[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M78W9JTC9QRAMG6XGYCZEYA3":3,"skill-same:01M78W9JTC9QRAMG6XGYCZEYA3":71},{"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},"01M78W9JTC9QRAMG6XGYCZEYA3","inference-and-deployment",null,"Use DLTK's TensorFlow 1.x prediction, full-volume sliding-window assembly, crop averaging, metrics, and metadata-preserving medical-image deployment workflows.","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\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: inference-and-deployment\ndescription: \"Use DLTK's TensorFlow 1.x prediction, full-volume sliding-window\n  assembly, crop averaging, metrics, and metadata-preserving medical-image\n  deployment workflows.\"\ndisable-model-invocation: true\nmetadata:\n  disco-role: operating\nlicense: Apache 2.0\n---\n\n# DLTK inference and deployment\n\nUse this route when the task is to run prediction with an exported DLTK model, assemble patch predictions into a full volume, average stochastic crops, calculate segmentation metrics, or write a NIfTI result. This is an operating guide for DLTK 0.2.1's TensorFlow 1.x graph-execution deployment path.\n\n## Compatibility gate\n\n- Expect Python 3.7 and TensorFlow 1.15-era graph execution. The documented deployment path is `tensorflow.contrib.predictor.from_saved_model`; do **not** claim that a modern TF2 installation provides this predictor or that these recipes are TF2-compatible.\n- Confirm the model input and fetched output static shapes before running a volume. The sliding-window utility expects rank-compatible static shapes and a channels-last tensor. The input sample normally has `[batch, spatial..., channels]`; a full-volume application adds a dummy batch dimension.\n- Keep model construction and output-key details in [model-building](..\u002Fmodel-building\u002FSKILL.md), input\u002Ffile\u002Freader assumptions in [data-pipelines](..\u002Fdata-pipelines\u002FSKILL.md), and export\u002Ftraining setup in [training-and-estimators](..\u002Ftraining-and-estimators\u002FSKILL.md).\n\n## Choose the workflow\n\n1. From the directory containing this file (`inference-and-deployment\u002F`), run `python scripts\u002Fsliding_window_plan.py --help`, then provide spatial `--input-shape`, `--window-shape`, and `--output-shape`. The helper performs validation without importing TensorFlow, reading data, or contacting a network.\n2. From that same sub-skill directory, run `python scripts\u002Fsliding_window_smoke.py`. It exercises the native test shape plus output-smaller, overlapping, and batched assembly using constant synthetic predictions; it does not load a model or dataset.\n3. For a real export, read [deployment-workflows.md](references\u002Fdeployment-workflows.md) and [api-reference.md](references\u002Fapi-reference.md) first. Discover a numeric export directory, load the TF1 predictor, inspect its tensors, assemble a volume, convert probabilities\u002Flogits to the required result, calculate metrics if labels exist, and write output only to an explicitly chosen directory.\n\n## Inference checklist\n\n- Validate all spatial dimensions, the predictor's input\u002Foutput keys, channel count, `batch_size`, and explicit stride before session execution. A malformed output\u002Fwindow combination can fail in padding or produce empty slices and a zero division counter.\n- Treat `sliding_window_segmentation_inference` as an averaging assembler: it runs each patch, adds each output to the corresponding output slice, increments a separate coverage counter, and returns `sum \u002F counter` for each requested op. It returns a list in `ops_list` order, not class labels.\n- For segmentation, convert a `[batch, spatial..., classes]` probability or logit result with `np.argmax(result, axis=-1)`. For probabilities, averaging overlapping patches before argmax is the DLTK MRBrainS pattern. For logits, a numerically stable softmax is useful when probabilities are needed, but argmax is unchanged.\n- Exclude or report background explicitly when aggregating Dice. DLTK's `dice` returns `NaN` for a class with no predicted and no reference voxels; `np.nanmean(dice[1:])` is an application policy, not a universal score.\n- For SimpleITK output, create the array image with the correct spatial order, call `CopyInformation` from the source image, and only then write the requested `.nii`\u002F`.nii.gz`. A shape or direction\u002Fspacing mismatch must be diagnosed rather than silently discarded.\n- Random-crop regression\u002Fclassification deployment is stochastic. Record the crop count, shape, seed policy (if any), and aggregation axis; the original IXI examples average four random crops and explicitly warn that results vary between runs.\n\nSee [troubleshooting.md](references\u002Ftroubleshooting.md) before changing geometry, predictor access, export discovery, or metric handling.\n","# DLTK inference and deployment\n\nUse this route when the task is to run prediction with an exported DLTK model, assemble patch predictions into a full volume, average stochastic crops, calculate segmentation metrics, or write a NIfTI result. This is an operating guide for DLTK 0.2.1's TensorFlow 1.x graph-execution deployment path.\n\n## Compatibility gate\n\n- Expect Python 3.7 and TensorFlow 1.15-era graph execution. The documented deployment path is `tensorflow.contrib.predictor.from_saved_model`; do **not** claim that a modern TF2 installation provides this predictor or that these recipes are TF2-compatible.\n- Confirm the model input and fetched output static shapes before running a volume. The sliding-window utility expects rank-compatible static shapes and a channels-last tensor. The input sample normally has `[batch, spatial..., channels]`; a full-volume application adds a dummy batch dimension.