[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-dino":3,"mdc--r9eznf-key":31,"related-repo-nvidia-tao-train-dino":1487,"related-org-nvidia-tao-train-dino":1591},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":20,"repoUrl":21,"updatedAt":22,"license":23,"forks":24,"topics":25,"repo":26,"sourceUrl":29,"mdContent":30},"tao-train-dino","train DINO object detection models","DINO (DETR with Improved DeNoising Anchor Boxes) for 2D object detection. Transformer-based detector with denoising training, multi-scale features, and optional distillation support. Use when training, evaluating, exporting, distilling, quantizing, or running inference for a TAO DINO detector. Trigger phrases include \"train DINO\", \"DETR object detection\", \"TAO 2D detection\", \"DINO with distillation\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,17],{"name":13,"slug":14,"type":15},"Deep Learning","deep-learning","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:29:58.100243","Apache-2.0",281,[],{"repoUrl":21,"stars":20,"forks":24,"topics":27,"description":28},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Ftao-train-dino","---\nname: tao-train-dino\ndescription: DINO (DETR with Improved DeNoising Anchor Boxes) for 2D object detection. Transformer-based detector with\n  denoising training, multi-scale features, and optional distillation support. Use when training, evaluating, exporting,\n  distilling, quantizing, or running inference for a TAO DINO detector. Trigger phrases include \"train DINO\", \"DETR object\n  detection\", \"TAO 2D detection\", \"DINO with distillation\".\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit.\nmetadata:\n  version: \"0.1.0\"\n  author: NVIDIA Corporation\nallowed-tools: Read Bash\ntags:\n- object\n- detection\n---\n\n# DINO\n\nDINO (DETR with Improved DeNoising Anchor Boxes) for 2D object detection. Transformer-based detector with denoising training, multi-scale features, and optional distillation support.\n\nUses pretrained backbone weights (e.g. ResNet-50 ImageNet). Set `model.pretrained_backbone_path` for backbone-only or `train.pretrained_model_path` for full model.\n\n## When To Use\n\nTrain, evaluate, export, distill, quantize, or run inference for a TAO DINO 2D object detector.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and\nTensorRT `inference`), read `references\u002Ftao-deploy-dino.md` first. Deploy spec templates live\nin this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml`\nprefix.\n\n## Reference Map\n\n- `references\u002Fdino-data-specs.md` — dataset contracts, per-action dataset requirements, per-action spec-override examples (train, evaluate, export, deploy\u002Fgen_trt_engine, inference, quantize, distill), data-source arrays, checkpoint inference, and dataset layout.\n- `references\u002Fdino-actions-errors.md` — important parameters, default values, evaluate\u002Fexport defaults, hardware, and the full error-pattern catalog.\n- `references\u002Fdino-tuning-multigpu.md` — full AutoML\u002FHPO notes (metrics, hyperparameters, extractor) and multi-GPU spec consistency.\n- `references\u002Fdino-automl-sdk.md` — AutoML metrics, SDK orchestration internals, data-source gap, and spec-param\u002Fparent-model inference.\n- `references\u002Ftao-deploy-dino.md` — TensorRT deploy workflow.\n- `references\u002Fdetailed-guide.md` — map to the detailed model guide.\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 such as `evaluate`, `inference`, `export`, and deploy flows stay in this model skill. The per-run `automl_policy` override does not change model metadata.\n\n## Training Requirements\n\nThe agent MUST read this section before generating any training or AutoML script for DINO.\n\n- **Dataset type:** object_detection\n- **Formats:** coco, coco_raw\n- **Accepted dataset intents:** training, evaluation, testing, calibration\n- **Monitoring metric:** mAP50 for quick operational checks; `val_mAP` for\n  COCO\u002Fpaper-style benchmark comparisons.