH       COCO root, not train2017\u002Fval2017\n-batch_size N         per-process\u002Fper-GPU batch size\n-eval                 evaluation-only mode\n-pretrained PREFIX    source appends .pdparams\n-resume PREFIX        source expects .pdparams and .pdopt\n-last_epoch N         resume epoch metadata\n-ngpus N              configured multi-GPU worker count\n```\n\n`-cfg=...` and `-cfg ...` are both acceptable. Pass checkpoint **prefixes**\nwithout `.pdparams` when following the source loader. Inspect launcher shell\nfiles rather than executing them blindly: paths, visible devices, output\nprefixes, and training duration are caller-owned. A representative command\nshape is:\n\n```bash\ncd \u003Csource-root>\u002Fobject_detection\u002FDETR\nCUDA_VISIBLE_DEVICES=0 python main_single_gpu.py \\\n  -cfg=.\u002Fconfigs\u002Fdetr_resnet50.yaml -dataset=coco \\\n  -batch_size=1 -data_path=\u003Ccoco-root> -eval \\\n  -pretrained=\u003Ccheckpoint-prefix>\n```\n\nTreat real training\u002Fevaluation as expensive and data\u002Fcheckpoint\u002FGPU dependent;\nuse the [workflow reference](references\u002Fworkflows.md) for gated sequencing.\n\n## Family contracts to preserve\n\n- **DETR:** expect logits shaped `[B,Q,C+1]`, normalized center-size boxes\n  `[B,Q,4]`, optional auxiliary decoder outputs, and losses including\n  classification, L1, and GIoU. Post-processing needs target sizes in\n  `[height,width]` order and emits absolute `xyxy` boxes, scores, and labels.\n- **Swin\u002FPVTv2:** expect hierarchical features, FPN\u002FRPN\u002FRoI losses during\n  training, and post-NMS rows shaped `[label, score, xmin, ymin, xmax, ymax]`\n  during evaluation. Their target path uses absolute boxes and contiguous\n  classes; the two families share the broad head\u002Fneck contract but not every\n  backbone channel\u002Fconfig value.\n- **All families:** COCO boxes begin as `[x,y,width,height]`; transformations\n  must update geometry and area. A valid synthetic training fixture needs at\n  least one non-empty target. Preserve original image IDs for COCO results.\n  Full transform and output schemas are in the linked references.\n\n## Verification tiers\n\nChoose the cheapest tier that answers the question and record command, device,\nPaddle version, config, and result:\n\n1. `check_coco_layout.py` for root layout, JSON arrays, IDs, boxes, files, and\n   optional Pillow\u002FCOCO-API checks.\n2. `detection_model_smoke.py` for standalone tiny shape and finite-value\n   contracts; a non-32-divisible size intentionally reports the padding caveat.\n3. Native DETR box\u002Ftransform\u002Fmodel tests when a source checkout and dependencies\n   are available. They are evidence candidates, not runtime prerequisites;\n   some are skipped or require an untracked fixture.\n4. A one-batch source build\u002Fforward using one family root and a tiny local\n   fixture, only after config\u002Fimport checks pass.\n5. Real COCO evaluation or training only with explicit data, checkpoint, GPU,\n   and time approval. Never call a smoke or random-weight forward mAP evidence.\n\nCPU can validate parsing, layout, box utilities, and tiny diagnostics. It does\nnot establish CUDA, AMP, distributed, or full detector claims. For those\nboundaries use [deployment-and-operations](..\u002Fdeployment-and-operations\u002FSKILL.md).\n\n## Stop and recover\n\n- **Missing files\u002FAPI:** `-data_path` must be the root; verify exact split\n  filenames. If `pycocotools` is absent, omit `--check-api` but do not claim\n  COCO evaluation is verified.\n- **Import\u002FYAML errors:** start a fresh process, isolate one family root,\n  resolve `BASE` relative to its YAML, compare keys to that family's config,\n  and inspect the final config. Do not mix module paths or claim a build that\n  never constructed a model.\n- **Shape\u002Fcheckpoint errors:** verify family, backbone\u002FFPN channels, attention\n  divisibility, query\u002Fclass counts, head spelling, and checkpoint provenance.\n  A prefix normally maps to `.pdparams`; resume also needs `.pdopt`. Do not\n  reshape incompatible state dictionaries.\n- **Bad boxes\u002Fempty batches:** inspect crowd flags, category mapping, positive\n  area, coordinate format, resize\u002Fflip\u002Fcrop order, and image IDs. Keep one\n  valid target in a fixture; an all-empty batch is not a valid smoke.\n- **NaN\u002FInf or no detections:** run CPU geometry checks, verify normalized DETR\n  boxes or absolute Swin\u002FPVTv2 boxes, then inspect thresholds, NMS, and loaded\n  keys. Random weights cannot support an accuracy claim.\n- **CUDA OOM\u002Funavailable:** lower per-GPU batch size or input scale for OOM.\n  If CUDA is unavailable, report CPU-only partial evidence; never substitute it\n  for a required GPU result. For multi-GPU, check visible devices, `-ngpus`,\n  per-process batch semantics, NCCL, and rank gathering before retrying.\n\nSee [troubleshooting](references\u002Ftroubleshooting.md) for expanded recovery\nbranches. Do not run network, full-native, long-training, or multi-GPU\nworkflows as part of this route unless explicitly authorized.\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\u002Fpaddlevit\u002Fsub-skills\u002Fdetection",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 detection",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill detection",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill detection","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002FSKILL.md",[47,53,57,61,65,71],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fdata-formats.md","REFERENCE","参考文档",3881,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002Freferences\u002Fdata-formats.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fmodel-overview.md",4052,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002Freferences\u002Fmodel-overview.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Ftroubleshooting.md",4816,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002Freferences\u002Ftroubleshooting.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Fworkflows.md",5571,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002Freferences\u002Fworkflows.md",{"path":66,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Fcheck_coco_layout.py","SCRIPT","脚本",11884,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002Fscripts\u002Fcheck_coco_layout.py",{"path":72,"resourceType":67,"resourceTypeDesc":68,"sizeBytes":73,"sourceHtmlUrl":74},"scripts\u002Fdetection_model_smoke.py",9103,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fpaddlevit\u002Fsub-skills\u002Fdetection\u002Fscripts\u002Fdetection_model_smoke.py",[76,83,90,97,104,111],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"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.","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":14,"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":14,"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":14,"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":14,"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":14,"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",[],[]]