[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7A7463FD59EPW0DDMB3EMWQ":3,"skill-same:01M7A7463FD59EPW0DDMB3EMWQ":138},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":36,"commitSha":37,"skillDirectory":38,"forksCount":39,"repositoryHtmlUrl":40,"repositoryHomepageUrl":41,"ownerHtmlUrl":42,"ownerBlogUrl":43,"ownerDescription":44,"updatedAt":45,"install":46,"resources":59},"01M7A7463FD59EPW0DDMB3EMWQ","tao-port-huggingface-model",null,"Integrate a HuggingFace Computer Vision model into the NVIDIA TAO Toolkit ecosystem (tao-core config, tao-pytorch trainer, tao-deploy TensorRT pipeline). Use when the user asks to \"integrate a HuggingFace model into TAO\", \"add an HF model to TAO Toolkit\", \"wire a HuggingFace ViT\u002FDETR\u002F SegFormer into tao-pytorch\", \"build a TAO trainer + deploy pipeline for an HF CV model\", or pastes a HuggingFace model URL\u002FID and wants it turned into a TAO model. Covers the full 7-phase loop: prerequisites check, HuggingFace inspection and validation, codebase exploration, tao-core configuration and native trainer implementation, ONNX export plus TensorRT deploy integration, packaging and L0 testing, container-based end-to-end validation, and (conditional) accuracy\u002Flatency tuning. Supports classification, object detection, semantic \u002F instance \u002F panoptic segmentation, zero-shot detection, and depth estimation.","themusebook\u002Fskills","themusebook","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F349927247?v=4","01M7A5C1RV5S3V5FF6ZACVDNRN","musebook","01M7A5C1SFC3KRZAXNN60XJ7KM",0,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: tao-port-huggingface-model\ndescription: >\n  Integrate a HuggingFace Computer Vision model into the NVIDIA TAO Toolkit\n  ecosystem (tao-core config, tao-pytorch trainer, tao-deploy TensorRT\n  pipeline). Use when the user asks to \"integrate a HuggingFace model into\n  TAO\", \"add an HF model to TAO Toolkit\", \"wire a HuggingFace ViT\u002FDETR\u002F\n  SegFormer into tao-pytorch\", \"build a TAO trainer + deploy pipeline for an\n  HF CV model\", or pastes a HuggingFace model URL\u002FID and wants it turned\n  into a TAO model. Covers the full 7-phase loop: prerequisites check,\n  HuggingFace inspection and validation, codebase exploration, tao-core\n  configuration and native trainer implementation, ONNX export plus TensorRT\n  deploy integration, packaging and L0 testing, container-based end-to-end\n  validation, and (conditional) accuracy\u002Flatency tuning. Supports\n  classification, object detection, semantic \u002F instance \u002F panoptic\n  segmentation, zero-shot detection, and depth estimation.\nlicense: Apache-2.0\ncompatibility: Requires Python 3.10+, NVIDIA driver, CUDA 13.0+, docker + nvidia-container-toolkit, an NGC API key (`docker login nvcr.io`), an HF_TOKEN, and access to the TAO Toolkit container images on `nvcr.io` for `tao-pytorch`, `tao-deploy`, and (optionally) `tao-dataservices` — Phase 0 asks the user for the exact image references and prepares them locally as `tao-pytorch-base:latest`, `tao-deploy-base:latest`, `tao-dataservices-base:latest`. Local clones of `tao-core`, `tao-pytorch`, `tao-deploy`,...\nmetadata:\n  author: NVIDIA Corporation\n  version: \"0.1.0\"\nallowed-tools: Read Bash Write Edit Grep Glob\ntags:\n- tao\n- huggingface\n- integration\n- computer-vision\n- deploy\n---\n\u003C!--\nCopyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.\n\nLicensed under the Apache License, Version 2.0 (the \"License\");\nyou may not use this file except in compliance with the License.\nYou may obtain a copy of the License at\n\n    http:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0\n\nUnless required by applicable law or agreed to in writing, software\ndistributed under the License is distributed on an \"AS IS\" BASIS,\nWITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\nSee the License for the specific language governing permissions and\nlimitations under the License.\n-->\n\n\n# TAO-HF Integration Skill\n\nIntegrate a HuggingFace (HF) Computer Vision model into the NVIDIA TAO Toolkit ecosystem. Work the phases iteratively — not purely linearly — via a **build → test → debug → fix → retest** loop at every step: when something fails, diagnose and fix before moving on; when it passes, move to the next step.\n\nThis SKILL.md is the workflow coordinator. Each phase has a dedicated `references\u002Fphase-N-*.md` with the full step-by-step content, code, docker invocations, and gates. Read the matching reference at the start of each phase — the summaries below are not sufficient.\n\n---\n\n## Local-Only Rule\n\nAll work is strictly local. Do NOT push\u002Fcommit\u002Fbranch on any remote (GitLab, GitHub, HuggingFace), create merge\u002Fpull requests or issues, or upload\u002Fpublish Docker images to any registry or artifact store. You may only read\u002Fclone from remotes — all edits, Docker builds, and test runs stay on the local machine.\n\n---\n\n## Submodule Override Strategy\n\nThe user clones the four TAO repos (`tao-core`, `tao-pytorch`, `tao-deploy`, `tao-dataservices`) independently into one working directory. The `tao-core\u002F` submodule nested inside each repo points to the **original unmodified commit**; modifications only exist in the top-level `tao-core\u002F`. **Always install from the top-level `tao-core\u002F`, never `\u003Crepo>\u002Ftao-core\u002F`** — the nested submodule silently ignores all modifications. Override rules: (1) mount the working directory `-v $(pwd):\u002Fworkspace`; (2) `pip install \u002Fworkspace\u002Ftao-core` FIRST, before tao-pytorch\u002Ftao-deploy; (3) PYTHONPATH top-level tao-core first, e.g. `-e PYTHONPATH=\u002Fworkspace\u002Ftao-core:\u002Fworkspace\u002Ftao-pytorch`. See `references\u002Fcross-cutting.md` for the directory tree.