[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-visual-changenet":3,"mdc-havoyr-key":34,"related-repo-nvidia-tao-train-visual-changenet":2881,"related-org-nvidia-tao-train-visual-changenet":2983},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":32,"mdContent":33},"tao-train-visual-changenet","train Visual ChangeNet for defect detection","Visual ChangeNet for binary image classification and segmentation in AOI defect detection. Use when training, evaluating, exporting, or running inference for PCB defect detection or visual inspection, comparing image pairs for PASS\u002FNO_PASS classification, or producing change-segmentation masks. Trigger phrases include \"train Visual ChangeNet\", \"ChangeNet classify\", \"ChangeNet segment\", \"AOI defect detection\", \"PCB inspection model\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"Deep Learning","deep-learning","tag",{"name":17,"slug":18,"type":15},"Automation","automation",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:30:35.38844","Apache-2.0",281,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":31},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Ftao-train-visual-changenet","---\nname: tao-train-visual-changenet\ndescription: Visual ChangeNet for binary image classification and segmentation in AOI defect detection. Use when training,\n  evaluating, exporting, or running inference for PCB defect detection or visual inspection, comparing image pairs for\n  PASS\u002FNO_PASS classification, or producing change-segmentation masks. Trigger phrases include \"train Visual ChangeNet\",\n  \"ChangeNet classify\", \"ChangeNet segment\", \"AOI defect detection\", \"PCB inspection model\".\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit.\nmetadata:\n  author: NVIDIA Corporation\n  version: \"0.1.0\"\nallowed-tools: Read Bash\ntags:\n- pcb\n- aoi\n- defect\n- classification\n- segmentation\n- siamese\n- visual-inspection\n---\n\n# Visual ChangeNet\n\nVisual ChangeNet is a TAO Toolkit model for visual inspection and defect detection. It supports two tasks:\n\n- **Classify** — Binary image classification using a siamese-style architecture with a shared backbone (C-RADIO ViT) and a learnable difference module. Compares image pairs to classify defects as PASS\u002FNO_PASS.\n- **Segment** — Pixel-level change segmentation using a ViT-Large NVDINOv2 backbone. Compares before\u002Fafter image pairs to produce a binary change mask.\n\nThe backbone weight (`c_radio_v2_vit_base_patch16_224`) is the `nvidia\u002FC-RADIOv2-B` model from HuggingFace, distributed as `model.safetensors` (~393 MB). **The TAO 7.0.0-rc container does not auto-fetch from HF URLs** — `ptm_utils.load_pretrained_weights()` hands the `pretrained_backbone_path` value to `torch.load(path)` \u002F `safetensors.torch.load_file(path)` directly. Passing an `https:\u002F\u002Fhuggingface.co\u002F...` URL or a repo id produces `FileNotFoundError` and the run fails with `Execution status: FAIL` within a few seconds. Stage the file locally before launch:\n\n```bash\npython3 -c \"from huggingface_hub import hf_hub_download; import shutil; \\\nshutil.copy(hf_hub_download('nvidia\u002FC-RADIOv2-B', 'model.safetensors'), '\u003Cworkspace>\u002Fbackbone\u002Fc_radio_v2_b.safetensors')\"\n```\n\nMount it into the container (`-v \u003Cworkspace>\u002Fbackbone\u002Fc_radio_v2_b.safetensors:\u002Fdata\u002Fpretrained_models\u002FC-RADIOv2_B.safetensors`) and set the spec `model.backbone.pretrained_backbone_path` to the container path. `HF_TOKEN` is only needed at staging time, not at training time.\n\nSegment specs use `model.backbone.type: vit_large_nvdinov2` and the NVDINOv2\ncheckpoint family. Keep the checkpoint architecture aligned with the backbone\ntype: `NV_DINOV2_518_16_256.ckpt` is compatible with the packaged segment\ntemplates, but it must not be used with `fan_small_12_p4_hybrid`. If you switch\nto a different segment backbone, use a matching checkpoint or leave\n`model.backbone.pretrained_backbone_path` empty for default initialization.\n\n## Dataclass Schemas\n\nGenerated TAO Core schemas are packaged in `schemas\u002F\u003Caction>.schema.json`, with `schemas\u002Fmanifest.json` listing available actions. Each generated schema also emits `references\u002Fspec_template_\u003Caction>.yaml` from the schema top-level `default` field. AutoML enablement is declared at the model layer in `references\u002Fskill_info.yaml` via `automl_enabled`. Runnable AutoML still requires `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` to exist and parse. Use the packaged train schema for `automl_default_parameters`, `automl_disabled_parameters`, defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect `~\u002Ftao-core` at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.\n\n## Train Action Policy\n\nThis model is AutoML-enabled at the model layer. Before handling any train-stage request, read `references\u002Fskill_info.yaml` and resolve the run override from either an explicit `automl_policy` value or the user's workflow request. Use `automl_policy: on` by default and only expose `on` \u002F `off` in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as `automl_policy: off` for this run only. When `automl_policy: on`, `automl_enabled: true`, and both `schemas\u002Ftrain.schema.json` and `references\u002Fspec_template_train.yaml` are packaged, route the train action through `tao-skill-bank:tao-run-automl` by default with this model's `skill_dir`. Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and `automl_policy`. Use direct model training only when `automl_policy: off` or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.\n\nNon-train actions declared by this model skill (`evaluate`, `inference`,\n`export`, `quantize`, `segment_evaluate`, and `segment_inference`) stay in this\nmodel skill. Do not present `segment_export` or `segment_quantize` as runnable\nparent-skill actions until matching entries are packaged in\n`schemas\u002Fmanifest.json`. Prune and retrain are not declared in the current\nparent `references\u002Fskill_info.yaml`; do not present them as runnable parent-skill\nactions unless the metadata is extended with matching action wiring and schemas.\nThe per-run `automl_policy` override does not change model metadata.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference` for classify and segment variants), read `references\u002Ftao-deploy-visual-changenet.md` first. Deploy spec templates live in this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml` prefix.\nDeploy requires an exported ONNX artifact as `parent_model`. If no ONNX artifact exists and the main skill does not expose an export action, report deploy as blocked instead of inventing an artifact.\n\n## Training Requirements\n\nVisual ChangeNet has two separate task modes with different dataset types and data source structures.\n\n### Classify\n\n- **Dataset type:** visual_changenet_classify\n- **Formats:** default\n- **Accepted dataset intents:** training, evaluation, testing, calibration\n- **Monitoring metric:** val_loss\n\n#### Per-Action Dataset Requirements (Classify)\n\nThe `quantize` and `gen_trt_engine` rows below describe TAO spec data requirements only. They are not parent-skill actions unless the corresponding action is declared in `references\u002Fskill_info.yaml` or `deploy\u002Fskill_info.yaml`.\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| train | dataset.classify.train_dataset.images_dir | train_datasets | images.tar.gz | No |\n| train | dataset.classify.train_dataset.csv_path | train_datasets | dataset.csv | No |\n| train | dataset.classify.validation_dataset.images_dir | eval_dataset | images.tar.gz | No |\n| train | dataset.classify.validation_dataset.csv_path | eval_dataset | dataset.csv | No |\n| quantize | dataset.classify.train_dataset.images_dir | train_datasets | images.tar.gz | No |\n| quantize | dataset.classify.train_dataset.csv_path | train_datasets | dataset.csv | No |\n| quantize | dataset.classify.validation_dataset.images_dir | eval_dataset | images.tar.gz | No |\n| quantize | dataset.classify.validation_dataset.csv_path | eval_dataset | dataset.csv | No |\n| quantize | dataset.classify.quant_calibration_dataset.images_dir | train_datasets | images.tar.gz | No |\n| evaluate | dataset.classify.validation_dataset.images_dir | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.classify.validation_dataset.csv_path | eval_dataset | dataset.csv | No |\n| evaluate | dataset.classify.test_dataset.images_dir | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.classify.test_dataset.csv_path | eval_dataset | dataset.csv | No |\n| inference | dataset.classify.infer_dataset.images_dir | inference_dataset | images.tar.gz | No |\n| inference | dataset.classify.infer_dataset.csv_path | inference_dataset | dataset.csv | No |\n| gen_trt_engine | gen_trt_engine.tensorrt.calibration.cal_image_dir | calibration_dataset | images.tar.gz | Yes |\n\n### Segment\n\n- **Dataset type:** visual_changenet_segment\n- **Formats:** default\n- **Accepted dataset intents:** training, calibration\n- **Monitoring metric:** val_loss\n\nSegment uses a paired directory structure (`A\u002F`, `B\u002F`, `list\u002F`, `label\u002F`) instead of CSV + images. The `root_dir` spec key points to the top-level directory containing all four subdirectories.\n\n**Required files per dataset:** `A.tar.gz`, `B.tar.gz`, `list.tar.gz`, `label.tar.gz`\n\n#### Per-Action Dataset Requirements (Segment)\n\nThe `quantize` and `gen_trt_engine` rows below describe TAO spec data requirements only. They are not parent-skill actions unless the corresponding action is declared in `references\u002Fskill_info.yaml` or `deploy\u002Fskill_info.yaml`.\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| train | dataset.segment.root_dir | train_datasets | (root directory) | No |\n| quantize | dataset.segment.root_dir | train_datasets | (root directory) | No |\n| quantize | dataset.segment.quant_calibration_dataset.images_dir | train_datasets | (root directory) | No |\n| evaluate | dataset.segment.root_dir | train_datasets | (root directory) | No |\n| inference | dataset.segment.root_dir | train_datasets | (root directory) | No |\n| gen_trt_engine | dataset.segment.root_dir | train_datasets | (root directory) | No |\n| gen_trt_engine | gen_trt_engine.tensorrt.calibration.cal_image_dir | calibration_dataset | images.tar.gz | Yes |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action** — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`.\n\n```python\nS3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\nS3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n```\n\n**train (classify, mandatory data sources):**\n```python\n{\n    \"train.num_epochs\": 30,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"train.use_distributed_sampler\": False,\n    \"train.sync_batchnorm\": False,\n    \"dataset.classify.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.classify.train_dataset.csv_path\": f\"{S3_TRAIN}\u002Fdataset.csv\",\n    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n}\n```\n\n**train (segment, mandatory data sources):**\n```python\n{\n    \"train.num_epochs\": 30,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"train.use_distributed_sampler\": False,\n    \"train.sync_batchnorm\": False,\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n}\n```\n\n**export (classify):**\n```python\n{\n    \"export.input_height\": 896,\n    \"export.input_width\": 224,\n}\n```\n\n**export (segment):**\n```python\n{\n    \"export.input_height\": 224,\n    \"export.input_width\": 224,\n}\n```\n\n**quantize (classify, mandatory data sources):**\n```python\n{\n    \"dataset.classify.