\n\n**Required datasets — MUST resolve both:**\n\n| Dataset | Required | Why |\n|---|---|---|\n| Train dataset URI | Yes | Training data (COCO format) |\n| Validation dataset URI | **Yes — ALWAYS** | DINO unconditionally builds a val dataloader. Omitting `val_data_sources` causes `FileNotFoundError` at startup regardless of the metric or workflow. If the user has no separate eval split, reuse the train URI. |\n\n**Required inputs before generating any training spec:**\n\n1. **Train dataset URI** — S3 path to COCO-format training data\n2. **Validation dataset URI** — S3 path to COCO-format val data (can be same as train)\n3. **`num_classes`** — How many object classes? Default 91 (COCO). Must be >= `max(category_id) + 1`. Too low causes `CUDA error: device-side assert triggered`.\n\nResolve these from the user request or the default profile below. Prompt only\nfor values that are still missing after applying the profile rules.\n\n**Bankable local default profile for DINO AutoML smoke runs:**\n\nUse this profile only when the user asks to run DINO AutoML and does not provide\ndataset or class-count inputs. This profile is intentionally small and local to\nthis skill bank; it is for smoke\u002Fiteration runs, not a production benchmark.\nDo not search previous runners, logs, session state, shell history, or the home\ndirectory to recover these values.\n\n```python\nDINO_AUTOML_PROFILE = {\n    \"train_dataset_uri\": \"s3:\u002F\u002Fnvcf-storage-handling\u002Fdata\u002Ftao_od_synthetic_subset_train_no_convert\",\n    \"validation_dataset_uri\": \"s3:\u002F\u002Fnvcf-storage-handling\u002Fdata\u002Ftao_od_synthetic_subset_val_no_convert\",\n    \"object_classes\": 4,\n    \"dataset_num_classes\": 5,\n    \"image_archive\": \"images.tar.gz\",\n    \"annotation_file\": \"annotations.json\",\n    \"max_recommendations\": 10,\n    \"train_num_epochs\": 10,\n    \"train_checkpoint_interval\": 10,\n    \"train_validation_interval\": 1,\n    \"train_num_gpus\": 1,\n}\n```\n\nIf the user supplies any dataset URI or class-count value, prefer the user value\nand ask for any remaining required DINO value. Do not partially mix a user's\ncustom dataset with this profile's class count unless the user confirms it.\n\n**Do not prompt for image layout for the standard DINO dataset.** The standard\nTAO DINO dataset artifact is `images.tar.gz` plus `annotations.json`. Use\n`images.tar.gz` in the remote `image_dir` spec override. The SDK downloads the\narchive and rewrites the runtime spec to the extracted folder named after the\narchive stem (`images.tar.gz` -> `images`). Only deviate if the user explicitly\nprovides a different image artifact name.\n\n## Core Workflow\n\nDINO supports train, evaluate, export, distill, quantize, and inference. Data-source\noverrides are **mandatory for every action** — DINO's `config.json` has empty\n`data_sources` because the runner cannot auto-resolve array-of-objects spec keys.\nThe agent MUST construct data source paths and include them in `spec_overrides`.\n\nSee `references\u002Fdino-data-specs.md` for the per-action dataset requirements table,\nthe standard dataset artifact (`images.tar.gz` + `annotations.json`) and runtime\nfolder rewrite rules, and the complete per-action `spec_overrides` examples for\ntrain, evaluate, export, deploy\u002Fgen_trt_engine, inference, quantize, and distill —\nincluding checkpoint inference via `parent_model`, the `results_dir\u002Ftrain\u002F`\ncheckpoint location, and the distillation FAN-teacher \u002F student rules.\n\n## Important Parameters And Defaults\n\nKey defaults: `num_epochs=10`, `batch_size=4`, `learning_rate=2e-4`,\n`lr_backbone=2e-5`, `num_classes=91`, `backbone=resnet_50`.\n\n- **dataset.num_classes**: Default 91 (COCO). Must be >= `max(category_id) + 1`. Too low causes `CUDA error: device-side assert triggered`. Set as `\u003Cnum_classes> + 1` in spec overrides.\n- **num_epochs**: default 10 (quick iteration); real datasets typically need 30-50+ epochs for good mAP.