\n\n---\n\n## Execution platform\n\nEvery test, smoke run, and end-to-end validation executes inside a locally prepared TAO Toolkit container (`tao-pytorch-base:latest`, `tao-deploy-base:latest`, optionally `tao-dataservices-base:latest` — all from Phase 0). The platform skills own *how* to run them; this skill specifies *what*. **Default platform:** `local-docker`. Phase 0 delegates the driver \u002F CUDA \u002F NCT preflight to `tao-setup-nvidia-gpu-host`. See `references\u002Fcross-cutting.md` for the authoritative-skill table, bind-mount rationale, and canonical docker-run flag set.\n\n---\n\n## Phase Map\n\n| Phase | Goal | Reference |\n|---|---|---|\n| 0 | Prerequisites + TAO Toolkit images + local image tags | [phase-0-prereqs.md](references\u002Fphase-0-prereqs.md) |\n| 1 | Inputs, HF-inspection container, validate model + dataset | [phase-1-inspection.md](references\u002Fphase-1-inspection.md), [hf-inspection.md](references\u002Fhf-inspection.md) |\n| 2 | Closest existing TAO reference model | [phase-2-codebase.md](references\u002Fphase-2-codebase.md), [task-type-guide.md](references\u002Ftask-type-guide.md) |\n| 3 | tao-core config + tao-pytorch trainer \u002F eval \u002F inference | [phase-3-implementation.md](references\u002Fphase-3-implementation.md), [tao-patterns.md](references\u002Ftao-patterns.md), [repo-structure.md](references\u002Frepo-structure.md) |\n| 4 | ONNX export + tao-deploy TRT engine \u002F inference \u002F eval | [phase-4-deploy.md](references\u002Fphase-4-deploy.md) |\n| 5 | Packaging (`console_scripts`) + L0 tests | [phase-5-packaging.md](references\u002Fphase-5-packaging.md) |\n| 6 | Container testing + end-to-end validation | [phase-6-container-tests.md](references\u002Fphase-6-container-tests.md), [docker-patterns.md](references\u002Fdocker-patterns.md) |\n| 7 | (conditional) Accuracy \u002F latency \u002F size tuning | [phase-7-optimization.md](references\u002Fphase-7-optimization.md) |\n\nCross-cutting refs: [workflow-consistency.md](references\u002Fworkflow-consistency.md) (CLI flow, config field paths, cross-phase dependencies); [cross-cutting.md](references\u002Fcross-cutting.md) (platform, isolation, module pitfalls, debugging).\n\n**IMPORTANT — Continuous Execution Through Phase 6:** do NOT stop after Phases 3–5 to wait for the user to run tests. Phase 6 is mandatory — not complete until tests pass inside the containers and the end-to-end pipeline is validated.\n\n---\n\n## Development Loop\n\nAt every step: write code → test immediately (import check, unit test, or dry-run) → if it fails, read traceback → diagnose → fix → retest; if it passes, move on. Do NOT accumulate untested code — testing only at the end compounds bugs.\n\n---\n\n## Debugging Playbook\n\nWhen something fails, consult the symptom → likely-cause → fix table in `references\u002Fcross-cutting.md` before trying random fixes — it covers `ModuleNotFoundError`, `BACKBONE_REGISTRY` `KeyError`, shape mismatch, NaN loss, ONNX\u002FTRT build failures, TRT-vs-PyTorch accuracy gaps, OOM, DDP hangs, checkpoint load failures, and stale-submodule config issues.\n\n---\n\n## Environment Isolation Strategy\n\nAll Python work runs **inside Docker containers** — no host venvs, no `pip install`s into host Python (the host needs only Docker, from `tao-setup-nvidia-gpu-host`). Three contexts: A (Phase 1 HF inspection in `tao-hf-inspect`, `python:3.12-slim` fallback), B (Phase 3\u002F4\u002F6 smoke\u002FL0\u002Fe2e in the prepared container, source via `pip install \u002Fworkspace\u002Ftao-core && python setup.py develop`), C (host-bind-mount scratch). See `references\u002Fcross-cutting.md` for the contexts in full and the four numbered rules verbatim (`--check-only` host packages; Phase 1 `--user $(id -u):$(id -g)` vs. root; `HOME=\u002Fworkspace`\u002F`PIP_USER=1` fallback; distro package-manager list; `root:root` trade-off; `tao-hf-inspect` cleanup).\n\n---\n\n## Phase 0 — Prerequisites Check\n\n**Goal:** verify Python 3.10+ and `git`; delegate the driver \u002F CUDA \u002F Docker \u002F NVIDIA Container Toolkit host check to `tao-setup-nvidia-gpu-host`; verify NGC `docker login` for `nvcr.io`. Then **ask the user** for the TAO Toolkit image references (tao-pytorch, tao-deploy, optionally tao-dataservices), pull, and prepare local tags `tao-pytorch-base:latest`, `tao-deploy-base:latest`, `tao-dataservices-base:latest` for later phases — preparation removes the pre-installed released TAO packages so the user's `\u002Fworkspace\u002F...` clones install\u002Fload via `pip install \u002Fworkspace\u002Ftao-core && python setup.py develop`. **Hard stop** on any failed check. Required user inputs: the image references + credentials (NGC login, `HF_TOKEN`). Full commands, prompt wording, and per-image `Dockerfile` snippets: the Phase 0 reference.\n\n**Gate:** all prerequisite checks pass; the user supplied the required image references; `tao-pytorch-base:latest` and `tao-deploy-base:latest` exist locally; `tao-dataservices-base:latest` exists if dataservices work is anticipated.\n\n---\n\n## Phase 1 — Information Gathering & Validation\n\n**Goal:** decide whether to proceed at all. Gather credentials, locate\u002Fclone the four TAO repos, create a consistent working branch, launch the `tao-hf-inspect` container (Context A), validate the HF model is CV with a supported `pipeline_tag`, extract config + state-dict schema, sanity-check ONNX export, clean up. Full steps: Phase 1 references.\n\n**Reject if:** `pipeline_tag` is NLP \u002F audio \u002F LLM (non-CV); `AutoConfig` raises; or ONNX export fundamentally cannot work (no rewrite path).\n\n**Gate:** all 4 TAO repos located\u002Fcloned with a consistent branch; `pipeline_tag` confirmed CV; `model_type`, `image_size`, `hidden_size`, `num_labels` extracted; state-dict keys documented + HF→TAO remapping plan drafted; ONNX export sanity check passed (or failure understood); user confirmed `model_short_name` + task type. (Full checklist: [phase-1-inspection.md](references\u002Fphase-1-inspection.md).) Present findings and get user confirmation first.