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.classify.train_dataset.csv_path\": f\"{S3_TRAIN}\u002Fdataset.csv\",\n    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n    \"dataset.classify.quant_calibration_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n}\n```\n\n**evaluate (classify, mandatory data sources):**\n```python\n{\n    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n    \"dataset.classify.test_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.test_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n}\n```\n\n**inference (classify, mandatory data sources):**\n```python\n{\n    \"dataset.classify.infer_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.infer_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n}\n```\n\n**gen_trt_engine (classify, mandatory data sources):**\n```python\n{\n    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n}\n```\n\n**quantize (segment, mandatory data sources):**\n```python\n{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n    \"dataset.segment.quant_calibration_dataset.images_dir\": f\"{S3_TRAIN}\",\n}\n```\n\n**evaluate (segment, mandatory data sources):**\n```python\n{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n}\n```\n\n**inference (segment, mandatory data sources):**\n```python\n{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n}\n```\n\n**gen_trt_engine (segment, mandatory data sources):**\n```python\n{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n}\n```\n## Optional: running via the TAO SDK\n\nWhen running without the TAO SDK (local docker), resolve the TAO pyt image from `versions.yaml` and invoke `visual_changenet \u003Ctrain|evaluate|inference|export|quantize>` directly. `--shm-size=8g` is required, the C-RADIO `.safetensors` must be mounted to `\u002Fdata\u002Fpretrained_models\u002FC-RADIOv2_B.safetensors`, and checkpoint\u002Fresults_dir can be overridden on the command line. See `references\u002Flocal-docker.md` for the full `docker run` command, mounts, and overrides.\n\n## Tasks\n\n### Classify (default)\n\nUses actions: `train`, `evaluate`, `inference`. Defaults template: `references\u002Fspec_template_train.yaml`.\n\n### Segment\n\nUses skill action names `segment_train`, `segment_evaluate`, and\n`segment_inference`. When invoking local Docker directly, run TAO CLI subcommands\n`train`, `evaluate`, and `inference` with `task: segment` in the spec. The\nschema-driven action templates are `references\u002Fspec_template_segment_train.yaml`,\n`references\u002Fspec_template_segment_evaluate.yaml`, and\n`references\u002Fspec_template_segment_inference.yaml`; the compact direct-Docker\nexample template is `references\u002Fspec_template_segment.yaml`.\n\nSegmentation requires compiling custom CUDA ops (`MultiScaleDeformableAttention`) on first run, which takes ~5 minutes. The ViT adapter backbone uses these for multi-scale feature extraction.\n\nDataset structure for segmentation differs from classify — uses paired directories (`A\u002F`, `B\u002F`, `list\u002F`, `label\u002F`) instead of CSV files. See `dataset.segment.root_dir` in the defaults.\n\n## Data Format\n\nClassify needs a 4-column CSV (`input_path,golden_path,label,object_name`) plus an images directory; segment uses a paired directory structure (`A\u002F`, `B\u002F`, `list\u002F`, `label\u002F`) under `dataset.segment.root_dir` instead of CSV. The `image_ext` field (default `.jpg`) must match the actual file extensions; if images are `.png`, set `dataset.classify.image_ext: .png`. Multi-lighting input is configured via `dataset.classify.input_map` (each lighting name maps to a channel index) with `dataset.classify.num_input` set to match. See `references\u002Fdata-formats.md` for the per-field input tables (classify train\u002Feval\u002Finference, segment), CSV column semantics, lighting\u002Fpath-concatenation conventions, the segment directory layout, and `input_map`\u002F`grid_map` examples.\n\n## Important Parameters\n\nKey knobs include `train.validation_interval` (default 50, must be ≤ num_epochs), `train.checkpoint_interval` (default 200, must be ≤ num_epochs), `train.num_epochs` (default 100), `model.classify.eval_margin` (default 0.3, the precision\u002Frecall threshold), `model.classify.train_margin_euclid` (default 2.0), `model.classify.embedding_vectors` (default 5), `dataset.classify.batch_size` (default 16, must be > 1), `dataset.classify.fpratio_sampling` (default 0.25), and `train.classify.cls_weight` (default [1.0, 10.0]). Hardware: minimum 1 GPU with 16GB+ VRAM, recommended 8 GPUs (DDP); do not set `gpu_spec_key` (GPU count is managed internally by TAO), `num_nodes` (default 1) controls multi-node. See `references\u002Ftuning-parameters.md` for the full per-parameter guidance and hardware detail.\n\n## Error Patterns\n\nFor checkpoint-not-found, CSV format mismatch, image extension mismatch, OOM, low evaluation accuracy, the contrastive-loss `AssertionError`, checkpoint load key mismatch at evaluate\u002Finference, non-convergence, segment-only backbone dimension mismatch, the `MultiScaleDeformableAttention` `OSError`, the Lightning `MisconfigurationException`, `ModuleNotFoundError: nvidia_tao_pytorch`, and epoch defaults, see `references\u002Ftroubleshooting.md` for the full symptom-and-fix list.\n\n## Spec Param \u002F Parent Model Inference\n\nModel-specific parent-model mappings are declared in `references\u002Fskill_info.yaml` under `spec_params`, so generated runners and agents resolve checkpoints before `create_job()` instead of guessing file names. For `parent_model` or `parent_model_folder`, pass the upstream train\u002Fexport\u002FAutoML child job id as `parent_job_id`; the SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. See `references\u002Fparent-model-inference.md` for the full per-action spec-field-to-inference-function mapping table.\n\n## Deployment\n\n- [tao-deploy-visual-changenet](references\u002Ftao-deploy-visual-changenet.md)\n",{"data":35,"body":49},{"name":4,"description":6,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"tags":41},"Requires docker + nvidia-container-toolkit.",{"author":38,"version":39},"NVIDIA Corporation","0.1.0","Read Bash",[42,43,44,45,46,47,48],"pcb","aoi","defect","classification","segmentation","siamese","visual-inspection",{"type":50,"children":51},"root",[52,61,67,93,186,246,275,311,318,411,417,525,612,671,677,682,688,731,738,770,1200,1205,1242,1284,1321,1327,1355,1551,1557,1576,1601,1609,1722,1730,1801,1809,1846,1854,1890,1898,1955,1963,2014,2022,2059,2067,2096,2104,2140,2148,2176,2184,2211,2219,2254,2260,2321,2327,2333,2363,2368,2450,2463,2500,2506,2623,2629,2737,2743,2793,2799,2857,2863,2875],{"type":53,"tag":54,"props":55,"children":57},"element","h1",{"id":56},"visual-changenet",[58],{"type":59,"value":60},"text","Visual ChangeNet",{"type":53,"tag":62,"props":63,"children":64},"p",{},[65],{"type":59,"value":66},"Visual ChangeNet is a TAO Toolkit model for visual inspection and defect detection. It supports two tasks:",{"type":53,"tag":68,"props":69,"children":70},"ul",{},[71,83],{"type":53,"tag":72,"props":73,"children":74},"li",{},[75,81],{"type":53,"tag":76,"props":77,"children":78},"strong",{},[79],{"type":59,"value":80},"Classify",{"type":59,"value":82}," — Binary image classification using a siamese-style architecture with a shared backbone (C-RADIO ViT) and a learnable difference module. Compares image pairs to classify defects as PASS\u002FNO_PASS.",{"type":53,"tag":72,"props":84,"children":85},{},[86,91],{"type":53,"tag":76,"props":87,"children":88},{},[89],{"type":59,"value":90},"Segment",{"type":59,"value":92}," — Pixel-level change segmentation using a ViT-Large NVDINOv2 backbone. Compares before\u002Fafter image pairs to produce a binary change mask.",{"type":53,"tag":62,"props":94,"children":95},{},[96,98,105,107,113,115,121,123,128,130,136,138,144,146,152,154,160,162,168,170,176,178,184],{"type":59,"value":97},"The backbone weight (",{"type":53,"tag":99,"props":100,"children":102},"code",{"className":101},[],[103],{"type":59,"value":104},"c_radio_v2_vit_base_patch16_224",{"type":59,"value":106},") is the ",{"type":53,"tag":99,"props":108,"children":110},{"className":109},[],[111],{"type":59,"value":112},"nvidia\u002FC-RADIOv2-B",{"type":59,"value":114}," model from HuggingFace, distributed as ",{"type":53,"tag":99,"props":116,"children":118},{"className":117},[],[119],{"type":59,"value":120},"model.safetensors",{"type":59,"value":122}," (~393 MB). ",{"type":53,"tag":76,"props":124,"children":125},{},[126],{"type":59,"value":127},"The TAO 7.0.0-rc container does not auto-fetch from HF URLs",{"type":59,"value":129}," — ",{"type":53,"tag":99,"props":131,"children":133},{"className":132},[],[134],{"type":59,"value":135},"ptm_utils.load_pretrained_weights()",{"type":59,"value":137}," hands the ",{"type":53,"tag":99,"props":139,"children":141},{"className":140},[],[142],{"type":59,"value":143},"pretrained_backbone_path",{"type":59,"value":145}," value to ",{"type":53,"tag":99,"props":147,"children":149},{"className":148},[],[150],{"type":59,"value":151},"torch.load(path)",{"type":59,"value":153}," \u002F ",{"type":53,"tag":99,"props":155,"children":157},{"className":156},[],[158],{"type":59,"value":159},"safetensors.torch.load_file(path)",{"type":59,"value":161}," directly. Passing an ",{"type":53,"tag":99,"props":163,"children":165},{"className":164},[],[166],{"type":59,"value":167},"https:\u002F\u002Fhuggingface.co\u002F...",{"type":59,"value":169}," URL or a repo id produces ",{"type":53,"tag":99,"props":171,"children":173},{"className":172},[],[174],{"type":59,"value":175},"FileNotFoundError",{"type":59,"value":177}," and the run fails with ",{"type":53,"tag":99,"props":179,"children":181},{"className":180},[],[182],{"type":59,"value":183},"Execution status: FAIL",{"type":59,"value":185}," within a few seconds. Stage the file locally before launch:",{"type":53,"tag":187,"props":188,"children":193},"pre",{"className":189,"code":190,"language":191,"meta":192,"style":192},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","python3 -c \"from huggingface_hub import hf_hub_download; import shutil; \\\nshutil.copy(hf_hub_download('nvidia\u002FC-RADIOv2-B', 'model.safetensors'), '\u003Cworkspace>\u002Fbackbone\u002Fc_radio_v2_b.safetensors')\"\n","bash","",[194],{"type":53,"tag":99,"props":195,"children":196},{"__ignoreMap":192},[197,232],{"type":53,"tag":198,"props":199,"children":202},"span",{"class":200,"line":201},"line",1,[203,209,215,221,226],{"type":53,"tag":198,"props":204,"children":206},{"style":205},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[207],{"type":59,"value":208},"python3",{"type":53,"tag":198,"props":210,"children":212},{"style":211},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[213],{"type":59,"value":214}," -c",{"type":53,"tag":198,"props":216,"children":218},{"style":217},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[219],{"type":59,"value":220}," \"",{"type":53,"tag":198,"props":222,"children":223},{"style":211},[224],{"type":59,"value":225},"from huggingface_hub import hf_hub_download; import shutil; ",{"type":53,"tag":198,"props":227,"children":229},{"style":228},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[230],{"type":59,"value":231},"\\\n",{"type":53,"tag":198,"props":233,"children":235},{"class":200,"line":234},2,[236,241],{"type":53,"tag":198,"props":237,"children":238},{"style":211},[239],{"type":59,"value":240},"shutil.copy(hf_hub_download('nvidia\u002FC-RADIOv2-B', 'model.safetensors'), '\u003Cworkspace>\u002Fbackbone\u002Fc_radio_v2_b.safetensors')",{"type":53,"tag":198,"props":242,"children":243},{"style":217},[244],{"type":59,"value":245},"\"\n",{"type":53,"tag":62,"props":247,"children":248},{},[249,251,257,259,265,267,273],{"type":59,"value":250},"Mount it into the container (",{"type":53,"tag":99,"props":252,"children":254},{"className":253},[],[255],{"type":59,"value":256},"-v \u003Cworkspace>\u002Fbackbone\u002Fc_radio_v2_b.safetensors:\u002Fdata\u002Fpretrained_models\u002FC-RADIOv2_B.safetensors",{"type":59,"value":258},") and set the spec ",{"type":53,"tag":99,"props":260,"children":262},{"className":261},[],[263],{"type":59,"value":264},"model.backbone.pretrained_backbone_path",{"type":59,"value":266}," to the container path. ",{"type":53,"tag":99,"props":268,"children":270},{"className":269},[],[271],{"type":59,"value":272},"HF_TOKEN",{"type":59,"value":274}," is only needed at staging time, not at training time.",{"type":53,"tag":62,"props":276,"children":277},{},[278,280,286,288,294,296,302,304,309],{"type":59,"value":279},"Segment specs use ",{"type":53,"tag":99,"props":281,"children":283},{"className":282},[],[284],{"type":59,"value":285},"model.backbone.type: vit_large_nvdinov2",{"type":59,"value":287}," and the NVDINOv2\ncheckpoint family. Keep the checkpoint architecture aligned with the backbone\ntype: ",{"type":53,"tag":99,"props":289,"children":291},{"className":290},[],[292],{"type":59,"value":293},"NV_DINOV2_518_16_256.ckpt",{"type":59,"value":295}," is compatible with the packaged segment\ntemplates, but it must not be used with ",{"type":53,"tag":99,"props":297,"children":299},{"className":298},[],[300],{"type":59,"value":301},"fan_small_12_p4_hybrid",{"type":59,"value":303},". If you switch\nto a different segment backbone, use a matching checkpoint or