\n\nSee `references\u002Fdino-actions-errors.md` for the full parameter list (backbone\noptions, `train.optim.lr`\u002F`lr_steps`, `model.num_queries`, `batch_size`),\ndefault values, evaluate defaults, export defaults (input 960x544, opset 17,\nTRT data types, workspace 1024 MB), and hardware requirements.\n\n## Multi-GPU And AutoML \u002F HPO\n\nWhen increasing `train.num_gpus`, also set `train.gpu_ids` to the same visible\ndevice range, or distributed startup can be inconsistent.\n\nAutoML runs training — all **Training Requirements** above apply. For no-input\nlocal smoke runs, use `DINO_AUTOML_PROFILE`. Recommended metric is `mAP50`\n(`val_mAP` for benchmark comparisons) with `direction=\"maximize\"` and a custom\n`metric_extractor`.\n\nSee `references\u002Fdino-tuning-multigpu.md` for the full multi-GPU spec-consistency\nrule (8-GPU example, NCCL timeout note) and the full AutoML\u002FHPO notes (metric\nselection, `metric_extractor`, recommended hyperparameters, `weight_decay`\nbehavior, dense-dataset resume guidance). See `references\u002Fdino-automl-sdk.md` for\nAutoML metric extractor code, SDK orchestration internals, and parent-model\ninference mappings.\n\n## Error Patterns\n\nCommon failures include CUDA OOM (reduce `batch_size`), missing `val_data_sources`\n(`FileNotFoundError` at startup — always supply val), `num_classes` too low (`CUDA\ndevice-side assert`), and the parent `dino gen_trt_engine` \u002F `dino convert` PyT-CLI\nrestrictions.\n\nSee `references\u002Fdino-actions-errors.md` for the complete error-pattern catalog\nwith diagnostics and fixes.\n\n## Spec Param \u002F Parent Model Inference\n\nModel-specific inference mappings belong in this MD file, not in `config.json`.\nGenerated runners read the mappings and apply them with SDK helpers before\n`create_job()`. For `parent_model`\u002F`parent_model_folder`, pass the upstream\ntrain\u002Fexport\u002FAutoML child job id as `parent_job_id`; the SDK lists the parent\nresult folder, filters checkpoint artifacts, and returns the selected model.\n\nSee `references\u002Fdino-automl-sdk.md` for the full inference-mapping table (per\naction: `parent_model`, `key`, `output_dir`, `ptm_if_no_resume_model`,\n`resume_model`, `create_onnx_file`) and the TensorRT-mapping note. TensorRT\nmappings live in the deploy workflow, not the PyT model skill.\n\n## Optional: running via the TAO SDK\n\nWhen running DINO through the TAO SDK (`script_runner` orchestration, S3 I\u002FO\nwrapping, AutoML), skills read `references\u002Fskill_info.yaml` for input and\nspec-param mappings. See `references\u002Fdino-automl-sdk.md` for SDK orchestration\ninternals, including the data-sources gap and the `[0]`-indexed `inputs`\ndeclarations. Skip this when running locally with `docker run`.\n\n## Deployment\n\n- [tao-deploy-dino](references\u002Ftao-deploy-dino.md)\n",{"data":32,"body":41},{"name":4,"description":6,"license":23,"compatibility":33,"metadata":34,"allowed-tools":37,"tags":38},"Requires docker + nvidia-container-toolkit.",{"version":35,"author":36},"0.1.0","NVIDIA Corporation","Read Bash",[39,40],"object","detection",{"type":42,"children":43},"root",[44,53,59,81,88,93,146,152,222,228,321,327,436,468,474,479,531,539,627,635,687,692,700,705,835,840,896,902,937,986,992,1040,1084,1125,1131,1152,1201,1234,1240,1295,1306,1312,1354,1407,1413,1463,1469,1481],{"type":45,"tag":46,"props":47,"children":49},"element","h1",{"id":48},"dino",[50],{"type":51,"value":52},"text","DINO",{"type":45,"tag":54,"props":55,"children":56},"p",{},[57],{"type":51,"value":58},"DINO (DETR with Improved DeNoising Anchor Boxes) for 2D object detection. Transformer-based detector with denoising training, multi-scale features, and optional distillation support.",{"type":45,"tag":54,"props":60,"children":61},{},[62,64,71,73,79],{"type":51,"value":63},"Uses pretrained backbone weights (e.g. ResNet-50 ImageNet). Set ",{"type":45,"tag":65,"props":66,"children":68},"code",{"className":67},[],[69],{"type":51,"value":70},"model.pretrained_backbone_path",{"type":51,"value":72}," for backbone-only