\n\n---\n\n## Phase 2 — Codebase Exploration\n\n**Goal:** find the closest existing TAO reference model for the detected `pipeline_tag`, read its implementation across `tao-core` \u002F `tao-pytorch` \u002F `tao-deploy`, and decide whether the backbone exists in `backbone_v2\u002F` or is new.\n\nThe HF `pipeline_tag` → TAO reference model mapping (classification → `classification_pyt`, detection → `dino`\u002F`rtdetr`, segmentation → `segformer`, instance → `mask2former`, panoptic → `oneformer`, zero-shot → `grounding_dino`, depth → `mono_depth`) drives **everything downstream** (config, architecture, loss, ONNX shape, TRT builder, deploy classes, metrics, dataset format). See the Phase 2 references for the full reference list (12 files per model), the `backbone_v2\u002F` and `tao-dataservices` coverage checks, and per-task architecture.\n\nIf a new backbone is needed, decide the strategy (timm wrap > re-implement > HF black-box wrap) before Phase 3 — it changes weight loading, ONNX export, deploy. **Never dual-inherit from `transformers.PreTrainedModel` and `BackboneBase`** (metaclass conflict — compose instead).\n\n**Gate:** reference TAO model identified + all 12 reference locations read; task-type implications understood (architecture, loss, ONNX outputs, deploy classes, metrics, dataset); backbone coverage decided (reuse \u002F wrap timm \u002F new); dataservices coverage checked. (Full checklist: [phase-2-codebase.md](references\u002Fphase-2-codebase.md).)\n\n---\n\n## Phase 3 — TAO Core Configuration & Native Implementation\n\n**Goal:** write the tao-core config schema + the tao-pytorch trainer \u002F native inference \u002F evaluation, smoke-testing between steps. (`\u003Cmodel_name>` = `snake_case` short-name; `\u003CModelName>` = `PascalCase`.)\n\nSteps 1–7 (each builds on the previous, smoke-test between): tao-core config (1), tao-pytorch trainer (2), multi-GPU\u002Fmulti-node (3), native inference → `result.csv` (4), native evaluation → `results.json` (5), MLOps for training and eval\u002Finfer → `status.json` (6–7). The `ExperimentConfig(CommonExperimentConfig)` must contain `model`, `dataset`, `train`, `evaluate`, `inference`, `export`, `gen_trt_engine`, `quantize`. All `???` fields are `MISSING` (user supplies via YAML\u002FCLI); the `augmentation.mean`\u002F`std`, `model.head.in_channels`, checkpoint-name, and `onnx_file` matches are in the checklist below.\n\nFull per-step bodies, code, the canonical `experiment_spec.yaml`, and smoke-test commands: the Phase 3 references.\n\n**Gates:** Step 1 — `ExperimentConfig` imports cleanly in-container; Step 2 — `build_model(cfg)` runs + PLModel instantiates in-container; Phase 3 — all 7 steps complete, smoke tests pass, no missing `__init__.py`.\n\n---\n\n## Phase 4 — Export, Deployment & TensorRT Integration\n\n**Goal:** ONNX export from tao-pytorch, then TRT engine builder + inference + evaluation in tao-deploy reusing the tao-core `ExperimentConfig`.\n\nSteps 8–11: ONNX exporter (8 — task-specific input\u002Foutput names, `batch_size=-1` ⇒ dynamic batch); TRT engine builder (9 — subclass `EngineBuilder` or reuse `ClassificationEngineBuilder`; write `specs\u002F{gen_trt_engine,inference,evaluate}.yaml`, same `ExperimentConfig` schema, `augmentation.mean`\u002F`std` MUST match training); TRT inference → `result.csv` (10); TRT eval → `results.json` (11). See the Phase 4 reference for full code and the Phase 3+4 gate (3 in-container checks: imports, model build + forward, ONNX round-trip).\n\n**Module pitfalls:** tao-pytorch and tao-deploy have **separate** `hydra_runner` and `monitor_status` — use the deploy versions in deploy scripts. `ExperimentConfig` comes from `nvidia_tao_core` in both (same schema\u002Ffield paths).\n\n**Phase 3+4 gate:** all three in-container checks pass (`tao-pytorch` imports + model + ONNX export; `tao-deploy` imports).\n\n---\n\n## Phase 5 — Packaging & L0 Testing\n\n**Goal:** register the model as a console_script in both repos and add unit tests.\n\nSteps 12–15: register `'\u003Cmodel_name>=...:main'` in `console_scripts` of `tao-pytorch\u002Fsetup.py` (12) and `tao-deploy\u002Fsetup.py` (13, creating the deploy `entrypoint\u002F\u003Cmodel_name>.py` via `entrypoint_hydra`); deploy L0 tests (14); trainer L0 tests — `Trainer(..., fast_dev_run=True)` + `@pytest.mark.cv_unit @pytest.mark.\u003Cmodel_name>` (15). See the Phase 5 reference for exact entry-point strings, code, and L0 test file lists.\n\n**Gate:** entrypoints registered; pytest files exist and follow the marker convention. **Do NOT stop — go directly to Phase 6.**\n\n---\n\n## Cross-Phase Data Flow & Consistency Verification\n\nBefore Docker testing, verify the chain `train → export → gen_trt_engine → inference \u002F evaluate` (the `*_model_latest.pth` → `.onnx` → `.engine` artifact flow + the config fields each stage reads\u002Fwrites — full diagram in `references\u002Fcross-cutting.md`).\n\nConsistency checklist (verify before proceeding): `self.checkpoint_filename` → the `*_latest.pth` name `evaluate.checkpoint` \u002F `export.checkpoint` reference; `augmentation.mean`\u002F`std` identical across training spec, `inference.yaml`, `evaluate.yaml`, engine-builder `preprocess_mode`; ONNX `input_names=['input']` \u002F `output_names=['output']` (detection\u002Finstance-seg use task-specific names); `export.input_width`\u002F`input_height` match `dataset.img_size`; `model.head.in_channels` matches `model_params_mapping.py`; `classes.txt` at `dataset.root_dir` readable by both repos; all `__init__.py` exist (incl. `scripts\u002F__init__.py` for `get_subtasks()` via `pkgutil`). Full paths: [workflow-consistency.md](references\u002Fworkflow-consistency.md).\n\n---\n\n## Phase 6 — Container Testing & End-to-End Validation\n\n**Mandatory — start immediately after Phase 5.