leave\n",{"type":53,"tag":99,"props":305,"children":307},{"className":306},[],[308],{"type":59,"value":264},{"type":59,"value":310}," empty for default initialization.",{"type":53,"tag":312,"props":313,"children":315},"h2",{"id":314},"dataclass-schemas",[316],{"type":59,"value":317},"Dataclass Schemas",{"type":53,"tag":62,"props":319,"children":320},{},[321,323,329,331,337,339,345,347,353,355,361,363,369,371,377,379,385,387,393,395,401,403,409],{"type":59,"value":322},"Generated TAO Core schemas are packaged in ",{"type":53,"tag":99,"props":324,"children":326},{"className":325},[],[327],{"type":59,"value":328},"schemas\u002F\u003Caction>.schema.json",{"type":59,"value":330},", with ",{"type":53,"tag":99,"props":332,"children":334},{"className":333},[],[335],{"type":59,"value":336},"schemas\u002Fmanifest.json",{"type":59,"value":338}," listing available actions. Each generated schema also emits ",{"type":53,"tag":99,"props":340,"children":342},{"className":341},[],[343],{"type":59,"value":344},"references\u002Fspec_template_\u003Caction>.yaml",{"type":59,"value":346}," from the schema top-level ",{"type":53,"tag":99,"props":348,"children":350},{"className":349},[],[351],{"type":59,"value":352},"default",{"type":59,"value":354}," field. AutoML enablement is declared at the model layer in ",{"type":53,"tag":99,"props":356,"children":358},{"className":357},[],[359],{"type":59,"value":360},"references\u002Fskill_info.yaml",{"type":59,"value":362}," via ",{"type":53,"tag":99,"props":364,"children":366},{"className":365},[],[367],{"type":59,"value":368},"automl_enabled",{"type":59,"value":370},". Runnable AutoML still requires ",{"type":53,"tag":99,"props":372,"children":374},{"className":373},[],[375],{"type":59,"value":376},"schemas\u002Ftrain.schema.json",{"type":59,"value":378}," and ",{"type":53,"tag":99,"props":380,"children":382},{"className":381},[],[383],{"type":59,"value":384},"references\u002Fspec_template_train.yaml",{"type":59,"value":386}," to exist and parse. Use the packaged train schema for ",{"type":53,"tag":99,"props":388,"children":390},{"className":389},[],[391],{"type":59,"value":392},"automl_default_parameters",{"type":59,"value":394},", ",{"type":53,"tag":99,"props":396,"children":398},{"className":397},[],[399],{"type":59,"value":400},"automl_disabled_parameters",{"type":59,"value":402},", defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ",{"type":53,"tag":99,"props":404,"children":406},{"className":405},[],[407],{"type":59,"value":408},"~\u002Ftao-core",{"type":59,"value":410}," at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.",{"type":53,"tag":312,"props":412,"children":414},{"id":413},"train-action-policy",[415],{"type":59,"value":416},"Train Action Policy",{"type":53,"tag":62,"props":418,"children":419},{},[420,422,427,429,435,437,443,445,451,452,458,460,466,468,473,474,480,482,487,488,493,495,501,503,509,511,516,518,523],{"type":59,"value":421},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":53,"tag":99,"props":423,"children":425},{"className":424},[],[426],{"type":59,"value":360},{"type":59,"value":428}," and resolve the run override from either an explicit ",{"type":53,"tag":99,"props":430,"children":432},{"className":431},[],[433],{"type":59,"value":434},"automl_policy",{"type":59,"value":436}," value or the user's workflow request. Use ",{"type":53,"tag":99,"props":438,"children":440},{"className":439},[],[441],{"type":59,"value":442},"automl_policy: on",{"type":59,"value":444}," by default and only expose ",{"type":53,"tag":99,"props":446,"children":448},{"className":447},[],[449],{"type":59,"value":450},"on",{"type":59,"value":153},{"type":53,"tag":99,"props":453,"children":455},{"className":454},[],[456],{"type":59,"value":457},"off",{"type":59,"value":459}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":53,"tag":99,"props":461,"children":463},{"className":462},[],[464],{"type":59,"value":465},"automl_policy: off",{"type":59,"value":467}," for this run only. When ",{"type":53,"tag":99,"props":469,"children":471},{"className":470},[],[472],{"type":59,"value":442},{"type":59,"value":394},{"type":53,"tag":99,"props":475,"children":477},{"className":476},[],[478],{"type":59,"value":479},"automl_enabled: true",{"type":59,"value":481},", and both ",{"type":53,"tag":99,"props":483,"children":485},{"className":484},[],[486],{"type":59,"value":376},{"type":59,"value":378},{"type":53,"tag":99,"props":489,"children":491},{"className":490},[],[492],{"type":59,"value":384},{"type":59,"value":494}," are packaged, route the train action through ",{"type":53,"tag":99,"props":496,"children":498},{"className":497},[],[499],{"type":59,"value":500},"tao-skill-bank:tao-run-automl",{"type":59,"value":502}," by default with this model's ",{"type":53,"tag":99,"props":504,"children":506},{"className":505},[],[507],{"type":59,"value":508},"skill_dir",{"type":59,"value":510},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":53,"tag":99,"props":512,"children":514},{"className":513},[],[515],{"type":59,"value":434},{"type":59,"value":517},". Use direct model training only when ",{"type":53,"tag":99,"props":519,"children":521},{"className":520},[],[522],{"type":59,"value":465},{"type":59,"value":524}," or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.",{"type":53,"tag":62,"props":526,"children":527},{},[528,530,536,537,543,545,551,552,558,559,565,567,573,575,581,583,589,591,596,598,603,605,610],{"type":59,"value":529},"Non-train actions declared by this model skill (",{"type":53,"tag":99,"props":531,"children":533},{"className":532},[],[534],{"type":59,"value":535},"evaluate",{"type":59,"value":394},{"type":53,"tag":99,"props":538,"children":540},{"className":539},[],[541],{"type":59,"value":542},"inference",{"type":59,"value":544},",\n",{"type":53,"tag":99,"props":546,"children":548},{"className":547},[],[549],{"type":59,"value":550},"export",{"type":59,"value":394},{"type":53,"tag":99,"props":553,"children":555},{"className":554},[],[556],{"type":59,"value":557},"quantize",{"type":59,"value":394},{"type":53,"tag":99,"props":560,"children":562},{"className":561},[],[563],{"type":59,"value":564},"segment_evaluate",{"type":59,"value":566},", and ",{"type":53,"tag":99,"props":568,"children":570},{"className":569},[],[571],{"type":59,"value":572},"segment_inference",{"type":59,"value":574},") stay in this\nmodel skill. Do not present ",{"type":53,"tag":99,"props":576,"children":578},{"className":577},[],[579],{"type":59,"value":580},"segment_export",{"type":59,"value":582}," or ",{"type":53,"tag":99,"props":584,"children":586},{"className":585},[],[587],{"type":59,"value":588},"segment_quantize",{"type":59,"value":590}," as runnable\nparent-skill actions until matching entries are packaged in\n",{"type":53,"tag":99,"props":592,"children":594},{"className":593},[],[595],{"type":59,"value":336},{"type":59,"value":597},". Prune and retrain are not declared in the current\nparent ",{"type":53,"tag":99,"props":599,"children":601},{"className":600},[],[602],{"type":59,"value":360},{"type":59,"value":604},"; do not present them as runnable parent-skill\nactions unless the metadata is extended with matching action wiring and schemas.\nThe per-run ",{"type":53,"tag":99,"props":606,"children":608},{"className":607},[],[609],{"type":59,"value":434},{"type":59,"value":611}," override does not change model metadata.",{"type":53,"tag":62,"props":613,"children":614},{},[615,617,623,625,630,632,637,639,645,647,653,655,661,663,669],{"type":59,"value":616},"For TAO Deploy TensorRT actions (",{"type":53,"tag":99,"props":618,"children":620},{"className":619},[],[621],{"type":59,"value":622},"gen_trt_engine",{"type":59,"value":624},", TensorRT ",{"type":53,"tag":99,"props":626,"children":628},{"className":627},[],[629],{"type":59,"value":535},{"type":59,"value":631},", and TensorRT ",{"type":53,"tag":99,"props":633,"children":635},{"className":634},[],[636],{"type":59,"value":542},{"type":59,"value":638}," for classify and segment variants), read ",{"type":53,"tag":99,"props":640,"children":642},{"className":641},[],[643],{"type":59,"value":644},"references\u002Ftao-deploy-visual-changenet.md",{"type":59,"value":646}," first. Deploy spec templates live in this skill's ",{"type":53,"tag":99,"props":648,"children":650},{"className":649},[],[651],{"type":59,"value":652},"references\u002F",{"type":59,"value":654}," folder with the ",{"type":53,"tag":99,"props":656,"children":658},{"className":657},[],[659],{"type":59,"value":660},"spec_template_deploy_*.yaml",{"type":59,"value":662}," prefix.\nDeploy requires an exported ONNX artifact as ",{"type":53,"tag":99,"props":664,"children":666},{"className":665},[],[667],{"type":59,"value":668},"parent_model",{"type":59,"value":670},". If no ONNX artifact exists and the main skill does not expose an export action, report deploy as blocked instead of inventing an artifact.",{"type":53,"tag":312,"props":672,"children":674},{"id":673},"training-requirements",[675],{"type":59,"value":676},"Training Requirements",{"type":53,"tag":62,"props":678,"children":679},{},[680],{"type":59,"value":681},"Visual ChangeNet has two separate task modes with different dataset types and data source structures.",{"type":53,"tag":683,"props":684,"children":686},"h3",{"id":685},"classify",[687],{"type":59,"value":80},{"type":53,"tag":68,"props":689,"children":690},{},[691,701,711,721],{"type":53,"tag":72,"props":692,"children":693},{},[694,699],{"type":53,"tag":76,"props":695,"children":696},{},[697],{"type":59,"value":698},"Dataset type:",{"type":59,"value":700}," visual_changenet_classify",{"type":53,"tag":72,"props":702,"children":703},{},[704,709],{"type":53,"tag":76,"props":705,"children":706},{},[707],{"type":59,"value":708},"Formats:",{"type":59,"value":710}," default",{"type":53,"tag":72,"props":712,"children":713},{},[714,719],{"type":53,"tag":76,"props":715,"children":716},{},[717],{"type":59,"value":718},"Accepted dataset intents:",{"type":59,"value":720}," training, evaluation, testing, calibration",{"type":53,"tag":72,"props":722,"children":723},{},[724,729],{"type":53,"tag":76,"props":725,"children":726},{},[727],{"type":59,"value":728},"Monitoring metric:",{"type":59,"value":730}," val_loss",{"type":53,"tag":732,"props":733,"children":735},"h4",{"id":734},"per-action-dataset-requirements-classify",[736],{"type":59,"value":737},"Per-Action Dataset Requirements (Classify)",{"type":53,"tag":62,"props":739,"children":740},{},[741,743,748,749,754,756,761,762,768],{"type":59,"value":742},"The ",{"type":53,"tag":99,"props":744,"children":746},{"className":745},[],[747],{"type":59,"value":557},{"type":59,"value":378},{"type":53,"tag":99,"props":750,"children":752},{"className":751},[],[753],{"type":59,"value":622},{"type":59,"value":755}," rows below describe TAO spec data requirements only. They are not parent-skill actions unless the corresponding action is declared in ",{"type":53,"tag":99,"props":757,"children":759},{"className":758},[],[760],{"type":59,"value":360},{"type":59,"value":582},{"type":53,"tag":99,"props":763,"children":765},{"className":764},[],[766],{"type":59,"value":767},"deploy\u002Fskill_info.yaml",{"type":59,"value":769},".",{"type":53,"tag":771,"props":772,"children":773},"table",{},[774,808],{"type":53,"tag":775,"props":776,"children":777},"thead",{},[778],{"type":53,"tag":779,"props":780,"children":781},"tr",{},[782,788,793,798,803],{"type":53,"tag":783,"props":784,"children":785},"th",{},[786],{"type":59,"value":787},"Action",{"type":53,"tag":783,"props":789,"children":790},{},[791],{"type":59,"value":792},"Spec