\n- Keep model construction and output-key details in [model-building](..\u002Fmodel-building\u002FSKILL.md), input\u002Ffile\u002Freader assumptions in [data-pipelines](..\u002Fdata-pipelines\u002FSKILL.md), and export\u002Ftraining setup in [training-and-estimators](..\u002Ftraining-and-estimators\u002FSKILL.md).\n\n## Choose the workflow\n\n1. From the directory containing this file (`inference-and-deployment\u002F`), run `python scripts\u002Fsliding_window_plan.py --help`, then provide spatial `--input-shape`, `--window-shape`, and `--output-shape`. The helper performs validation without importing TensorFlow, reading data, or contacting a network.\n2. From that same sub-skill directory, run `python scripts\u002Fsliding_window_smoke.py`. It exercises the native test shape plus output-smaller, overlapping, and batched assembly using constant synthetic predictions; it does not load a model or dataset.\n3. For a real export, read [deployment-workflows.md](references\u002Fdeployment-workflows.md) and [api-reference.md](references\u002Fapi-reference.md) first. Discover a numeric export directory, load the TF1 predictor, inspect its tensors, assemble a volume, convert probabilities\u002Flogits to the required result, calculate metrics if labels exist, and write output only to an explicitly chosen directory.\n\n## Inference checklist\n\n- Validate all spatial dimensions, the predictor's input\u002Foutput keys, channel count, `batch_size`, and explicit stride before session execution. A malformed output\u002Fwindow combination can fail in padding or produce empty slices and a zero division counter.\n- Treat `sliding_window_segmentation_inference` as an averaging assembler: it runs each patch, adds each output to the corresponding output slice, increments a separate coverage counter, and returns `sum \u002F counter` for each requested op. It returns a list in `ops_list` order, not class labels.\n- For segmentation, convert a `[batch, spatial..., classes]` probability or logit result with `np.argmax(result, axis=-1)`. For probabilities, averaging overlapping patches before argmax is the DLTK MRBrainS pattern. For logits, a numerically stable softmax is useful when probabilities are needed, but argmax is unchanged.\n- Exclude or report background explicitly when aggregating Dice. DLTK's `dice` returns `NaN` for a class with no predicted and no reference voxels; `np.nanmean(dice[1:])` is an application policy, not a universal score.\n- For SimpleITK output, create the array image with the correct spatial order, call `CopyInformation` from the source image, and only then write the requested `.nii`\u002F`.nii.gz`. A shape or direction\u002Fspacing mismatch must be diagnosed rather than silently discarded.\n- Random-crop regression\u002Fclassification deployment is stochastic. Record the crop count, shape, seed policy (if any), and aggregation axis; the original IXI examples average four random crops and explicitly warn that results vary between runs.\n\nSee [troubleshooting.md](references\u002Ftroubleshooting.md) before changing geometry, predictor access, export discovery, or metric handling.\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\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment",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 inference-and-deployment",{"tool":40,"command":41},"bunx","bunx skills add VectorSpaceLab\u002FAREX-Skill --skill inference-and-deployment",{"tool":43,"command":44},"pnpm","pnpm dlx skills add VectorSpaceLab\u002FAREX-Skill --skill inference-and-deployment","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002FSKILL.md",[47,53,57,61,67],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fapi-reference.md","REFERENCE","参考文档",8459,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002Freferences\u002Fapi-reference.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fdeployment-workflows.md",7159,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002Freferences\u002Fdeployment-workflows.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Ftroubleshooting.md",6200,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002Freferences\u002Ftroubleshooting.md",{"path":62,"resourceType":63,"resourceTypeDesc":64,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Fsliding_window_plan.py","SCRIPT","脚本",7132,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002Fscripts\u002Fsliding_window_plan.py",{"path":68,"resourceType":63,"resourceTypeDesc":64,"sizeBytes":69,"sourceHtmlUrl":70},"scripts\u002Fsliding_window_smoke.py",10350,"https:\u002F\u002Fgithub.com\u002FVectorSpaceLab\u002FAREX-Skill\u002Fblob\u002Fmain\u002Fskills\u002Frepositories\u002Frepo-skills\u002Fdltk\u002Fsub-skills\u002Finference-and-deployment\u002Fscripts\u002Fsliding_window_smoke.py",[72,79,86,93,100,107],{"skillNo":73,"name":74,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"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":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"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":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"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":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"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":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"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":108,"name":109,"nameZh":6,"description":110,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":111,"categories":112,"keywords":113,"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",[],[]]