or ",{"type":45,"tag":65,"props":74,"children":76},{"className":75},[],[77],{"type":51,"value":78},"train.pretrained_model_path",{"type":51,"value":80}," for full model.",{"type":45,"tag":82,"props":83,"children":85},"h2",{"id":84},"when-to-use",[86],{"type":51,"value":87},"When To Use",{"type":45,"tag":54,"props":89,"children":90},{},[91],{"type":51,"value":92},"Train, evaluate, export, distill, quantize, or run inference for a TAO DINO 2D object detector.",{"type":45,"tag":54,"props":94,"children":95},{},[96,98,104,106,112,114,120,122,128,130,136,138,144],{"type":51,"value":97},"For TAO Deploy TensorRT actions (",{"type":45,"tag":65,"props":99,"children":101},{"className":100},[],[102],{"type":51,"value":103},"gen_trt_engine",{"type":51,"value":105},", TensorRT ",{"type":45,"tag":65,"props":107,"children":109},{"className":108},[],[110],{"type":51,"value":111},"evaluate",{"type":51,"value":113},", and\nTensorRT ",{"type":45,"tag":65,"props":115,"children":117},{"className":116},[],[118],{"type":51,"value":119},"inference",{"type":51,"value":121},"), read ",{"type":45,"tag":65,"props":123,"children":125},{"className":124},[],[126],{"type":51,"value":127},"references\u002Ftao-deploy-dino.md",{"type":51,"value":129}," first. Deploy spec templates live\nin this skill's ",{"type":45,"tag":65,"props":131,"children":133},{"className":132},[],[134],{"type":51,"value":135},"references\u002F",{"type":51,"value":137}," folder with the ",{"type":45,"tag":65,"props":139,"children":141},{"className":140},[],[142],{"type":51,"value":143},"spec_template_deploy_*.yaml",{"type":51,"value":145},"\nprefix.",{"type":45,"tag":82,"props":147,"children":149},{"id":148},"reference-map",[150],{"type":51,"value":151},"Reference Map",{"type":45,"tag":153,"props":154,"children":155},"ul",{},[156,168,179,190,201,211],{"type":45,"tag":157,"props":158,"children":159},"li",{},[160,166],{"type":45,"tag":65,"props":161,"children":163},{"className":162},[],[164],{"type":51,"value":165},"references\u002Fdino-data-specs.md",{"type":51,"value":167}," — dataset contracts, per-action dataset requirements, per-action spec-override examples (train, evaluate, export, deploy\u002Fgen_trt_engine, inference, quantize, distill), data-source arrays, checkpoint inference, and dataset layout.",{"type":45,"tag":157,"props":169,"children":170},{},[171,177],{"type":45,"tag":65,"props":172,"children":174},{"className":173},[],[175],{"type":51,"value":176},"references\u002Fdino-actions-errors.md",{"type":51,"value":178}," — important parameters, default values, evaluate\u002Fexport defaults, hardware, and the full error-pattern catalog.",{"type":45,"tag":157,"props":180,"children":181},{},[182,188],{"type":45,"tag":65,"props":183,"children":185},{"className":184},[],[186],{"type":51,"value":187},"references\u002Fdino-tuning-multigpu.md",{"type":51,"value":189}," — full AutoML\u002FHPO notes (metrics, hyperparameters, extractor) and multi-GPU spec consistency.",{"type":45,"tag":157,"props":191,"children":192},{},[193,199],{"type":45,"tag":65,"props":194,"children":196},{"className":195},[],[197],{"type":51,"value":198},"references\u002Fdino-automl-sdk.md",{"type":51,"value":200}," — AutoML metrics, SDK orchestration internals, data-source gap, and spec-param\u002Fparent-model inference.",{"type":45,"tag":157,"props":202,"children":203},{},[204,209],{"type":45,"tag":65,"props":205,"children":207},{"className":206},[],[208],{"type":51,"value":127},{"type":51,"value":210}," — TensorRT deploy workflow.",{"type":45,"tag":157,"props":212,"children":213},{},[214,220],{"type":45,"tag":65,"props":215,"children":217},{"className":216},[],[218],{"type":51,"value":219},"references\u002Fdetailed-guide.md",{"type":51,"value":221}," — map to the detailed model guide.",{"type":45,"tag":82,"props":223,"children":225},{"id":224},"dataclass-schemas",[226],{"type":51,"value":227},"Dataclass Schemas",{"type":45,"tag":54,"props":229,"children":230},{},[231,233,239,241,247,249,255,257,263,265,271,273,279,281,287,289,295,297,303,305,311,313,319],{"type":51,"value":232},"Generated