Key",{"type":53,"tag":783,"props":794,"children":795},{},[796],{"type":59,"value":797},"Source",{"type":53,"tag":783,"props":799,"children":800},{},[801],{"type":59,"value":802},"Files",{"type":53,"tag":783,"props":804,"children":805},{},[806],{"type":59,"value":807},"List?",{"type":53,"tag":809,"props":810,"children":811},"tbody",{},[812,841,866,891,915,938,961,984,1007,1031,1054,1077,1101,1125,1150,1174],{"type":53,"tag":779,"props":813,"children":814},{},[815,821,826,831,836],{"type":53,"tag":816,"props":817,"children":818},"td",{},[819],{"type":59,"value":820},"train",{"type":53,"tag":816,"props":822,"children":823},{},[824],{"type":59,"value":825},"dataset.classify.train_dataset.images_dir",{"type":53,"tag":816,"props":827,"children":828},{},[829],{"type":59,"value":830},"train_datasets",{"type":53,"tag":816,"props":832,"children":833},{},[834],{"type":59,"value":835},"images.tar.gz",{"type":53,"tag":816,"props":837,"children":838},{},[839],{"type":59,"value":840},"No",{"type":53,"tag":779,"props":842,"children":843},{},[844,848,853,857,862],{"type":53,"tag":816,"props":845,"children":846},{},[847],{"type":59,"value":820},{"type":53,"tag":816,"props":849,"children":850},{},[851],{"type":59,"value":852},"dataset.classify.train_dataset.csv_path",{"type":53,"tag":816,"props":854,"children":855},{},[856],{"type":59,"value":830},{"type":53,"tag":816,"props":858,"children":859},{},[860],{"type":59,"value":861},"dataset.csv",{"type":53,"tag":816,"props":863,"children":864},{},[865],{"type":59,"value":840},{"type":53,"tag":779,"props":867,"children":868},{},[869,873,878,883,887],{"type":53,"tag":816,"props":870,"children":871},{},[872],{"type":59,"value":820},{"type":53,"tag":816,"props":874,"children":875},{},[876],{"type":59,"value":877},"dataset.classify.validation_dataset.images_dir",{"type":53,"tag":816,"props":879,"children":880},{},[881],{"type":59,"value":882},"eval_dataset",{"type":53,"tag":816,"props":884,"children":885},{},[886],{"type":59,"value":835},{"type":53,"tag":816,"props":888,"children":889},{},[890],{"type":59,"value":840},{"type":53,"tag":779,"props":892,"children":893},{},[894,898,903,907,911],{"type":53,"tag":816,"props":895,"children":896},{},[897],{"type":59,"value":820},{"type":53,"tag":816,"props":899,"children":900},{},[901],{"type":59,"value":902},"dataset.classify.validation_dataset.csv_path",{"type":53,"tag":816,"props":904,"children":905},{},[906],{"type":59,"value":882},{"type":53,"tag":816,"props":908,"children":909},{},[910],{"type":59,"value":861},{"type":53,"tag":816,"props":912,"children":913},{},[914],{"type":59,"value":840},{"type":53,"tag":779,"props":916,"children":917},{},[918,922,926,930,934],{"type":53,"tag":816,"props":919,"children":920},{},[921],{"type":59,"value":557},{"type":53,"tag":816,"props":923,"children":924},{},[925],{"type":59,"value":825},{"type":53,"tag":816,"props":927,"children":928},{},[929],{"type":59,"value":830},{"type":53,"tag":816,"props":931,"children":932},{},[933],{"type":59,"value":835},{"type":53,"tag":816,"props":935,"children":936},{},[937],{"type":59,"value":840},{"type":53,"tag":779,"props":939,"children":940},{},[941,945,949,953,957],{"type":53,"tag":816,"props":942,"children":943},{},[944],{"type":59,"value":557},{"type":53,"tag":816,"props":946,"children":947},{},[948],{"type":59,"value":852},{"type":53,"tag":816,"props":950,"children":951},{},[952],{"type":59,"value":830},{"type":53,"tag":816,"props":954,"children":955},{},[956],{"type":59,"value":861},{"type":53,"tag":816,"props":958,"children":959},{},[960],{"type":59,"value":840},{"type":53,"tag":779,"props":962,"children":963},{},[964,968,972,976,980],{"type":53,"tag":816,"props":965,"children":966},{},[967],{"type":59,"value":557},{"type":53,"tag":816,"props":969,"children":970},{},[971],{"type":59,"value":877},{"type":53,"tag":816,"props":973,"children":974},{},[975],{"type":59,"value":882},{"type":53,"tag":816,"props":977,"children":978},{},[979],{"type":59,"value":835},{"type":53,"tag":816,"props":981,"children":982},{},[983],{"type":59,"value":840},{"type":53,"tag":779,"props":985,"children":986},{},[987,991,995,999,1003],{"type":53,"tag":816,"props":988,"children":989},{},[990],{"type":59,"value":557},{"type":53,"tag":816,"props":992,"children":993},{},[994],{"type":59,"value":902},{"type":53,"tag":816,"props":996,"children":997},{},[998],{"type":59,"value":882},{"type":53,"tag":816,"props":1000,"children":1001},{},[1002],{"type":59,"value":861},{"type":53,"tag":816,"props":1004,"children":1005},{},[1006],{"type":59,"value":840},{"type":53,"tag":779,"props":1008,"children":1009},{},[1010,1014,1019,1023,1027],{"type":53,"tag":816,"props":1011,"children":1012},{},[1013],{"type":59,"value":557},{"type":53,"tag":816,"props":1015,"children":1016},{},[1017],{"type":59,"value":1018},"dataset.classify.quant_calibration_dataset.images_dir",{"type":53,"tag":816,"props":1020,"children":1021},{},[1022],{"type":59,"value":830},{"type":53,"tag":816,"props":1024,"children":1025},{},[1026],{"type":59,"value":835},{"type":53,"tag":816,"props":1028,"children":1029},{},[1030],{"type":59,"value":840},{"type":53,"tag":779,"props":1032,"children":1033},{},[1034,1038,1042,1046,1050],{"type":53,"tag":816,"props":1035,"children":1036},{},[1037],{"type":59,"value":535},{"type":53,"tag":816,"props":1039,"children":1040},{},[1041],{"type":59,"value":877},{"type":53,"tag":816,"props":1043,"children":1044},{},[1045],{"type":59,"value":882},{"type":53,"tag":816,"props":1047,"children":1048},{},[1049],{"type":59,"value":835},{"type":53,"tag":816,"props":1051,"children":1052},{},[1053],{"type":59,"value":840},{"type":53,"tag":779,"props":1055,"children":1056},{},[1057,1061,1065,1069,1073],{"type":53,"tag":816,"props":1058,"children":1059},{},[1060],{"type":59,"value":535},{"type":53,"tag":816,"props":1062,"children":1063},{},[1064],{"type":59,"value":902},{"type":53,"tag":816,"props":1066,"children":1067},{},[1068],{"type":59,"value":882},{"type":53,"tag":816,"props":1070,"children":1071},{},[1072],{"type":59,"value":861},{"type":53,"tag":816,"props":1074,"children":1075},{},[1076],{"type":59,"value":840},{"type":53,"tag":779,"props":1078,"children":1079},{},[1080,1084,1089,1093,1097],{"type":53,"tag":816,"props":1081,"children":1082},{},[1083],{"type":59,"value":535},{"type":53,"tag":816,"props":1085,"children":1086},{},[1087],{"type":59,"value":1088},"dataset.classify.test_dataset.images_dir",{"type":53,"tag":816,"props":1090,"children":1091},{},[1092],{"type":59,"value":882},{"type":53,"tag":816,"props":1094,"children":1095},{},[1096],{"type":59,"value":835},{"type":53,"tag":816,"props":1098,"children":1099},{},[1100],{"type":59,"value":840},{"type":53,"tag":779,"props":1102,"children":1103},{},[1104,1108,1113,1117,1121],{"type":53,"tag":816,"props":1105,"children":1106},{},[1107],{"type":59,"value":535},{"type":53,"tag":816,"props":1109,"children":1110},{},[1111],{"type":59,"value":1112},"dataset.classify.test_dataset.csv_path",{"type":53,"tag":816,"props":1114,"children":1115},{},[1116],{"type":59,"value":882},{"type":53,"tag":816,"props":1118,"children":1119},{},[1120],{"type":59,"value":861},{"type":53,"tag":816,"props":1122,"children":1123},{},[1124],{"type":59,"value":840},{"type":53,"tag":779,"props":1126,"children":1127},{},[1128,1132,1137,1142,1146],{"type":53,"tag":816,"props":1129,"children":1130},{},[1131],{"type":59,"value":542},{"type":53,"tag":816,"props":1133,"children":1134},{},[1135],{"type":59,"value":1136},"dataset.classify.infer_dataset.images_dir",{"type":53,"tag":816,"props":1138,"children":1139},{},[1140],{"type":59,"value":1141},"inference_dataset",{"type":53,"tag":816,"props":1143,"children":1144},{},[1145],{"type":59,"value":835},{"type":53,"tag":816,"props":1147,"children":1148},{},[1149],{"type":59,"value":840},{"type":53,"tag":779,"props":1151,"children":1152},{},[1153,1157,1162,1166,1170],{"type":53,"tag":816,"props":1154,"children":1155},{},[1156],{"type":59,"value":542},{"type":53,"tag":816,"props":1158,"children":1159},{},[1160],{"type":59,"value":1161},"dataset.classify.infer_dataset.csv_path",{"type":53,"tag":816,"props":1163,"children":1164},{},[1165],{"type":59,"value":1141},{"type":53,"tag":816,"props":1167,"children":1168},{},[1169],{"type":59,"value":861},{"type":53,"tag":816,"props":1171,"children":1172},{},[1173],{"type":59,"value":840},{"type":53,"tag":779,"props":1175,"children":1176},{},[1177,1181,1186,1191,1195],{"type":53,"tag":816,"props":1178,"children":1179},{},[1180],{"type":59,"value":622},{"type":53,"tag":816,"props":1182,"children":1183},{},[1184],{"type":59,"value":1185},"gen_trt_engine.tensorrt.calibration.cal_image_dir",{"type":53,"tag":816,"props":1187,"children":1188},{},[1189],{"type":59,"value":1190},"calibration_dataset",{"type":53,"tag":816,"props":1192,"children":1193},{},[1194],{"type":59,"value":835},{"type":53,"tag":816,"props":1196,"children":1197},{},[1198],{"type":59,"value":1199},"Yes",{"type":53,"tag":683,"props":1201,"children":1203},{"id":1202},"segment",[1204],{"type":59,"value":90},{"type":53,"tag":68,"props":1206,"children":1207},{},[1208,1217,1225,1234],{"type":53,"tag":72,"props":1209,"children":1210},{},[1211,1215],{"type":53,"tag":76,"props":1212,"children":1213},{},[1214],{"type":59,"value":698},{"type":59,"value":1216}," visual_changenet_segment",{"type":53,"tag":72,"props":1218,"children":1219},{},[1220,1224],{"type":53,"tag":76,"props":1221,"children":1222},{},[1223],{"type":59,"value":708},{"type":59,"value":710},{"type":53,"tag":72,"props":1226,"children":1227},{},[1228,1232],{"type":53,"tag":76,"props":1229,"children":1230},{},[1231],{"type":59,"value":718},{"type":59,"value":1233}," training, calibration",{"type":53,"tag":72,"props":1235,"children":1236},{},[1237,1241],{"type":53,"tag":76,"props":1238,"children":1239},{},[1240],{"type":59,"value":728},{"type":59,"value":730},{"type":53,"tag":62,"props":1243,"children":1244},{},[1245,1247,1253,1254,1260,1261,1267,1268,1274,1276,1282],{"type":59,"value":1246},"Segment uses a paired directory structure (",{"type":53,"tag":99,"props":1248,"children":1250},{"className":1249},[],[1251],{"type":59,"value":1252},"A\u002F",{"type":59,"value":394},{"type":53,"tag":99,"props":1255,"children":1257},{"className":1256},[],[1258],{"type":59,"value":1259},"B\u002F",{"type":59,"value":394},{"type":53,"tag":99,"props":1262,"children":1264},{"className":1263},[],[1265],{"type":59,"value":1266},"list\u002F",{"type":59,"value":394},{"type":53,"tag":99,"props":1269,"children":1271},{"className":1270},[],[1272],{"type":59,"value":1273},"label\u002F",{"type":59,"value":1275},") instead of CSV + images. The ",{"type":53,"tag":99,"props":1277,"children":1279},{"className":1278},[],[1280],{"type":59,"value":1281},"root_dir",{"type":59,"value":1283}," spec key points to the top-level directory containing all four subdirectories.",{"type":53,"tag":62,"props":1285,"children":1286},{},[1287,1292,1294,1300,1301,1307,1308,1314,1315],{"type":53,"tag":76,"props":1288,"children":1289},{},[1290],{"type":59,"value":1291},"Required files per dataset:",{"type":59,"value":1293}," ",{"type":53,"tag":99,"props":1295,"children":1297},{"className":1296},[],[1298],{"type":59,"value":1299},"A.tar.gz",{"type":59,"value":394},{"type":53,"tag":99,"props":1302,"children":1304},{"className":1303},[],[1305],{"type":59,"value":1306},"B.tar.gz",{"type":59,"value":394},{"type":53,"tag":99,"props":1309,"children":1311},{"className":1310},[],[1312],{"type":59,"value":1313},"list.tar.gz",{"type":59,"value":394},{"type":53,"tag":99,"props":1316,"children":1318},{"className":1317},[],[1319],{"type":59,"value":1320},"label.tar.gz",{"type":53,"tag":732,"props":1322,"children":1324},{"id":1323},"per-action-dataset-requirements-segment",[1325],{"type":59,"value":1326},"Per-Action Dataset Requirements (Segment)",{"type":53,"tag":62,"props":1328,"children":1329},{},[1330,1331,1336,1337,1342,1343,1348,1349,1354],{"type":59,"value":742},{"type":53,"tag":99,"props":1332,"children":1334},{"className":1333},[],[1335],{"type":59,"value":557},{"type":59,"value":378},{"type":53,"tag":99,"props":1338,"children":1340},{"className":1339},[],[1341],{"type":59,"value":622},{"type":59,"value":755},{"type":53,"tag":99,"props":1344,"children":1346},{"className":1345},[],[1347],{"type":59,"value":360},{"type":59,"value":582},{"type":53,"tag":99,"props":1350,"children":1352},{"className":1351},[],[1353],{"type":59,"value":767},{"type":59,"value":769},{"type":53,"tag":771,"props":1356,"children":1357},{},[1358,1384],{"type":53,"tag":775,"props":1359,"children":1360},{},[1361],{"type":53,"tag":779,"props":1362,"children":1363},{},[1364,1368,1372,1376,1380],{"type":53,"tag":783,"props":1365,"children":1366},{},[1367],{"type":59,"value":787},{"type":53,"tag":783,"props":1369,"children":1370},{},[1371],{"type":59,"value":792},{"type":53,"tag":783,"props":1373,"children":1374},{},[1375],{"type":59,"value":797},{"type":53,"tag":783,"props":1377,"children":1378},{},[1379],{"type":59,"value":802},{"type":53,"tag":783,"props":1381,"children":1382},{},[1383],{"type":59,"value":807},{"type":53,"tag":809,"props":1385,"children":1386},{},[1387,1412,1435,1459,1482,1505,1528],{"type":53,"tag":779,"props":1388,"children":1389},{},[1390,1394,1399,1403,1408],{"type":53,"tag":816,"props":1391,"children":1392},{},[1393],{"type":59,"value":820},{"type":53,"tag":816,"props":1395,"children":1396},{},[1397],{"type":59,"value":1398},"dataset.segment.root_dir",{"type":53,"tag":816,"props":1400,"children":1401},{},[1402],{"type":59,"value":830},{"type":53,"tag":816,"props":1404,"children":1405},{},[1406],{"type":59,"value":1407},"(root