** All TAO models ship as Docker images; code that only works outside a container is incomplete. Testing runs **directly inside the TAO Toolkit container** — no image build in the loop: mount → install source (`setup.py develop`) → run `pytest` \u002F `pylint` \u002F `pydocstyle` \u002F `flake8` directly. Use vanilla commands, NOT the `ci\u002Frun_functional_tests.py` \u002F `ci\u002Frun_static_tests.py` wrappers (internal-mirror-only; public `github.com\u002FNVIDIA-TAO\u002F` mirrors have no `ci\u002F` dir).\n\nSteps 16–25: verify local image tags exist (16); unit tests for tao-core \u002F tao-pytorch (`-m cv_unit`, `--shm-size=16G`) \u002F tao-deploy (17–19); lint (20); wheels (21); end-to-end — train dry-run + export in **one** tao-pytorch session, then gen_trt_engine + inference + evaluate in **one** tao-deploy session (same session critical — `--rm` discards installs) (22); cross-check native vs TRT (23); debug shells (24); optional release images (25).\n\nFull commands (every `docker run`, per-container env-vars, exact pytest \u002F lint invocations + full-suite variants, the train\u002Fexport\u002Fgen_trt_engine\u002Finference\u002Fevaluate one-liner with all CLI overrides, the `ci\u002F` note, the fix-and-retest loop) and build scripts \u002F runner patterns: see the Phase 6 references.\n\n**Phase 6 gate (Done criteria):** tao-core \u002F tao-pytorch \u002F tao-deploy unit tests pass in their containers; static tests pass (or only legacy lint warnings); wheels build; end-to-end `\u003Cmodel_name>_model_latest.pth` → `model.onnx` → `model.engine` → non-empty `result.csv` + `results.json`; native vs TRT agree within tolerance.\n\n---\n\n## Phase 7 — Optimization & Tuning (conditional)\n\nEnter only if Phase 6 passes but accuracy \u002F latency \u002F size needs improvement. **Ask the user for target metrics first.**\n\nDiagnostic categories: accuracy too low; TRT-vs-native gap; training too slow; inference too slow. Techniques: Step 27 — hyperparameter tuning; Step 28 — INT8 quantization (PTQ via torchao \u002F modelopt, TRT INT8 + calibration); Step 29 — channel pruning + retrain; Step 30 — knowledge distillation; Step 31 — resolution tuning (TAO interpolates ViT positional embeddings automatically). See the Phase 7 reference for each category's checks, config blocks, YAML overrides, decision tree, and rationale.\n\n---\n\n## Argument\n\n`$ARGUMENTS`\n\nIf provided, interpret `$ARGUMENTS` as the HuggingFace model ID or URL to start Phase 1. If credentials or model short-name are not included, ask the user for them before proceeding.\n","\u003C!--\nCopyright (c) 2026, NVIDIA CORPORATION.  All rights reserved.\n\nLicensed under the Apache License, Version 2.0 (the \"License\");\nyou may not use this file except in compliance with the License.\nYou may obtain a copy of the License at\n\n    http:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0\n\nUnless required by applicable law or agreed to in writing, software\ndistributed under the License is distributed on an \"AS IS\" BASIS,\nWITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\nSee the License for the specific language governing permissions and\nlimitations under the License.\n-->\n\n\n# TAO-HF Integration Skill\n\nIntegrate a HuggingFace (HF) Computer Vision model into the NVIDIA TAO Toolkit ecosystem. Work the phases iteratively — not purely linearly — via a **build → test → debug → fix → retest** loop at every step: when something fails, diagnose and fix before moving on; when it passes, move to the next step.\n\nThis SKILL.md is the workflow coordinator. Each phase has a dedicated `references\u002Fphase-N-*.md` with the full step-by-step content, code, docker invocations, and gates. Read the matching reference at the start of each phase — the summaries below are not sufficient.\n\n---\n\n## Local-Only Rule\n\nAll work is strictly local. Do NOT push\u002Fcommit\u002Fbranch on any remote (GitLab, GitHub, HuggingFace), create merge\u002Fpull requests or issues, or upload\u002Fpublish Docker images to any registry or artifact store. You may only read\u002Fclone from remotes — all edits, Docker builds, and test runs stay on the local machine.\n\n---\n\n## Submodule Override Strategy\n\nThe user clones the four TAO repos (`tao-core`, `tao-pytorch`, `tao-deploy`, `tao-dataservices`) independently into one working directory. The `tao-core\u002F` submodule nested inside each repo points to the **original unmodified commit**; modifications only exist in the top-level `tao-core\u002F`. **Always install from the top-level `tao-core\u002F`, never `\u003Crepo>\u002Ftao-core\u002F`** — the nested submodule silently ignores all modifications. Override rules: (1) mount the working directory `-v $(pwd):\u002Fworkspace`; (2) `pip install \u002Fworkspace\u002Ftao-core` FIRST, before tao-pytorch\u002Ftao-deploy; (3) PYTHONPATH top-level tao-core first, e.g. `-e PYTHONPATH=\u002Fworkspace\u002Ftao-core:\u002Fworkspace\u002Ftao-pytorch`. See `references\u002Fcross-cutting.md` for the directory tree.\n\n---\n\n## Execution platform\n\nEvery test, smoke run, and end-to-end validation executes inside a locally prepared TAO Toolkit container (`tao-pytorch-base:latest`, `tao-deploy-base:latest`, optionally `tao-dataservices-base:latest` — all from Phase 0). The platform skills own *how* to run them; this skill specifies *what*. **Default platform:** `local-docker`. Phase 0 delegates the driver \u002F CUDA \u002F NCT preflight to `tao-setup-nvidia-gpu-host`. See `references\u002Fcross-cutting.md` for the authoritative-skill table, bind-mount rationale, and canonical docker-run flag set.