TAO Core schemas are packaged in ",{"type":45,"tag":65,"props":234,"children":236},{"className":235},[],[237],{"type":51,"value":238},"schemas\u002F\u003Caction>.schema.json",{"type":51,"value":240},", with ",{"type":45,"tag":65,"props":242,"children":244},{"className":243},[],[245],{"type":51,"value":246},"schemas\u002Fmanifest.json",{"type":51,"value":248}," listing available actions. Each generated schema also emits ",{"type":45,"tag":65,"props":250,"children":252},{"className":251},[],[253],{"type":51,"value":254},"references\u002Fspec_template_\u003Caction>.yaml",{"type":51,"value":256}," from the schema top-level ",{"type":45,"tag":65,"props":258,"children":260},{"className":259},[],[261],{"type":51,"value":262},"default",{"type":51,"value":264}," field. AutoML enablement is declared at the model layer in ",{"type":45,"tag":65,"props":266,"children":268},{"className":267},[],[269],{"type":51,"value":270},"references\u002Fskill_info.yaml",{"type":51,"value":272}," via ",{"type":45,"tag":65,"props":274,"children":276},{"className":275},[],[277],{"type":51,"value":278},"automl_enabled",{"type":51,"value":280},". Runnable AutoML still requires ",{"type":45,"tag":65,"props":282,"children":284},{"className":283},[],[285],{"type":51,"value":286},"schemas\u002Ftrain.schema.json",{"type":51,"value":288}," and ",{"type":45,"tag":65,"props":290,"children":292},{"className":291},[],[293],{"type":51,"value":294},"references\u002Fspec_template_train.yaml",{"type":51,"value":296}," to exist and parse. Use the packaged train schema for ",{"type":45,"tag":65,"props":298,"children":300},{"className":299},[],[301],{"type":51,"value":302},"automl_default_parameters",{"type":51,"value":304},", ",{"type":45,"tag":65,"props":306,"children":308},{"className":307},[],[309],{"type":51,"value":310},"automl_disabled_parameters",{"type":51,"value":312},", defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ",{"type":45,"tag":65,"props":314,"children":316},{"className":315},[],[317],{"type":51,"value":318},"~\u002Ftao-core",{"type":51,"value":320}," at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.",{"type":45,"tag":82,"props":322,"children":324},{"id":323},"train-action-policy",[325],{"type":51,"value":326},"Train Action Policy",{"type":45,"tag":54,"props":328,"children":329},{},[330,332,337,339,345,347,353,355,361,363,369,371,377,379,384,385,391,393,398,399,404,406,412,414,420,422,427,429,434],{"type":51,"value":331},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":45,"tag":65,"props":333,"children":335},{"className":334},[],[336],{"type":51,"value":270},{"type":51,"value":338}," and resolve the run override from either an explicit ",{"type":45,"tag":65,"props":340,"children":342},{"className":341},[],[343],{"type":51,"value":344},"automl_policy",{"type":51,"value":346}," value or the user's workflow request. Use ",{"type":45,"tag":65,"props":348,"children":350},{"className":349},[],[351],{"type":51,"value":352},"automl_policy: on",{"type":51,"value":354}," by default and only expose ",{"type":45,"tag":65,"props":356,"children":358},{"className":357},[],[359],{"type":51,"value":360},"on",{"type":51,"value":362}," \u002F ",{"type":45,"tag":65,"props":364,"children":366},{"className":365},[],[367],{"type":51,"value":368},"off",{"type":51,"value":370}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":45,"tag":65,"props":372,"children":374},{"className":373},[],[375],{"type":51,"value":376},"automl_policy: off",{"type":51,"value":378}," for this run only. When ",{"type":45,"tag":65,"props":380,"children":382},{"className":381},[],[383],{"type":51,"value":352},{"type":51,"value":304},{"type":45,"tag":65,"props":386,"children":388},{"className":387},[],[389],{"type":51,"value":390},"automl_enabled: true",{"type":51,"value":392},", and both ",{"type":45,"tag":65,"props":394,"children":396},{"className":395},[],[397],{"type":51,"value":286},{"type":51,"value":288},{"type":45,"tag":65,"props":400,"children":402},{"className":401},[],[403],{"type":51,"value":294},{"type":51,"value":405}," are