directory)",{"type":53,"tag":816,"props":1409,"children":1410},{},[1411],{"type":59,"value":840},{"type":53,"tag":779,"props":1413,"children":1414},{},[1415,1419,1423,1427,1431],{"type":53,"tag":816,"props":1416,"children":1417},{},[1418],{"type":59,"value":557},{"type":53,"tag":816,"props":1420,"children":1421},{},[1422],{"type":59,"value":1398},{"type":53,"tag":816,"props":1424,"children":1425},{},[1426],{"type":59,"value":830},{"type":53,"tag":816,"props":1428,"children":1429},{},[1430],{"type":59,"value":1407},{"type":53,"tag":816,"props":1432,"children":1433},{},[1434],{"type":59,"value":840},{"type":53,"tag":779,"props":1436,"children":1437},{},[1438,1442,1447,1451,1455],{"type":53,"tag":816,"props":1439,"children":1440},{},[1441],{"type":59,"value":557},{"type":53,"tag":816,"props":1443,"children":1444},{},[1445],{"type":59,"value":1446},"dataset.segment.quant_calibration_dataset.images_dir",{"type":53,"tag":816,"props":1448,"children":1449},{},[1450],{"type":59,"value":830},{"type":53,"tag":816,"props":1452,"children":1453},{},[1454],{"type":59,"value":1407},{"type":53,"tag":816,"props":1456,"children":1457},{},[1458],{"type":59,"value":840},{"type":53,"tag":779,"props":1460,"children":1461},{},[1462,1466,1470,1474,1478],{"type":53,"tag":816,"props":1463,"children":1464},{},[1465],{"type":59,"value":535},{"type":53,"tag":816,"props":1467,"children":1468},{},[1469],{"type":59,"value":1398},{"type":53,"tag":816,"props":1471,"children":1472},{},[1473],{"type":59,"value":830},{"type":53,"tag":816,"props":1475,"children":1476},{},[1477],{"type":59,"value":1407},{"type":53,"tag":816,"props":1479,"children":1480},{},[1481],{"type":59,"value":840},{"type":53,"tag":779,"props":1483,"children":1484},{},[1485,1489,1493,1497,1501],{"type":53,"tag":816,"props":1486,"children":1487},{},[1488],{"type":59,"value":542},{"type":53,"tag":816,"props":1490,"children":1491},{},[1492],{"type":59,"value":1398},{"type":53,"tag":816,"props":1494,"children":1495},{},[1496],{"type":59,"value":830},{"type":53,"tag":816,"props":1498,"children":1499},{},[1500],{"type":59,"value":1407},{"type":53,"tag":816,"props":1502,"children":1503},{},[1504],{"type":59,"value":840},{"type":53,"tag":779,"props":1506,"children":1507},{},[1508,1512,1516,1520,1524],{"type":53,"tag":816,"props":1509,"children":1510},{},[1511],{"type":59,"value":622},{"type":53,"tag":816,"props":1513,"children":1514},{},[1515],{"type":59,"value":1398},{"type":53,"tag":816,"props":1517,"children":1518},{},[1519],{"type":59,"value":830},{"type":53,"tag":816,"props":1521,"children":1522},{},[1523],{"type":59,"value":1407},{"type":53,"tag":816,"props":1525,"children":1526},{},[1527],{"type":59,"value":840},{"type":53,"tag":779,"props":1529,"children":1530},{},[1531,1535,1539,1543,1547],{"type":53,"tag":816,"props":1532,"children":1533},{},[1534],{"type":59,"value":622},{"type":53,"tag":816,"props":1536,"children":1537},{},[1538],{"type":59,"value":1185},{"type":53,"tag":816,"props":1540,"children":1541},{},[1542],{"type":59,"value":1190},{"type":53,"tag":816,"props":1544,"children":1545},{},[1546],{"type":59,"value":835},{"type":53,"tag":816,"props":1548,"children":1549},{},[1550],{"type":59,"value":1199},{"type":53,"tag":683,"props":1552,"children":1554},{"id":1553},"typical-spec-overrides",[1555],{"type":59,"value":1556},"Typical Spec Overrides",{"type":53,"tag":62,"props":1558,"children":1559},{},[1560,1562,1567,1569,1575],{"type":59,"value":1561},"Data source overrides are ",{"type":53,"tag":76,"props":1563,"children":1564},{},[1565],{"type":59,"value":1566},"mandatory for every action",{"type":59,"value":1568}," — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in ",{"type":53,"tag":99,"props":1570,"children":1572},{"className":1571},[],[1573],{"type":59,"value":1574},"spec_overrides",{"type":59,"value":769},{"type":53,"tag":187,"props":1577,"children":1581},{"className":1578,"code":1579,"language":1580,"meta":192,"style":192},"language-python shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","S3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\nS3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n","python",[1582],{"type":53,"tag":99,"props":1583,"children":1584},{"__ignoreMap":192},[1585,1593],{"type":53,"tag":198,"props":1586,"children":1587},{"class":200,"line":201},[1588],{"type":53,"tag":198,"props":1589,"children":1590},{},[1591],{"type":59,"value":1592},"S3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\n",{"type":53,"tag":198,"props":1594,"children":1595},{"class":200,"line":234},[1596],{"type":53,"tag":198,"props":1597,"children":1598},{},[1599],{"type":59,"value":1600},"S3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n",{"type":53,"tag":62,"props":1602,"children":1603},{},[1604],{"type":53,"tag":76,"props":1605,"children":1606},{},[1607],{"type":59,"value":1608},"train (classify, mandatory data sources):",{"type":53,"tag":187,"props":1610,"children":1612},{"className":1578,"code":1611,"language":1580,"meta":192,"style":192},"{\n    \"train.num_epochs\": 30,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"train.use_distributed_sampler\": False,\n    \"train.sync_batchnorm\": False,\n    \"dataset.classify.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.classify.train_dataset.csv_path\": f\"{S3_TRAIN}\u002Fdataset.csv\",\n    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n}\n",[1613],{"type":53,"tag":99,"props":1614,"children":1615},{"__ignoreMap":192},[1616,1624,1632,1641,1650,1659,1668,1677,1686,1695,1704,1713],{"type":53,"tag":198,"props":1617,"children":1618},{"class":200,"line":201},[1619],{"type":53,"tag":198,"props":1620,"children":1621},{},[1622],{"type":59,"value":1623},"{\n",{"type":53,"tag":198,"props":1625,"children":1626},{"class":200,"line":234},[1627],{"type":53,"tag":198,"props":1628,"children":1629},{},[1630],{"type":59,"value":1631},"    \"train.num_epochs\": 30,\n",{"type":53,"tag":198,"props":1633,"children":1635},{"class":200,"line":1634},3,[1636],{"type":53,"tag":198,"props":1637,"children":1638},{},[1639],{"type":59,"value":1640},"    \"train.checkpoint_interval\": 10,\n",{"type":53,"tag":198,"props":1642,"children":1644},{"class":200,"line":1643},4,[1645],{"type":53,"tag":198,"props":1646,"children":1647},{},[1648],{"type":59,"value":1649},"    \"train.validation_interval\": 10,\n",{"type":53,"tag":198,"props":1651,"children":1653},{"class":200,"line":1652},5,[1654],{"type":53,"tag":198,"props":1655,"children":1656},{},[1657],{"type":59,"value":1658},"    \"train.num_gpus\": 1,\n",{"type":53,"tag":198,"props":1660,"children":1662},{"class":200,"line":1661},6,[1663],{"type":53,"tag":198,"props":1664,"children":1665},{},[1666],{"type":59,"value":1667},"    \"train.use_distributed_sampler\": False,\n",{"type":53,"tag":198,"props":1669,"children":1671},{"class":200,"line":1670},7,[1672],{"type":53,"tag":198,"props":1673,"children":1674},{},[1675],{"type":59,"value":1676},"    \"train.sync_batchnorm\": False,\n",{"type":53,"tag":198,"props":1678,"children":1680},{"class":200,"line":1679},8,[1681],{"type":53,"tag":198,"props":1682,"children":1683},{},[1684],{"type":59,"value":1685},"    \"dataset.classify.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n",{"type":53,"tag":198,"props":1687,"children":1689},{"class":200,"line":1688},9,[1690],{"type":53,"tag":198,"props":1691,"children":1692},{},[1693],{"type":59,"value":1694},"    \"dataset.classify.train_dataset.csv_path\": f\"{S3_TRAIN}\u002Fdataset.csv\",\n",{"type":53,"tag":198,"props":1696,"children":1698},{"class":200,"line":1697},10,[1699],{"type":53,"tag":198,"props":1700,"children":1701},{},[1702],{"type":59,"value":1703},"    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n",{"type":53,"tag":198,"props":1705,"children":1707},{"class":200,"line":1706},11,[1708],{"type":53,"tag":198,"props":1709,"children":1710},{},[1711],{"type":59,"value":1712},"    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n",{"type":53,"tag":198,"props":1714,"children":1716},{"class":200,"line":1715},12,[1717],{"type":53,"tag":198,"props":1718,"children":1719},{},[1720],{"type":59,"value":1721},"}\n",{"type":53,"tag":62,"props":1723,"children":1724},{},[1725],{"type":53,"tag":76,"props":1726,"children":1727},{},[1728],{"type":59,"value":1729},"train (segment, mandatory data sources):",{"type":53,"tag":187,"props":1731,"children":1733},{"className":1578,"code":1732,"language":1580,"meta":192,"style":192},"{\n    \"train.num_epochs\": 30,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"train.use_distributed_sampler\": False,\n    \"train.sync_batchnorm\": False,\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n}\n",[1734],{"type":53,"tag":99,"props":1735,"children":1736},{"__ignoreMap":192},[1737,1744,1751,1758,1765,1772,1779,1786,1794],{"type":53,"tag":198,"props":1738,"children":1739},{"class":200,"line":201},[1740],{"type":53,"tag":198,"props":1741,"children":1742},{},[1743],{"type":59,"value":1623},{"type":53,"tag":198,"props":1745,"children":1746},{"class":200,"line":234},[1747],{"type":53,"tag":198,"props":1748,"children":1749},{},[1750],{"type":59,"value":1631},{"type":53,"tag":198,"props":1752,"children":1753},{"class":200,"line":1634},[1754],{"type":53,"tag":198,"props":1755,"children":1756},{},[1757],{"type":59,"value":1640},{"type":53,"tag":198,"props":1759,"children":1760},{"class":200,"line":1643},[1761],{"type":53,"tag":198,"props":1762,"children":1763},{},[1764],{"type":59,"value":1649},{"type":53,"tag":198,"props":1766,"children":1767},{"class":200,"line":1652},[1768],{"type":53,"tag":198,"props":1769,"children":1770},{},[1771],{"type":59,"value":1658},{"type":53,"tag":198,"props":1773,"children":1774},{"class":200,"line":1661},[1775],{"type":53,"tag":198,"props":1776,"children":1777},{},[1778],{"type":59,"value":1667},{"type":53,"tag":198,"props":1780,"children":1781},{"class":200,"line":1670},[1782],{"type":53,"tag":198,"props":1783,"children":1784},{},[1785],{"type":59,"value":1676},{"type":53,"tag":198,"props":1787,"children":1788},{"class":200,"line":1679},[1789],{"type":53,"tag":198,"props":1790,"children":1791},{},[1792],{"type":59,"value":1793},"    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n",{"type":53,"tag":198,"props":1795,"children":1796},{"class":200,"line":1688},[1797],{"type":53,"tag":198,"props":1798,"children":1799},{},[1800],{"type":59,"value":1721},{"type":53,"tag":62,"props":1802,"children":1803},{},[1804],{"type":53,"tag":76,"props":1805,"children":1806},{},[1807],{"type":59,"value":1808},"export (classify):",{"type":53,"tag":187,"props":1810,"children":1812},{"className":1578,"code":1811,"language":1580,"meta":192,"style":192},"{\n    \"export.input_height\": 896,\n    \"export.input_width\": 224,\n}\n",[1813],{"type":53,"tag":99,"props":1814,"children":1815},{"__ignoreMap":192},[1816,1823,1831,1839],{"type":53,"tag":198,"props":1817,"children":1818},{"class":200,"line":201},[1819],{"type":53,"tag":198,"props":1820,"children":1821},{},[1822],{"type":59,"value":1623},{"type":53,"tag":198,"props":1824,"children":1825},{"class":200,"line":234},[1826],{"type":53,"tag":198,"props":1827,"children":1828},{},[1829],{"type":59,"value":1830},"    \"export.input_height\": 896,\n",{"type":53,"tag":198,"props":1832,"children":1833},{"class":200,"line":1634},[1834],{"type":53,"tag":198,"props":1835,"children":1836},{},[1837],{"type":59,"value":1838},"    \"export.input_width\": 224,\n",{"type":53,"tag":198,"props":1840,"children":1841},{"class":200,"line":1643},[1842],{"type":53,"tag":198,"props":1843,"children":1844},{},[1845],{"type":59,"value":1721},{"type":53,"tag":62,"props":1847,"children":1848},{},[1849],{"type":53,"tag":76,"props":1850,"children":1851},{},[1852],{"type":59,"value":1853},"export (segment):",{"type":53,"tag":187,"props":1855,"children":1857},{"className":1578,"code":1856,"language":1580,"meta":192,"style":192},"{\n    \"export.input_height\": 224,\n    \"export.input_width\": 224,\n}\n",[1858],{"type":53,"tag":99,"props":1859,"children":1860},{"__ignoreMap":192},[1861,1868,1876,1883],{"type":53,"tag":198,"props":1862,"children":1863},{"class":200,"line":201},[1864],{"type":53,"tag":198,"props":1865,"children":1866},{},[1867],{"type":59,"value":1623},{"type":53,"tag":198,"props":1869,"children":1870},{"class":200,"line":234},[1871],{"type":53,"tag":198,"props":1872,"children":1873},{},[1874],{"type":59,"value":1875},"    \"export.input_height\": 224,\n",{"type":53,"tag":198,"props":1877,"children":1878},{"class":200,"line":1634},[1879],{"type":53,"tag":198,"props":1880,"children":1881},{},[1882],{"type":59,"value":1838},{"type":53,"tag":198,"props":1884,"children":1885},{"class":200,"line":1643},[1886],{"type":53,"tag":198,"props":1887,"children":1888},{},[1889],{"type":59,"value":1721},{"type":53,"tag":62,"props":1891,"children":1892},{},[1893],{"type":53,"tag":76,"props":1894,"children":1895},{},[1896],{"type":59,"value":1897},"quantize (classify, mandatory data sources):",{"type":53,"tag":187,"props":1899,"children":1901},{"className":1578,"code":1900,"language":1580,"meta":192,"style":192},"{\n    \"dataset.classify.