\n\n---\n\n## Phase Map\n\n| Phase | Goal | Reference |\n|---|---|---|\n| 0 | Prerequisites + TAO Toolkit images + local image tags | [phase-0-prereqs.md](references\u002Fphase-0-prereqs.md) |\n| 1 | Inputs, HF-inspection container, validate model + dataset | [phase-1-inspection.md](references\u002Fphase-1-inspection.md), [hf-inspection.md](references\u002Fhf-inspection.md) |\n| 2 | Closest existing TAO reference model | [phase-2-codebase.md](references\u002Fphase-2-codebase.md), [task-type-guide.md](references\u002Ftask-type-guide.md) |\n| 3 | tao-core config + tao-pytorch trainer \u002F eval \u002F inference | [phase-3-implementation.md](references\u002Fphase-3-implementation.md), [tao-patterns.md](references\u002Ftao-patterns.md), [repo-structure.md](references\u002Frepo-structure.md) |\n| 4 | ONNX export + tao-deploy TRT engine \u002F inference \u002F eval | [phase-4-deploy.md](references\u002Fphase-4-deploy.md) |\n| 5 | Packaging (`console_scripts`) + L0 tests | [phase-5-packaging.md](references\u002Fphase-5-packaging.md) |\n| 6 | Container testing + end-to-end validation | [phase-6-container-tests.md](references\u002Fphase-6-container-tests.md), [docker-patterns.md](references\u002Fdocker-patterns.md) |\n| 7 | (conditional) Accuracy \u002F latency \u002F size tuning | [phase-7-optimization.md](references\u002Fphase-7-optimization.md) |\n\nCross-cutting refs: [workflow-consistency.md](references\u002Fworkflow-consistency.md) (CLI flow, config field paths, cross-phase dependencies); [cross-cutting.md](references\u002Fcross-cutting.md) (platform, isolation, module pitfalls, debugging).\n\n**IMPORTANT — Continuous Execution Through Phase 6:** do NOT stop after Phases 3–5 to wait for the user to run tests. Phase 6 is mandatory — not complete until tests pass inside the containers and the end-to-end pipeline is validated.\n\n---\n\n## Development Loop\n\nAt every step: write code → test immediately (import check, unit test, or dry-run) → if it fails, read traceback → diagnose → fix → retest; if it passes, move on. Do NOT accumulate untested code — testing only at the end compounds bugs.\n\n---\n\n## Debugging Playbook\n\nWhen something fails, consult the symptom → likely-cause → fix table in `references\u002Fcross-cutting.md` before trying random fixes — it covers `ModuleNotFoundError`, `BACKBONE_REGISTRY` `KeyError`, shape mismatch, NaN loss, ONNX\u002FTRT build failures, TRT-vs-PyTorch accuracy gaps, OOM, DDP hangs, checkpoint load failures, and stale-submodule config issues.\n\n---\n\n## Environment Isolation Strategy\n\nAll Python work runs **inside Docker containers** — no host venvs, no `pip install`s into host Python (the host needs only Docker, from `tao-setup-nvidia-gpu-host`). Three contexts: A (Phase 1 HF inspection in `tao-hf-inspect`, `python:3.12-slim` fallback), B (Phase 3\u002F4\u002F6 smoke\u002FL0\u002Fe2e in the prepared container, source via `pip install \u002Fworkspace\u002Ftao-core && python setup.py develop`), C (host-bind-mount scratch). See `references\u002Fcross-cutting.md` for the contexts in full and the four numbered rules verbatim (`--check-only` host packages; Phase 1 `--user $(id -u):$(id -g)` vs. root; `HOME=\u002Fworkspace`\u002F`PIP_USER=1` fallback; distro package-manager list; `root:root` trade-off; `tao-hf-inspect` cleanup).\n\n---\n\n## Phase 0 — Prerequisites Check\n\n**Goal:** verify Python 3.10+ and `git`; delegate the driver \u002F CUDA \u002F Docker \u002F NVIDIA Container Toolkit host check to `tao-setup-nvidia-gpu-host`; verify NGC `docker login` for `nvcr.io`. Then **ask the user** for the TAO Toolkit image references (tao-pytorch, tao-deploy, optionally tao-dataservices), pull, and prepare local tags `tao-pytorch-base:latest`, `tao-deploy-base:latest`, `tao-dataservices-base:latest` for later phases — preparation removes the pre-installed released TAO packages so the user's `\u002Fworkspace\u002F...` clones install\u002Fload via `pip install \u002Fworkspace\u002Ftao-core && python setup.py develop`. **Hard stop** on any failed check. Required user inputs: the image references + credentials (NGC login, `HF_TOKEN`). Full commands, prompt wording, and per-image `Dockerfile` snippets: the Phase 0 reference.\n\n**Gate:** all prerequisite checks pass; the user supplied the required image references; `tao-pytorch-base:latest` and `tao-deploy-base:latest` exist locally; `tao-dataservices-base:latest` exists if dataservices work is anticipated.\n\n---\n\n## Phase 1 — Information Gathering & Validation\n\n**Goal:** decide whether to proceed at all. Gather credentials, locate\u002Fclone the four TAO repos, create a consistent working branch, launch the `tao-hf-inspect` container (Context A), validate the HF model is CV with a supported `pipeline_tag`, extract config + state-dict schema, sanity-check ONNX export, clean up. Full steps: Phase 1 references.\n\n**Reject if:** `pipeline_tag` is NLP \u002F audio \u002F LLM (non-CV); `AutoConfig` raises; or ONNX export fundamentally cannot work (no rewrite path).\n\n**Gate:** all 4 TAO repos located\u002Fcloned with a consistent branch; `pipeline_tag` confirmed CV; `model_type`, `image_size`, `hidden_size`, `num_labels` extracted; state-dict keys documented + HF→TAO remapping plan drafted; ONNX export sanity check passed (or failure understood); user confirmed `model_short_name` + task type. (Full checklist: [phase-1-inspection.md](references\u002Fphase-1-inspection.md).) Present findings and get user confirmation first.\n\n---\n\n## Phase 2 — Codebase Exploration\n\n**Goal:** find the closest existing TAO reference model for the detected `pipeline_tag`, read its implementation across `tao-core` \u002F `tao-pytorch` \u002F `tao-deploy`, and decide whether the backbone exists in `backbone_v2\u002F` or is new.\n\nThe HF `pipeline_tag` → TAO reference model mapping (classification → `classification_pyt`, detection → `dino`\u002F`rtdetr`, segmentation → `segformer`, instance → `mask2former`, panoptic → `oneformer`, zero-shot → `grounding_dino`, depth → `mono_depth`) drives **everything downstream** (config, architecture, loss, ONNX shape, TRT builder, deploy classes, metrics, dataset format). See the Phase 2 references for the full reference list (12 files per model), the `backbone_v2\u002F` and `tao-dataservices` coverage checks, and per-task architecture.