packaged, route the train action through ",{"type":45,"tag":65,"props":407,"children":409},{"className":408},[],[410],{"type":51,"value":411},"tao-skill-bank:tao-run-automl",{"type":51,"value":413}," by default with this model's ",{"type":45,"tag":65,"props":415,"children":417},{"className":416},[],[418],{"type":51,"value":419},"skill_dir",{"type":51,"value":421},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":45,"tag":65,"props":423,"children":425},{"className":424},[],[426],{"type":51,"value":344},{"type":51,"value":428},". Use direct model training only when ",{"type":45,"tag":65,"props":430,"children":432},{"className":431},[],[433],{"type":51,"value":376},{"type":51,"value":435}," 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":45,"tag":54,"props":437,"children":438},{},[439,441,446,447,452,453,459,461,466],{"type":51,"value":440},"Non-train actions such as ",{"type":45,"tag":65,"props":442,"children":444},{"className":443},[],[445],{"type":51,"value":111},{"type":51,"value":304},{"type":45,"tag":65,"props":448,"children":450},{"className":449},[],[451],{"type":51,"value":119},{"type":51,"value":304},{"type":45,"tag":65,"props":454,"children":456},{"className":455},[],[457],{"type":51,"value":458},"export",{"type":51,"value":460},", and deploy flows stay in this model skill. The per-run ",{"type":45,"tag":65,"props":462,"children":464},{"className":463},[],[465],{"type":51,"value":344},{"type":51,"value":467}," override does not change model metadata.",{"type":45,"tag":82,"props":469,"children":471},{"id":470},"training-requirements",[472],{"type":51,"value":473},"Training Requirements",{"type":45,"tag":54,"props":475,"children":476},{},[477],{"type":51,"value":478},"The agent MUST read this section before generating any training or AutoML script for DINO.",{"type":45,"tag":153,"props":480,"children":481},{},[482,493,503,513],{"type":45,"tag":157,"props":483,"children":484},{},[485,491],{"type":45,"tag":486,"props":487,"children":488},"strong",{},[489],{"type":51,"value":490},"Dataset type:",{"type":51,"value":492}," object_detection",{"type":45,"tag":157,"props":494,"children":495},{},[496,501],{"type":45,"tag":486,"props":497,"children":498},{},[499],{"type":51,"value":500},"Formats:",{"type":51,"value":502}," coco, coco_raw",{"type":45,"tag":157,"props":504,"children":505},{},[506,511],{"type":45,"tag":486,"props":507,"children":508},{},[509],{"type":51,"value":510},"Accepted dataset intents:",{"type":51,"value":512}," training, evaluation, testing, calibration",{"type":45,"tag":157,"props":514,"children":515},{},[516,521,523,529],{"type":45,"tag":486,"props":517,"children":518},{},[519],{"type":51,"value":520},"Monitoring metric:",{"type":51,"value":522}," mAP50 for quick operational checks; ",{"type":45,"tag":65,"props":524,"children":526},{"className":525},[],[527],{"type":51,"value":528},"val_mAP",{"type":51,"value":530}," for\nCOCO\u002Fpaper-style benchmark comparisons.",{"type":45,"tag":54,"props":532,"children":533},{},[534],{"type":45,"tag":486,"props":535,"children":536},{},[537],{"type":51,"value":538},"Required datasets — MUST resolve both:",{"type":45,"tag":540,"props":541,"children":542},"table",{},[543,567],{"type":45,"tag":544,"props":545,"children":546},"thead",{},[547],{"type":45,"tag":548,"props":549,"children":550},"tr",{},[551,557,562],{"type":45,"tag":552,"props":553,"children":554},"th",{},[555],{"type":51,"value":556},"Dataset",{"type":45,"tag":552,"props":558,"children":559},{},[560],{"type":51,"value":561},"Required",{"type":45,"tag":552,"props":563,"children":564},{},[565],{"type":51,"value":566},"Why",{"type":45,"tag":568,"props":569,"children":570},"tbody",{},[571,590],{"type":45,"tag":548,"props":572,"children":573},{},[574,580,585],{"type":45,"tag":575,"props":576,"children":577},"td",{},[578],{"type":51,"value":579},"Train dataset