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.classify.train_dataset.csv_path\": f\"{S3_TRAIN}\u002Fdataset.csv\",\n    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n    \"dataset.classify.quant_calibration_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n}\n",[1902],{"type":53,"tag":99,"props":1903,"children":1904},{"__ignoreMap":192},[1905,1912,1919,1926,1933,1940,1948],{"type":53,"tag":198,"props":1906,"children":1907},{"class":200,"line":201},[1908],{"type":53,"tag":198,"props":1909,"children":1910},{},[1911],{"type":59,"value":1623},{"type":53,"tag":198,"props":1913,"children":1914},{"class":200,"line":234},[1915],{"type":53,"tag":198,"props":1916,"children":1917},{},[1918],{"type":59,"value":1685},{"type":53,"tag":198,"props":1920,"children":1921},{"class":200,"line":1634},[1922],{"type":53,"tag":198,"props":1923,"children":1924},{},[1925],{"type":59,"value":1694},{"type":53,"tag":198,"props":1927,"children":1928},{"class":200,"line":1643},[1929],{"type":53,"tag":198,"props":1930,"children":1931},{},[1932],{"type":59,"value":1703},{"type":53,"tag":198,"props":1934,"children":1935},{"class":200,"line":1652},[1936],{"type":53,"tag":198,"props":1937,"children":1938},{},[1939],{"type":59,"value":1712},{"type":53,"tag":198,"props":1941,"children":1942},{"class":200,"line":1661},[1943],{"type":53,"tag":198,"props":1944,"children":1945},{},[1946],{"type":59,"value":1947},"    \"dataset.classify.quant_calibration_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n",{"type":53,"tag":198,"props":1949,"children":1950},{"class":200,"line":1670},[1951],{"type":53,"tag":198,"props":1952,"children":1953},{},[1954],{"type":59,"value":1721},{"type":53,"tag":62,"props":1956,"children":1957},{},[1958],{"type":53,"tag":76,"props":1959,"children":1960},{},[1961],{"type":59,"value":1962},"evaluate (classify, mandatory data sources):",{"type":53,"tag":187,"props":1964,"children":1966},{"className":1578,"code":1965,"language":1580,"meta":192,"style":192},"{\n    \"dataset.classify.validation_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.validation_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n    \"dataset.classify.test_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.test_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n}\n",[1967],{"type":53,"tag":99,"props":1968,"children":1969},{"__ignoreMap":192},[1970,1977,1984,1991,1999,2007],{"type":53,"tag":198,"props":1971,"children":1972},{"class":200,"line":201},[1973],{"type":53,"tag":198,"props":1974,"children":1975},{},[1976],{"type":59,"value":1623},{"type":53,"tag":198,"props":1978,"children":1979},{"class":200,"line":234},[1980],{"type":53,"tag":198,"props":1981,"children":1982},{},[1983],{"type":59,"value":1703},{"type":53,"tag":198,"props":1985,"children":1986},{"class":200,"line":1634},[1987],{"type":53,"tag":198,"props":1988,"children":1989},{},[1990],{"type":59,"value":1712},{"type":53,"tag":198,"props":1992,"children":1993},{"class":200,"line":1643},[1994],{"type":53,"tag":198,"props":1995,"children":1996},{},[1997],{"type":59,"value":1998},"    \"dataset.classify.test_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n",{"type":53,"tag":198,"props":2000,"children":2001},{"class":200,"line":1652},[2002],{"type":53,"tag":198,"props":2003,"children":2004},{},[2005],{"type":59,"value":2006},"    \"dataset.classify.test_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n",{"type":53,"tag":198,"props":2008,"children":2009},{"class":200,"line":1661},[2010],{"type":53,"tag":198,"props":2011,"children":2012},{},[2013],{"type":59,"value":1721},{"type":53,"tag":62,"props":2015,"children":2016},{},[2017],{"type":53,"tag":76,"props":2018,"children":2019},{},[2020],{"type":59,"value":2021},"inference (classify, mandatory data sources):",{"type":53,"tag":187,"props":2023,"children":2025},{"className":1578,"code":2024,"language":1580,"meta":192,"style":192},"{\n    \"dataset.classify.infer_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.classify.infer_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n}\n",[2026],{"type":53,"tag":99,"props":2027,"children":2028},{"__ignoreMap":192},[2029,2036,2044,2052],{"type":53,"tag":198,"props":2030,"children":2031},{"class":200,"line":201},[2032],{"type":53,"tag":198,"props":2033,"children":2034},{},[2035],{"type":59,"value":1623},{"type":53,"tag":198,"props":2037,"children":2038},{"class":200,"line":234},[2039],{"type":53,"tag":198,"props":2040,"children":2041},{},[2042],{"type":59,"value":2043},"    \"dataset.classify.infer_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n",{"type":53,"tag":198,"props":2045,"children":2046},{"class":200,"line":1634},[2047],{"type":53,"tag":198,"props":2048,"children":2049},{},[2050],{"type":59,"value":2051},"    \"dataset.classify.infer_dataset.csv_path\": f\"{S3_EVAL}\u002Fdataset.csv\",\n",{"type":53,"tag":198,"props":2053,"children":2054},{"class":200,"line":1643},[2055],{"type":53,"tag":198,"props":2056,"children":2057},{},[2058],{"type":59,"value":1721},{"type":53,"tag":62,"props":2060,"children":2061},{},[2062],{"type":53,"tag":76,"props":2063,"children":2064},{},[2065],{"type":59,"value":2066},"gen_trt_engine (classify, mandatory data sources):",{"type":53,"tag":187,"props":2068,"children":2070},{"className":1578,"code":2069,"language":1580,"meta":192,"style":192},"{\n    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n}\n",[2071],{"type":53,"tag":99,"props":2072,"children":2073},{"__ignoreMap":192},[2074,2081,2089],{"type":53,"tag":198,"props":2075,"children":2076},{"class":200,"line":201},[2077],{"type":53,"tag":198,"props":2078,"children":2079},{},[2080],{"type":59,"value":1623},{"type":53,"tag":198,"props":2082,"children":2083},{"class":200,"line":234},[2084],{"type":53,"tag":198,"props":2085,"children":2086},{},[2087],{"type":59,"value":2088},"    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n",{"type":53,"tag":198,"props":2090,"children":2091},{"class":200,"line":1634},[2092],{"type":53,"tag":198,"props":2093,"children":2094},{},[2095],{"type":59,"value":1721},{"type":53,"tag":62,"props":2097,"children":2098},{},[2099],{"type":53,"tag":76,"props":2100,"children":2101},{},[2102],{"type":59,"value":2103},"quantize (segment, mandatory data sources):",{"type":53,"tag":187,"props":2105,"children":2107},{"className":1578,"code":2106,"language":1580,"meta":192,"style":192},"{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n    \"dataset.segment.quant_calibration_dataset.images_dir\": f\"{S3_TRAIN}\",\n}\n",[2108],{"type":53,"tag":99,"props":2109,"children":2110},{"__ignoreMap":192},[2111,2118,2125,2133],{"type":53,"tag":198,"props":2112,"children":2113},{"class":200,"line":201},[2114],{"type":53,"tag":198,"props":2115,"children":2116},{},[2117],{"type":59,"value":1623},{"type":53,"tag":198,"props":2119,"children":2120},{"class":200,"line":234},[2121],{"type":53,"tag":198,"props":2122,"children":2123},{},[2124],{"type":59,"value":1793},{"type":53,"tag":198,"props":2126,"children":2127},{"class":200,"line":1634},[2128],{"type":53,"tag":198,"props":2129,"children":2130},{},[2131],{"type":59,"value":2132},"    \"dataset.segment.quant_calibration_dataset.images_dir\": f\"{S3_TRAIN}\",\n",{"type":53,"tag":198,"props":2134,"children":2135},{"class":200,"line":1643},[2136],{"type":53,"tag":198,"props":2137,"children":2138},{},[2139],{"type":59,"value":1721},{"type":53,"tag":62,"props":2141,"children":2142},{},[2143],{"type":53,"tag":76,"props":2144,"children":2145},{},[2146],{"type":59,"value":2147},"evaluate (segment, mandatory data sources):",{"type":53,"tag":187,"props":2149,"children":2151},{"className":1578,"code":2150,"language":1580,"meta":192,"style":192},"{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n}\n",[2152],{"type":53,"tag":99,"props":2153,"children":2154},{"__ignoreMap":192},[2155,2162,2169],{"type":53,"tag":198,"props":2156,"children":2157},{"class":200,"line":201},[2158],{"type":53,"tag":198,"props":2159,"children":2160},{},[2161],{"type":59,"value":1623},{"type":53,"tag":198,"props":2163,"children":2164},{"class":200,"line":234},[2165],{"type":53,"tag":198,"props":2166,"children":2167},{},[2168],{"type":59,"value":1793},{"type":53,"tag":198,"props":2170,"children":2171},{"class":200,"line":1634},[2172],{"type":53,"tag":198,"props":2173,"children":2174},{},[2175],{"type":59,"value":1721},{"type":53,"tag":62,"props":2177,"children":2178},{},[2179],{"type":53,"tag":76,"props":2180,"children":2181},{},[2182],{"type":59,"value":2183},"inference (segment, mandatory data sources):",{"type":53,"tag":187,"props":2185,"children":2186},{"className":1578,"code":2150,"language":1580,"meta":192,"style":192},[2187],{"type":53,"tag":99,"props":2188,"children":2189},{"__ignoreMap":192},[2190,2197,2204],{"type":53,"tag":198,"props":2191,"children":2192},{"class":200,"line":201},[2193],{"type":53,"tag":198,"props":2194,"children":2195},{},[2196],{"type":59,"value":1623},{"type":53,"tag":198,"props":2198,"children":2199},{"class":200,"line":234},[2200],{"type":53,"tag":198,"props":2201,"children":2202},{},[2203],{"type":59,"value":1793},{"type":53,"tag":198,"props":2205,"children":2206},{"class":200,"line":1634},[2207],{"type":53,"tag":198,"props":2208,"children":2209},{},[2210],{"type":59,"value":1721},{"type":53,"tag":62,"props":2212,"children":2213},{},[2214],{"type":53,"tag":76,"props":2215,"children":2216},{},[2217],{"type":59,"value":2218},"gen_trt_engine (segment, mandatory data sources):",{"type":53,"tag":187,"props":2220,"children":2222},{"className":1578,"code":2221,"language":1580,"meta":192,"style":192},"{\n    \"dataset.segment.root_dir\": f\"{S3_TRAIN}\",\n    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n}\n",[2223],{"type":53,"tag":99,"props":2224,"children":2225},{"__ignoreMap":192},[2226,2233,2240,2247],{"type":53,"tag":198,"props":2227,"children":2228},{"class":200,"line":201},[2229],{"type":53,"tag":198,"props":2230,"children":2231},{},[2232],{"type":59,"value":1623},{"type":53,"tag":198,"props":2234,"children":2235},{"class":200,"line":234},[2236],{"type":53,"tag":198,"props":2237,"children":2238},{},[2239],{"type":59,"value":1793},{"type":53,"tag":198,"props":2241,"children":2242},{"class":200,"line":1634},[2243],{"type":53,"tag":198,"props":2244,"children":2245},{},[2246],{"type":59,"value":2088},{"type":53,"tag":198,"props":2248,"children":2249},{"class":200,"line":1643},[2250],{"type":53,"tag":198,"props":2251,"children":2252},{},[2253],{"type":59,"value":1721},{"type":53,"tag":312,"props":2255,"children":2257},{"id":2256},"optional-running-via-the-tao-sdk",[2258],{"type":59,"value":2259},"Optional: running via the TAO SDK",{"type":53,"tag":62,"props":2261,"children":2262},{},[2263,2265,2271,2273,2279,2281,2287,2289,2295,2297,2303,2305,2311,2313,2319],{"type":59,"value":2264},"When running without the TAO SDK (local docker), resolve the TAO pyt image from ",{"type":53,"tag":99,"props":2266,"children":2268},{"className":2267},[],[2269],{"type":59,"value":2270},"versions.yaml",{"type":59,"value":2272}," and invoke ",{"type":53,"tag":99,"props":2274,"children":2276},{"className":2275},[],[2277],{"type":59,"value":2278},"visual_changenet \u003Ctrain|evaluate|inference|export|quantize>",{"type":59,"value":2280}," directly. ",{"type":53,"tag":99,"props":2282,"children":2284},{"className":2283},[],[2285],{"type":59,"value":2286},"--shm-size=8g",{"type":59,"value":2288}," is required, the C-RADIO ",{"type":53,"tag":99,"props":2290,"children":2292},{"className":2291},[],[2293],{"type":59,"value":2294},".safetensors",{"type":59,"value":2296}," must be mounted to ",{"type":53,"tag":99,"props":2298,"children":2300},{"className":2299},[],[2301],{"type":59,"value":2302},"\u002Fdata\u002Fpretrained_models\u002FC-RADIOv2_B.safetensors",{"type":59,"value":2304},", and checkpoint\u002Fresults_dir can be overridden on the command line. See ",{"type":53,"tag":99,"props":2306,"children":2308},{"className":2307},[],[2309],{"type":59,"value":2310},"references\u002Flocal-docker.md",{"type":59,"value":2312}," for