\n\nIf a new backbone is needed, decide the strategy (timm wrap > re-implement > HF black-box wrap) before Phase 3 — it changes weight loading, ONNX export, deploy. **Never dual-inherit from `transformers.PreTrainedModel` and `BackboneBase`** (metaclass conflict — compose instead).\n\n**Gate:** reference TAO model identified + all 12 reference locations read; task-type implications understood (architecture, loss, ONNX outputs, deploy classes, metrics, dataset); backbone coverage decided (reuse \u002F wrap timm \u002F new); dataservices coverage checked. (Full checklist: [phase-2-codebase.md](references\u002Fphase-2-codebase.md).)\n\n---\n\n## Phase 3 — TAO Core Configuration & Native Implementation\n\n**Goal:** write the tao-core config schema + the tao-pytorch trainer \u002F native inference \u002F evaluation, smoke-testing between steps. (`\u003Cmodel_name>` = `snake_case` short-name; `\u003CModelName>` = `PascalCase`.)\n\nSteps 1–7 (each builds on the previous, smoke-test between): tao-core config (1), tao-pytorch trainer (2), multi-GPU\u002Fmulti-node (3), native inference → `result.csv` (4), native evaluation → `results.json` (5), MLOps for training and eval\u002Finfer → `status.json` (6–7). The `ExperimentConfig(CommonExperimentConfig)` must contain `model`, `dataset`, `train`, `evaluate`, `inference`, `export`, `gen_trt_engine`, `quantize`. All `???` fields are `MISSING` (user supplies via YAML\u002FCLI); the `augmentation.mean`\u002F`std`, `model.head.in_channels`, checkpoint-name, and `onnx_file` matches are in the checklist below.\n\nFull per-step bodies, code, the canonical `experiment_spec.yaml`, and smoke-test commands: the Phase 3 references.\n\n**Gates:** Step 1 — `ExperimentConfig` imports cleanly in-container; Step 2 — `build_model(cfg)` runs + PLModel instantiates in-container; Phase 3 — all 7 steps complete, smoke tests pass, no missing `__init__.py`.\n\n---\n\n## Phase 4 — Export, Deployment & TensorRT Integration\n\n**Goal:** ONNX export from tao-pytorch, then TRT engine builder + inference + evaluation in tao-deploy reusing the tao-core `ExperimentConfig`.\n\nSteps 8–11: ONNX exporter (8 — task-specific input\u002Foutput names, `batch_size=-1` ⇒ dynamic batch); TRT engine builder (9 — subclass `EngineBuilder` or reuse `ClassificationEngineBuilder`; write `specs\u002F{gen_trt_engine,inference,evaluate}.yaml`, same `ExperimentConfig` schema, `augmentation.mean`\u002F`std` MUST match training); TRT inference → `result.csv` (10); TRT eval → `results.json` (11). See the Phase 4 reference for full code and the Phase 3+4 gate (3 in-container checks: imports, model build + forward, ONNX round-trip).\n\n**Module pitfalls:** tao-pytorch and tao-deploy have **separate** `hydra_runner` and `monitor_status` — use the deploy versions in deploy scripts. `ExperimentConfig` comes from `nvidia_tao_core` in both (same schema\u002Ffield paths).\n\n**Phase 3+4 gate:** all three in-container checks pass (`tao-pytorch` imports + model + ONNX export; `tao-deploy` imports).\n\n---\n\n## Phase 5 — Packaging & L0 Testing\n\n**Goal:** register the model as a console_script in both repos and add unit tests.\n\nSteps 12–15: register `'\u003Cmodel_name>=...:main'` in `console_scripts` of `tao-pytorch\u002Fsetup.py` (12) and `tao-deploy\u002Fsetup.py` (13, creating the deploy `entrypoint\u002F\u003Cmodel_name>.py` via `entrypoint_hydra`); deploy L0 tests (14); trainer L0 tests — `Trainer(..., fast_dev_run=True)` + `@pytest.mark.cv_unit @pytest.mark.\u003Cmodel_name>` (15). See the Phase 5 reference for exact entry-point strings, code, and L0 test file lists.\n\n**Gate:** entrypoints registered; pytest files exist and follow the marker convention. **Do NOT stop — go directly to Phase 6.**\n\n---\n\n## Cross-Phase Data Flow & Consistency Verification\n\nBefore Docker testing, verify the chain `train → export → gen_trt_engine → inference \u002F evaluate` (the `*_model_latest.pth` → `.onnx` → `.engine` artifact flow + the config fields each stage reads\u002Fwrites — full diagram in `references\u002Fcross-cutting.md`).\n\nConsistency checklist (verify before proceeding): `self.checkpoint_filename` → the `*_latest.pth` name `evaluate.checkpoint` \u002F `export.checkpoint` reference; `augmentation.mean`\u002F`std` identical across training spec, `inference.yaml`, `evaluate.yaml`, engine-builder `preprocess_mode`; ONNX `input_names=['input']` \u002F `output_names=['output']` (detection\u002Finstance-seg use task-specific names); `export.input_width`\u002F`input_height` match `dataset.img_size`; `model.head.in_channels` matches `model_params_mapping.py`; `classes.txt` at `dataset.root_dir` readable by both repos; all `__init__.py` exist (incl. `scripts\u002F__init__.py` for `get_subtasks()` via `pkgutil`). Full paths: [workflow-consistency.md](references\u002Fworkflow-consistency.md).\n\n---\n\n## Phase 6 — Container Testing & End-to-End Validation\n\n**Mandatory — start immediately after Phase 5.** All TAO models ship as Docker images; code that only works outside a container is incomplete. Testing runs **directly inside the TAO Toolkit container** — no image build in the loop: mount → install source (`setup.py develop`) → run `pytest` \u002F `pylint` \u002F `pydocstyle` \u002F `flake8` directly. Use vanilla commands, NOT the `ci\u002Frun_functional_tests.py` \u002F `ci\u002Frun_static_tests.py` wrappers (internal-mirror-only; public `github.com\u002FNVIDIA-TAO\u002F` mirrors have no `ci\u002F` dir).\n\nSteps 16–25: verify local image tags exist (16); unit tests for tao-core \u002F tao-pytorch (`-m cv_unit`, `--shm-size=16G`) \u002F tao-deploy (17–19); lint (20); wheels (21); end-to-end — train dry-run + export in **one** tao-pytorch session, then gen_trt_engine + inference + evaluate in **one** tao-deploy session (same session critical — `--rm` discards installs) (22); cross-check native vs TRT (23); debug shells (24); optional release images (25).\n\nFull commands (every `docker run`, per-container env-vars, exact pytest \u002F lint invocations + full-suite variants, the train\u002Fexport\u002Fgen_trt_engine\u002Finference\u002Fevaluate one-liner with all CLI overrides, the `ci\u002F` note, the fix-and-retest loop) and build scripts \u002F runner patterns: see the Phase 6 references.\n\n**Phase 6 gate (Done criteria):** tao-core \u002F tao-pytorch \u002F tao-deploy unit tests pass in their containers; static tests pass (or only legacy lint warnings); wheels build; end-to-end `\u003Cmodel_name>_model_latest.pth` → `model.onnx` → `model.engine` → non-empty `result.csv` + `results.json`; native vs TRT agree within tolerance.\n\n---\n\n## Phase 7 — Optimization & Tuning (conditional)\n\nEnter only if Phase 6 passes but accuracy \u002F latency \u002F size needs improvement. **Ask the user for target metrics first.**\n\nDiagnostic categories: accuracy too low; TRT-vs-native gap; training too slow; inference too slow. Techniques: Step 27 — hyperparameter tuning; Step 28 — INT8 quantization (PTQ via torchao \u002F modelopt, TRT INT8 + calibration); Step 29 — channel pruning + retrain; Step 30 — knowledge distillation; Step 31 — resolution tuning (TAO interpolates ViT positional embeddings automatically). See the Phase 7 reference for each category's checks, config blocks, YAML overrides, decision tree, and rationale.\n\n---\n\n## Argument\n\n`$ARGUMENTS`\n\nIf provided, interpret `$ARGUMENTS` as the HuggingFace model ID or URL to start Phase 1. If credentials or model short-name are not included, ask the user for them before proceeding.\n",{"name":5,"tags":24,"license":15,"metadata":30,"description":33,"allowed-tools":34,"compatibility":35},[25,26,27,28,29],"tao","huggingface","integration","computer-vision","deploy",{"author":31,"version":32},"NVIDIA Corporation","0.1.0","Integrate a HuggingFace Computer Vision model into the NVIDIA TAO Toolkit ecosystem (tao-core config, tao-pytorch trainer, tao-deploy TensorRT pipeline). Use when the user asks to \"integrate a HuggingFace model into TAO\", \"add an HF model to TAO Toolkit\", \"wire a HuggingFace ViT\u002FDETR\u002F SegFormer into tao-pytorch\", \"build a TAO trainer + deploy pipeline for an HF CV model\", or pastes a HuggingFace model URL\u002FID and wants it turned into a TAO model. Covers the full 7-phase loop: prerequisites check, HuggingFace inspection and validation, codebase exploration, tao-core configuration and native trainer implementation, ONNX export plus TensorRT deploy integration, packaging and L0 testing, container-based end-to-end validation, and (conditional) accuracy\u002Flatency tuning. Supports classification, object detection, semantic \u002F instance \u002F panoptic segmentation, zero-shot detection, and depth estimation.\n","Read Bash Write Edit Grep Glob","Requires Python 3.10+, NVIDIA driver, CUDA 13.0+, docker + nvidia-container-toolkit, an NGC API key (`docker login nvcr.io`), an HF_TOKEN, and access to the TAO Toolkit container images on `nvcr.io` for `tao-pytorch`, `tao-deploy`, and (optionally) `tao-dataservices` — Phase 0 asks the user for the exact image references and prepares them locally as `tao-pytorch-base:latest`, `tao-deploy-base:latest`, `tao-dataservices-base:latest`. Local clones of `tao-core`, `tao-pytorch`, `tao-deploy`,...","main","1c523aba5cdf867a79f078847075492d7778092a","public\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model",1,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills","https:\u002F\u002Fskills.x402.wtf","https:\u002F\u002Fgithub.com\u002Fthemusebook","https:\u002F\u002Fmusebook.trade","The official Github of musebook","2026-11-02 23:42:26",{"source":8,"skillName":5,"commands":47,"prompt":57,"zipAvailable":58},[48,51,54],{"tool":49,"command":50},"npx","npx skills add themusebook\u002Fskills --skill tao-port-huggingface-model",{"tool":52,"command":53},"bunx","bunx skills add themusebook\u002Fskills --skill tao-port-huggingface-model",{"tool":55,"command":56},"pnpm","pnpm dlx skills add themusebook\u002Fskills --skill tao-port-huggingface-model","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002FSKILL.md",true,[60,66,70,74,78,82,86,90,94,98,102,106,110,114,118,122,126,130,134],{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"BENCHMARK.md","REFERENCE","参考文档",4026,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002FBENCHMARK.md",{"path":67,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":68,"sourceHtmlUrl":69},"metadata.json",1140,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Fmetadata.json",{"path":71,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":72,"sourceHtmlUrl":73},"references\u002Fcross-cutting.md",10415,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fcross-cutting.md",{"path":75,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":76,"sourceHtmlUrl":77},"references\u002Fdocker-patterns.md",16237,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fdocker-patterns.md",{"path":79,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":80,"sourceHtmlUrl":81},"references\u002Fhf-inspection.md",6320,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fhf-inspection.md",{"path":83,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":84,"sourceHtmlUrl":85},"references\u002Fphase-0-prereqs.md",9944,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-0-prereqs.md",{"path":87,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":88,"sourceHtmlUrl":89},"references\u002Fphase-1-inspection.md",10245,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-1-inspection.md",{"path":91,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":92,"sourceHtmlUrl":93},"references\u002Fphase-2-codebase.md",7469,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