the full ",{"type":53,"tag":99,"props":2314,"children":2316},{"className":2315},[],[2317],{"type":59,"value":2318},"docker run",{"type":59,"value":2320}," command, mounts, and overrides.",{"type":53,"tag":312,"props":2322,"children":2324},{"id":2323},"tasks",[2325],{"type":59,"value":2326},"Tasks",{"type":53,"tag":683,"props":2328,"children":2330},{"id":2329},"classify-default",[2331],{"type":59,"value":2332},"Classify (default)",{"type":53,"tag":62,"props":2334,"children":2335},{},[2336,2338,2343,2344,2349,2350,2355,2357,2362],{"type":59,"value":2337},"Uses actions: ",{"type":53,"tag":99,"props":2339,"children":2341},{"className":2340},[],[2342],{"type":59,"value":820},{"type":59,"value":394},{"type":53,"tag":99,"props":2345,"children":2347},{"className":2346},[],[2348],{"type":59,"value":535},{"type":59,"value":394},{"type":53,"tag":99,"props":2351,"children":2353},{"className":2352},[],[2354],{"type":59,"value":542},{"type":59,"value":2356},". Defaults template: ",{"type":53,"tag":99,"props":2358,"children":2360},{"className":2359},[],[2361],{"type":59,"value":384},{"type":59,"value":769},{"type":53,"tag":683,"props":2364,"children":2366},{"id":2365},"segment-1",[2367],{"type":59,"value":90},{"type":53,"tag":62,"props":2369,"children":2370},{},[2371,2373,2379,2380,2385,2387,2392,2394,2399,2400,2405,2406,2411,2413,2419,2421,2427,2428,2434,2435,2441,2443,2449],{"type":59,"value":2372},"Uses skill action names ",{"type":53,"tag":99,"props":2374,"children":2376},{"className":2375},[],[2377],{"type":59,"value":2378},"segment_train",{"type":59,"value":394},{"type":53,"tag":99,"props":2381,"children":2383},{"className":2382},[],[2384],{"type":59,"value":564},{"type":59,"value":2386},", and\n",{"type":53,"tag":99,"props":2388,"children":2390},{"className":2389},[],[2391],{"type":59,"value":572},{"type":59,"value":2393},". When invoking local Docker directly, run TAO CLI subcommands\n",{"type":53,"tag":99,"props":2395,"children":2397},{"className":2396},[],[2398],{"type":59,"value":820},{"type":59,"value":394},{"type":53,"tag":99,"props":2401,"children":2403},{"className":2402},[],[2404],{"type":59,"value":535},{"type":59,"value":566},{"type":53,"tag":99,"props":2407,"children":2409},{"className":2408},[],[2410],{"type":59,"value":542},{"type":59,"value":2412}," with ",{"type":53,"tag":99,"props":2414,"children":2416},{"className":2415},[],[2417],{"type":59,"value":2418},"task: segment",{"type":59,"value":2420}," in the spec. The\nschema-driven action templates are ",{"type":53,"tag":99,"props":2422,"children":2424},{"className":2423},[],[2425],{"type":59,"value":2426},"references\u002Fspec_template_segment_train.yaml",{"type":59,"value":544},{"type":53,"tag":99,"props":2429,"children":2431},{"className":2430},[],[2432],{"type":59,"value":2433},"references\u002Fspec_template_segment_evaluate.yaml",{"type":59,"value":2386},{"type":53,"tag":99,"props":2436,"children":2438},{"className":2437},[],[2439],{"type":59,"value":2440},"references\u002Fspec_template_segment_inference.yaml",{"type":59,"value":2442},"; the compact direct-Docker\nexample template is ",{"type":53,"tag":99,"props":2444,"children":2446},{"className":2445},[],[2447],{"type":59,"value":2448},"references\u002Fspec_template_segment.yaml",{"type":59,"value":769},{"type":53,"tag":62,"props":2451,"children":2452},{},[2453,2455,2461],{"type":59,"value":2454},"Segmentation requires compiling custom CUDA ops (",{"type":53,"tag":99,"props":2456,"children":2458},{"className":2457},[],[2459],{"type":59,"value":2460},"MultiScaleDeformableAttention",{"type":59,"value":2462},") on first run, which takes ~5 minutes. The ViT adapter backbone uses these for multi-scale feature extraction.",{"type":53,"tag":62,"props":2464,"children":2465},{},[2466,2468,2473,2474,2479,2480,2485,2486,2491,2493,2498],{"type":59,"value":2467},"Dataset structure for segmentation differs from classify — uses paired directories (",{"type":53,"tag":99,"props":2469,"children":2471},{"className":2470},[],[2472],{"type":59,"value":1252},{"type":59,"value":394},{"type":53,"tag":99,"props":2475,"children":2477},{"className":2476},[],[2478],{"type":59,"value":1259},{"type":59,"value":394},{"type":53,"tag":99,"props":2481,"children":2483},{"className":2482},[],[2484],{"type":59,"value":1266},{"type":59,"value":394},{"type":53,"tag":99,"props":2487,"children":2489},{"className":2488},[],[2490],{"type":59,"value":1273},{"type":59,"value":2492},") instead of CSV files. See ",{"type":53,"tag":99,"props":2494,"children":2496},{"className":2495},[],[2497],{"type":59,"value":1398},{"type":59,"value":2499}," in the defaults.",{"type":53,"tag":312,"props":2501,"children":2503},{"id":2502},"data-format",[2504],{"type":59,"value":2505},"Data Format",{"type":53,"tag":62,"props":2507,"children":2508},{},[2509,2511,2517,2519,2524,2525,2530,2531,2536,2537,2542,2544,2549,2551,2557,2559,2565,2567,2573,2575,2581,2583,2589,2591,2597,2599,2605,2607,2613,2615,2621],{"type":59,"value":2510},"Classify needs a 4-column CSV (",{"type":53,"tag":99,"props":2512,"children":2514},{"className":2513},[],[2515],{"type":59,"value":2516},"input_path,golden_path,label,object_name",{"type":59,"value":2518},") plus an images directory; segment uses a paired directory structure (",{"type":53,"tag":99,"props":2520,"children":2522},{"className":2521},[],[2523],{"type":59,"value":1252},{"type":59,"value":394},{"type":53,"tag":99,"props":2526,"children":2528},{"className":2527},[],[2529],{"type":59,"value":1259},{"type":59,"value":394},{"type":53,"tag":99,"props":2532,"children":2534},{"className":2533},[],[2535],{"type":59,"value":1266},{"type":59,"value":394},{"type":53,"tag":99,"props":2538,"children":2540},{"className":2539},[],[2541],{"type":59,"value":1273},{"type":59,"value":2543},") under ",{"type":53,"tag":99,"props":2545,"children":2547},{"className":2546},[],[2548],{"type":59,"value":1398},{"type":59,"value":2550}," instead of CSV. The ",{"type":53,"tag":99,"props":2552,"children":2554},{"className":2553},[],[2555],{"type":59,"value":2556},"image_ext",{"type":59,"value":2558}," field (default ",{"type":53,"tag":99,"props":2560,"children":2562},{"className":2561},[],[2563],{"type":59,"value":2564},".jpg",{"type":59,"value":2566},") must match the actual file extensions; if images are ",{"type":53,"tag":99,"props":2568,"children":2570},{"className":2569},[],[2571],{"type":59,"value":2572},".png",{"type":59,"value":2574},", set ",{"type":53,"tag":99,"props":2576,"children":2578},{"className":2577},[],[2579],{"type":59,"value":2580},"dataset.classify.image_ext: .png",{"type":59,"value":2582},". Multi-lighting input is configured via ",{"type":53,"tag":99,"props":2584,"children":2586},{"className":2585},[],[2587],{"type":59,"value":2588},"dataset.classify.input_map",{"type":59,"value":2590}," (each lighting name maps to a channel index) with ",{"type":53,"tag":99,"props":2592,"children":2594},{"className":2593},[],[2595],{"type":59,"value":2596},"dataset.classify.num_input",{"type":59,"value":2598}," set to match. See ",{"type":53,"tag":99,"props":2600,"children":2602},{"className":2601},[],[2603],{"type":59,"value":2604},"references\u002Fdata-formats.md",{"type":59,"value":2606}," for the per-field input tables (classify train\u002Feval\u002Finference, segment), CSV column semantics, lighting\u002Fpath-concatenation conventions, the segment directory layout, and ",{"type":53,"tag":99,"props":2608,"children":2610},{"className":2609},[],[2611],{"type":59,"value":2612},"input_map",{"type":59,"value":2614},"\u002F",{"type":53,"tag":99,"props":2616,"children":2618},{"className":2617},[],[2619],{"type":59,"value":2620},"grid_map",{"type":59,"value":2622}," examples.",{"type":53,"tag":312,"props":2624,"children":2626},{"id":2625},"important-parameters",[2627],{"type":59,"value":2628},"Important Parameters",{"type":53,"tag":62,"props":2630,"children":2631},{},[2632,2634,2640,2642,2648,2650,2656,2658,2664,2666,2672,2674,2680,2682,2688,2690,2696,2698,2704,2706,2711,2713,2719,2721,2727,2729,2735],{"type":59,"value":2633},"Key knobs include ",{"type":53,"tag":99,"props":2635,"children":2637},{"className":2636},[],[2638],{"type":59,"value":2639},"train.validation_interval",{"type":59,"value":2641}," (default 50, must be ≤ num_epochs), ",{"type":53,"tag":99,"props":2643,"children":2645},{"className":2644},[],[2646],{"type":59,"value":2647},"train.checkpoint_interval",{"type":59,"value":2649}," (default 200, must be ≤ num_epochs), ",{"type":53,"tag":99,"props":2651,"children":2653},{"className":2652},[],[2654],{"type":59,"value":2655},"train.num_epochs",{"type":59,"value":2657}," (default 100), ",{"type":53,"tag":99,"props":2659,"children":2661},{"className":2660},[],[2662],{"type":59,"value":2663},"model.classify.eval_margin",{"type":59,"value":2665}," (default 0.3, the precision\u002Frecall threshold), ",{"type":53,"tag":99,"props":2667,"children":2669},{"className":2668},[],[2670],{"type":59,"value":2671},"model.classify.train_margin_euclid",{"type":59,"value":2673}," (default 2.0), ",{"type":53,"tag":99,"props":2675,"children":2677},{"className":2676},[],[2678],{"type":59,"value":2679},"model.classify.embedding_vectors",{"type":59,"value":2681}," (default 5), ",{"type":53,"tag":99,"props":2683,"children":2685},{"className":2684},[],[2686],{"type":59,"value":2687},"dataset.classify.batch_size",{"type":59,"value":2689}," (default 16, must be > 1), ",{"type":53,"tag":99,"props":2691,"children":2693},{"className":2692},[],[2694],{"type":59,"value":2695},"dataset.classify.fpratio_sampling",{"type":59,"value":2697}," (default 0.25), and ",{"type":53,"tag":99,"props":2699,"children":2701},{"className":2700},[],[2702],{"type":59,"value":2703},"train.classify.cls_weight",{"type":59,"value":2705}," (default ",{"type":53,"tag":198,"props":2707,"children":2708},{},[2709],{"type":59,"value":2710},"1.0, 10.0",{"type":59,"value":2712},"). Hardware: minimum 1 GPU with 16GB+ VRAM, recommended 8 GPUs (DDP); do not set ",{"type":53,"tag":99,"props":2714,"children":2716},{"className":2715},[],[2717],{"type":59,"value":2718},"gpu_spec_key",{"type":59,"value":2720}," (GPU count is managed internally by TAO), ",{"type":53,"tag":99,"props":2722,"children":2724},{"className":2723},[],[2725],{"type":59,"value":2726},"num_nodes",{"type":59,"value":2728}," (default 1) controls multi-node. See ",{"type":53,"tag":99,"props":2730,"children":2732},{"className":2731},[],[2733],{"type":59,"value":2734},"references\u002Ftuning-parameters.md",{"type":59,"value":2736}," for the full per-parameter guidance and hardware detail.",{"type":53,"tag":312,"props":2738,"children":2740},{"id":2739},"error-patterns",[2741],{"type":59,"value":2742},"Error Patterns",{"type":53,"tag":62,"props":2744,"children":2745},{},[2746,2748,2754,2756,2761,2762,2768,2770,2776,2777,2783,2785,2791],{"type":59,"value":2747},"For checkpoint-not-found, CSV format mismatch, image extension mismatch, OOM, low evaluation accuracy, the contrastive-loss ",{"type":53,"tag":99,"props":2749,"children":2751},{"className":2750},[],[2752],{"type":59,"value":2753},"AssertionError",{"type":59,"value":2755},", checkpoint load key mismatch at evaluate\u002Finference, non-convergence, segment-only backbone dimension mismatch, the ",{"type":53,"tag":99,"props":2757,"children":2759},{"className":2758},[],[2760],{"type":59,"value":2460},{"type":59,"value":1293},{"type":53,"tag":99,"props":2763,"children":2765},{"className":2764},[],[2766],{"type":59,"value":2767},"OSError",{"type":59,"value":2769},", the Lightning ",{"type":53,"tag":99,"props":2771,"children":2773},{"className":2772},[],[2774],{"type":59,"value":2775},"MisconfigurationException",{"type":59,"value":394},{"type":53,"tag":99,"props":2778,"children":2780},{"className":2779},[],[2781],{"type":59,"value":2782},"ModuleNotFoundError: nvidia_tao_pytorch",{"type":59,"value":2784},", and epoch defaults, see ",{"type":53,"tag":99,"props":2786,"children":2788},{"className":2787},[],[2789],{"type":59,"value":2790},"references\u002Ftroubleshooting.md",{"type":59,"value":2792}," for the full symptom-and-fix list.",{"type":53,"tag":312,"props":2794,"children":2796},{"id":2795},"spec-param-parent-model-inference",[2797],{"type":59,"value":2798},"Spec Param \u002F Parent Model Inference",{"type":53,"tag":62,"props":2800,"children":2801},{},[2802,2804,2809,2811,2817,2819,2825,2827,2832,2833,2839,2841,2847,2849,2855],{"type":59,"value":2803},"Model-specific parent-model mappings are declared in ",{"type":53,"tag":99,"props":2805,"children":2807},{"className":2806},[],[2808],{"type":59,"value":360},{"type":59,"value":2810}," under ",{"type":53,"tag":99,"props":2812,"children":2814},{"className":2813},[],[2815],{"type":59,"value":2816},"spec_params",{"type":59,"value":2818},", so generated runners and agents resolve checkpoints before ",{"type":53,"tag":99,"props":2820,"children":2822},{"className":2821},[],[2823],{"type":59,"value":2824},"create_job()",{"type":59,"value":2826}," instead of guessing file names. For ",{"type":53,"tag":99,"props":2828,"children":2830},{"className":2829},[],[2831],{"type":59,"value":668},{"type":59,"value":582},{"type":53,"tag":99,"props":2834,"children":2836},{"className":2835},[],[2837],{"type":59,"value":2838},"parent_model_folder",{"type":59,"value":2840},", pass the upstream train\u002Fexport\u002FAutoML child job id as ",{"type":53,"tag":99,"props":2842,"children":2844},{"className":2843},[],[2845],{"type":59,"value":2846},"parent_job_id",{"type":59,"value":2848},"; the SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. See ",{"type":53,"tag":99,"props":2850,"children":2852},{"className":2851},[],[2853],{"type":59,"value":2854},"references\u002Fparent-model-inference.md",{"type":59,"value":2856}," for the full per-action spec-field-to-inference-function mapping table.",{"type":53,"tag":312,"props":2858,"children":2860},{"id":2859},"deployment",[2861],{"type":59,"value":2862},"Deployment",{"type":53,"tag":68,"props":2864,"children":2865},{},[2866],{"type":53,"tag":72,"props":2867,"children":2868},{},[2869],{"type":53,"tag":2870,"props":2871,"children":2872},"a",{"href":644},[2873],{"type":59,"value":2874},"tao-deploy-visual-changenet",{"type":53,"tag":2876,"props":2877,"children":2878},"style",{},[2879],{"type":59,"value":2880},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":2882,"total":2982},[2883,2900,2912,2926,2938,2953,2968],{"slug":2884,"name":2884,"fn":2885,"description":2886,"org":2887,"tags":2888,"stars":23,"repoUrl":24,"updatedAt":2899},"accelerated-computing-cudf","accelerate data processing with cuDF","Official NVIDIA-authored guidance for NVIDIA cuDF GPU DataFrames, pandas acceleration, dask-cuDF, ETL, joins, groupby, CSV\u002FParquet I\u002FO, nullable semantics, and multi-GPU DataFrame workloads.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2889,2892,2895,2896],{"name":2890,"slug":2891,"type":15},"Data Analysis","data-analysis",{"name":2893,"slug":2894,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2897,"slug":2898,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2901,"name":2901,"fn":2902,"description":2903,"org":2904,"tags":2905,"stars":23,"repoUrl":24,"updatedAt":2911},"aiq-deploy","deploy and manage NVIDIA AI-Q infrastructure","Use when asked to install, deploy, run, validate, troubleshoot, or stop NVIDIA AI-Q Blueprint infrastructure.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2906,2907,2910],{"name":2862,"slug":2859,"type":15},{"name":2908,"slug":2909,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2913,"name":2913,"fn":2914,"description":2915,"org":2916,"tags":2917,"stars":23,"repoUrl":24,"updatedAt":2925},"aiq-research","conduct deep research with AI-Q","Use when asked to run deep research or AI-Q research through a reachable NVIDIA AI-Q Blueprint backend.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2918,2921,2922],{"name":2919,"slug":2920,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2923,"slug":2924,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2927,"name":2927,"fn":2928,"description":2929,"org":2930,"tags":2931,"stars":23,"repoUrl":24,"updatedAt":2937},"amc-run-sample-calibration","run AMC sample dataset calibration","Run end-to-end calibration on the shipped sample dataset (sdg_08_2_sample_data_010926.zip) against a running AMC microservice. Use when user says 'test sample dataset', 'run sample calibration', 'verify AMC install', or 'launch and test'.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2932,2933,2934],{"name":2890,"slug":2891,"type":15},{"name":9,"slug":8,"type":15},{"name":2935,"slug":2936,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":2939,"name":2939,"fn":2940,"description":2941,"org":2942,"tags":2943,"stars":23,"repoUrl":24,"updatedAt":2952},"amc-run-video-calibration","calibrate video datasets with AutoMagicCalib","Calibrate a new dataset from pre-recorded video files via the AutoMagicCalib REST API. Use when user has local MP4s and says 'calibrate my videos', 'run AMC on these videos', or similar. For RTSP\u002Flive streams, use amc-run-rtsp-calibration instead.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2944,2945,2948,2949],{"name":17,"slug":18,"type":15},{"name":2946,"slug":2947,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":2950,"slug":2951,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2954,"name":2954,"fn":2955,"description":2956,"org":2957,"tags":2958,"stars":23,"repoUrl":24,"updatedAt":2967},"amc-setup-calibration-stack","deploy AutoMagicCalib microservice with Docker","Launch AutoMagicCalib microservice and web UI from NGC release images via Docker Compose. Use when user says 'deploy auto calibration', 'launch auto calibration', 'launch AMC', 'start MS+UI', or 'set up auto-magic-calib'. Requires NGC API key.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2959,2960,2963,2964],{"name":2862,"slug":2859,"type":15},{"name":2961,"slug":2962,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2965,"slug":2966,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2969,"name":2969,"fn":2970,"description":2971,"org":2972,"tags":2973,"stars":23,"repoUrl":24,"updatedAt":2981},"cudaq-guide","develop quantum applications with CUDA-Q","CUDA-Q onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2974,2975,2978],{"name":9,"slug":8,"type":15},{"name":2976,"slug":2977,"type":15},"Quantum Computing","quantum-computing",{"name":2979,"slug":2980,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":2984,"total":3134},[2985,3003,3018,3029,3041,3055,3068,3082,3093,3102,3116,3125],{"slug":2986,"name":2986,"fn":2987,"description":2988,"org":2989,"tags":2990,"stars":3000,"repoUrl":3001,"updatedAt":3002},"nemoclaw-user-guide","retrieve NemoClaw documentation and configuration","Guides human users' AI agents to the NemoClaw docs MCP server and canonical Fern documentation in Markdown form. Use when users ask how to install, configure, operate, troubleshoot, secure, or learn NemoClaw with an AI coding assistant. Trigger keywords - nemoclaw docs, use nemoclaw with ai agent, nemoclaw mcp docs, nemoclaw install help, nemoclaw quickstart, nemoclaw markdown docs, llms.txt, agent skills.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[2991,2994,2997],{"name":2992,"slug":2993,"type":15},"Documentation","documentation",{"name":2995,"slug":2996,"type":15},"MCP","mcp",{"name":2998,"slug":2999,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":3004,"name":3004,"fn":3005,"description":3006,"org":3007,"tags":3008,"stars":3015,"repoUrl":3016,"updatedAt":3017},"mcore-build-and-dependency","manage Megatron-LM development environments","Container-based dev environment setup and dependency management for Megatron-LM. Covers acquiring and launching the CI container, uv package management, and updating uv.lock.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3009,3012,3013],{"name":3010,"slug":3011,"type":15},"Containers","containers",{"name":2862,"slug":2859,"type":15},{"name":3014,"slug":1580,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":3019,"name":3019,"fn":3020,"description":3021,"org":3022,"tags":3023,"stars":3015,"repoUrl":3016,"updatedAt":3028},"mcore-bump-base-image","update NVIDIA PyTorch base images","Bump the NVIDIA PyTorch base image (`nvcr.io\u002Fnvidia\u002Fpytorch:YY.MM-py3`) used by Megatron-LM CI. Covers the two pin sites (GitHub CI in `docker\u002F.ngc_version.dev` and GitLab CI in `.gitlab\u002Fstages\u002F01.build.yml`), the post-bump CI loop (re-run functional tests, refresh golden values, mark broken tests), and the gotchas that bit PRs",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3024,3027],{"name":3025,"slug":3026,"type":15},"CI\u002FCD","ci-cd",{"name":2862,"slug":2859,"type":15},"2026-07-14T05:25:59.97109",{"slug":3030,"name":3030,"fn":3031,"description":3032,"org":3033,"tags":3034,"stars":3015,"repoUrl":3016,"updatedAt":3040},"mcore-cicd","manage CI\u002FCD pipelines for Megatron-LM","CI\u002FCD reference for Megatron-LM. Covers CI pipeline structure, PR scope labels, triggering internal GitLab CI (which force-pushes the current branch to a pull-request\u002FBRANCH ref — always dry-run and verify the destination first; never run against shared or protected branches), and CI failure investigation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3035,3036,3037],{"name":3025,"slug":3026,"type":15},{"name":2862,"slug":2859,"type":15},{"name":3038,"slug":3039,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":3042,"name":3042,"fn":3043,"description":3044,"org":3045,"tags":3046,"stars":3015,"repoUrl":3016,"updatedAt":3054},"mcore-create-issue","investigate CI failures and create issues","Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3047,3050,3051],{"name":3048,"slug":3049,"type":15},"Debugging","debugging",{"name":3038,"slug":3039,"type":15},{"name":3052,"slug":3053,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":3056,"name":3056,"fn":3057,"description":3058,"org":3059,"tags":3060,"stars":3015,"repoUrl":3016,"updatedAt":3067},"mcore-linting-and-formatting","lint and format Megatron-LM code","Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3061,3064],{"name":3062,"slug":3063,"type":15},"Best Practices","best-practices",{"name":3065,"slug":3066,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":3069,"name":3069,"fn":3070,"description":3071,"org":3072,"tags":3073,"stars":3015,"repoUrl":3016,"updatedAt":3081},"mcore-migrate-gpt-to-hybrid","migrate Megatron-LM models to HybridModel","Migration guide for moving Megatron Core GPTModel checkpoints, model providers, training commands, and layer mappings to HybridModel.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3074,3077,3080],{"name":3075,"slug":3076,"type":15},"Machine Learning","machine-learning",{"name":3078,"slug":3079,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":3083,"name":3083,"fn":3084,"description":3085,"org":3086,"tags":3087,"stars":3015,"repoUrl":3016,"updatedAt":3092},"mcore-onboard-gb200-1node-tests","onboard functional tests for GB200","Onboard 1-node GitHub MR functional tests for GB200 from existing mr-scoped 2-node tests.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3088,3091],{"name":3089,"slug":3090,"type":15},"QA","qa",{"name":2935,"slug":2936,"type":15},"2026-07-14T05:25:53.673039",{"slug":3094,"name":3094,"fn":3095,"description":3096,"org":3097,"tags":3098,"stars":3015,"repoUrl":3016,"updatedAt":3101},"mcore-run-on-slurm","launch distributed training jobs on SLURM","How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3099,3100],{"name":2862,"slug":2859,"type":15},{"name":2908,"slug":2909,"type":15},"2026-07-14T05:25:49.362534",{"slug":3103,"name":3103,"fn":3104,"description":3105,"org":3106,"tags":3107,"stars":3015,"repoUrl":3016,"updatedAt":3115},"mcore-split-pr","split pull requests to reduce review load","Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3108,3111,3112],{"name":3109,"slug":3110,"type":15},"Code Review","code-review",{"name":3038,"slug":3039,"type":15},{"name":3113,"slug":3114,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":3117,"name":3117,"fn":3118,"description":3119,"org":3120,"tags":3121,"stars":3015,"repoUrl":3016,"updatedAt":3124},"mcore-testing","run and manage Megatron-LM tests","Test system for Megatron-LM. Covers test layout, recipe YAML structure, adding and running unit and functional tests, golden values, marker filters, and CI parity.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3122,3123],{"name":3089,"slug":3090,"type":15},{"name":2935,"slug":2936,"type":15},"2026-07-14T05:25:54.928983",{"slug":3126,"name":3126,"fn":3127,"description":3128,"org":3129,"tags":3130,"stars":3015,"repoUrl":3016,"updatedAt":3133},"nightly-sync","manage nightly main-to-dev sync workflows","Domain knowledge for the nightly main-to-dev sync workflow. Covers merge strategy, CI architecture, failure investigation, and known issues.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},[3131,3132],{"name":17,"slug":18,"type":15},{"name":3025,"slug":3026,"type":15},"2026-07-30T05:29:03.275638",496]