-2-codebase.md",{"path":95,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":96,"sourceHtmlUrl":97},"references\u002Fphase-3-implementation.md",35497,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-3-implementation.md",{"path":99,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":100,"sourceHtmlUrl":101},"references\u002Fphase-4-deploy.md",13025,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-4-deploy.md",{"path":103,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":104,"sourceHtmlUrl":105},"references\u002Fphase-5-packaging.md",3420,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-5-packaging.md",{"path":107,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":108,"sourceHtmlUrl":109},"references\u002Fphase-6-container-tests.md",15742,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-6-container-tests.md",{"path":111,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":112,"sourceHtmlUrl":113},"references\u002Fphase-7-optimization.md",7972,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fphase-7-optimization.md",{"path":115,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":116,"sourceHtmlUrl":117},"references\u002Frepo-structure.md",10817,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Frepo-structure.md",{"path":119,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":120,"sourceHtmlUrl":121},"references\u002Ftao-patterns.md",20574,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Ftao-patterns.md",{"path":123,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":124,"sourceHtmlUrl":125},"references\u002Ftask-type-guide.md",16829,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Ftask-type-guide.md",{"path":127,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":128,"sourceHtmlUrl":129},"references\u002Fworkflow-consistency.md",28153,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Freferences\u002Fworkflow-consistency.md",{"path":131,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":132,"sourceHtmlUrl":133},"skill-card.md",4406,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Fskill-card.md",{"path":135,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":136,"sourceHtmlUrl":137},"verification.json",4744,"https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fpublic\u002Fapi\u002Fskills\u002Fnvidia\u002Ftao-port-huggingface-model\u002Fverification.json",[139,148,155,162,169,176],{"skillNo":140,"name":141,"nameZh":6,"description":142,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":143,"licenseFileUrl":144,"sourceHtmlUrl":145,"categories":146,"keywords":147,"publishedAt":19},"01M7AB34SPKVSDRRG009PM1CZH","youtube-clipper","YouTube 视频智能剪辑工具。下载视频和字幕，AI 分析生成精细章节（几分钟级别）， 用户选择片段后自动剪辑、翻译字幕为中英双语、烧录字幕到视频，并生成总结文案。 使用场景：当用户需要剪辑 YouTube 视频、生成短视频片段、制作双语字幕版本时。 关键词：视频剪辑、YouTube、字幕翻译、双语字幕、视频下载、clip video","MIT","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fyoutube-clipper\u002FSKILL.md",[],[],{"skillNo":149,"name":150,"nameZh":6,"description":151,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":143,"licenseFileUrl":6,"sourceHtmlUrl":152,"categories":153,"keywords":154,"publishedAt":19},"01M7AB18BH6N7KR7939HRNWVMQ","viem-integration","Integrate EVM blockchains using viem. Use when user says \"read blockchain data\", \"send transaction\", \"interact with smart contract\", \"connect to Ethereum\", \"use viem\", \"use wagmi\", \"wallet integration\", \"viem setup\", or mentions blockchain\u002FEVM development with TypeScript.","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fviem-integration\u002FSKILL.md",[],[],{"skillNo":156,"name":157,"nameZh":6,"description":158,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":143,"licenseFileUrl":6,"sourceHtmlUrl":159,"categories":160,"keywords":161,"publishedAt":19},"01M7AB12GHGP48WCNXDQQ9222H","v4-security-foundations","Security-first Uniswap v4 hook development. Use when user mentions \"v4 hooks\", \"hook security\", \"PoolManager\", \"beforeSwap\", \"afterSwap\", or asks about V4 hook best practices, vulnerabilities, or audit requirements.","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fv4-security-foundations\u002FSKILL.md",[],[],{"skillNo":163,"name":164,"nameZh":6,"description":165,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":143,"licenseFileUrl":6,"sourceHtmlUrl":166,"categories":167,"keywords":168,"publishedAt":19},"01M7AB0XM3PSR7V3Q2ZRR0Q0FB","v4-sdk-integration","App-layer SDK guide for building swap and liquidity experiences directly with the Uniswap v4 SDK. Use when user asks about \"v4 sdk\", \"uniswap v4\", \"v4 swap\", \"v4 liquidity\", \"PoolManager\", \"V4Planner\", \"StateView\", \"PositionManager\", \"pool state\", \"v4 position\", \"uniswap sdk\", or when building swap\u002Fliquidity UX directly with SDKs rather than via the Trading API.","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fv4-sdk-integration\u002FSKILL.md",[],[],{"skillNo":170,"name":171,"nameZh":6,"description":172,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":143,"licenseFileUrl":6,"sourceHtmlUrl":173,"categories":174,"keywords":175,"publishedAt":19},"01M7AB0SGA93ASA588DJW4P9WT","v4-hook-generator","Generate Uniswap v4 hook contracts via OpenZeppelin MCP. Use when building custom swap logic, async swaps, hook-owned liquidity, custom curves, dynamic fees, MEV protection, limit orders, or oracle hooks.","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fv4-hook-generator\u002FSKILL.md",[],[],{"skillNo":177,"name":178,"nameZh":6,"description":179,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":143,"licenseFileUrl":6,"sourceHtmlUrl":180,"categories":181,"keywords":182,"publishedAt":19},"01M7AAZYH75H4CS03WSW2EZ8DM","testing","For testing with Light Protocol programs and clients on localnet, devnet, and mainnet validation.","https:\u002F\u002Fgithub.com\u002Fthemusebook\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftesting\u002FSKILL.md",[],[]]