URI",{"type":45,"tag":575,"props":581,"children":582},{},[583],{"type":51,"value":584},"Yes",{"type":45,"tag":575,"props":586,"children":587},{},[588],{"type":51,"value":589},"Training data (COCO format)",{"type":45,"tag":548,"props":591,"children":592},{},[593,598,606],{"type":45,"tag":575,"props":594,"children":595},{},[596],{"type":51,"value":597},"Validation dataset URI",{"type":45,"tag":575,"props":599,"children":600},{},[601],{"type":45,"tag":486,"props":602,"children":603},{},[604],{"type":51,"value":605},"Yes — ALWAYS",{"type":45,"tag":575,"props":607,"children":608},{},[609,611,617,619,625],{"type":51,"value":610},"DINO unconditionally builds a val dataloader. Omitting ",{"type":45,"tag":65,"props":612,"children":614},{"className":613},[],[615],{"type":51,"value":616},"val_data_sources",{"type":51,"value":618}," causes ",{"type":45,"tag":65,"props":620,"children":622},{"className":621},[],[623],{"type":51,"value":624},"FileNotFoundError",{"type":51,"value":626}," at startup regardless of the metric or workflow. If the user has no separate eval split, reuse the train URI.",{"type":45,"tag":54,"props":628,"children":629},{},[630],{"type":45,"tag":486,"props":631,"children":632},{},[633],{"type":51,"value":634},"Required inputs before generating any training spec:",{"type":45,"tag":636,"props":637,"children":638},"ol",{},[639,648,657],{"type":45,"tag":157,"props":640,"children":641},{},[642,646],{"type":45,"tag":486,"props":643,"children":644},{},[645],{"type":51,"value":579},{"type":51,"value":647}," — S3 path to COCO-format training data",{"type":45,"tag":157,"props":649,"children":650},{},[651,655],{"type":45,"tag":486,"props":652,"children":653},{},[654],{"type":51,"value":597},{"type":51,"value":656}," — S3 path to COCO-format val data (can be same as train)",{"type":45,"tag":157,"props":658,"children":659},{},[660,669,671,677,679,685],{"type":45,"tag":486,"props":661,"children":662},{},[663],{"type":45,"tag":65,"props":664,"children":666},{"className":665},[],[667],{"type":51,"value":668},"num_classes",{"type":51,"value":670}," — How many object classes? Default 91 (COCO). Must be >= ",{"type":45,"tag":65,"props":672,"children":674},{"className":673},[],[675],{"type":51,"value":676},"max(category_id) + 1",{"type":51,"value":678},". Too low causes ",{"type":45,"tag":65,"props":680,"children":682},{"className":681},[],[683],{"type":51,"value":684},"CUDA error: device-side assert triggered",{"type":51,"value":686},".",{"type":45,"tag":54,"props":688,"children":689},{},[690],{"type":51,"value":691},"Resolve these from the user request or the default profile below. Prompt only\nfor values that are still missing after applying the profile rules.",{"type":45,"tag":54,"props":693,"children":694},{},[695],{"type":45,"tag":486,"props":696,"children":697},{},[698],{"type":51,"value":699},"Bankable local default profile for DINO AutoML smoke runs:",{"type":45,"tag":54,"props":701,"children":702},{},[703],{"type":51,"value":704},"Use this profile only when the user asks to run DINO AutoML and does not provide\ndataset or class-count inputs. This profile is intentionally small and local to\nthis skill bank; it is for smoke\u002Fiteration runs, not a production benchmark.\nDo not search previous runners, logs, session state, shell history, or the home\ndirectory to recover these values.",{"type":45,"tag":706,"props":707,"children":712},"pre",{"className":708,"code":709,"language":710,"meta":711,"style":711},"language-python shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","DINO_AUTOML_PROFILE = {\n    \"train_dataset_uri\": \"s3:\u002F\u002Fnvcf-storage-handling\u002Fdata\u002Ftao_od_synthetic_subset_train_no_convert\",\n    \"validation_dataset_uri\": \"s3:\u002F\u002Fnvcf-storage-handling\u002Fdata\u002Ftao_od_synthetic_subset_val_no_convert\",\n    \"object_classes\": 4,\n    \"dataset_num_classes\": 5,\n    \"image_archive\": \"images.tar.gz\",\n    \"annotation_file\": \"annotations.json\",\n    \"max_recommendations\": 10,\n    \"train_num_epochs\": 10,\n    \"train_checkpoint_interval\": 10,\n    \"train_validation_interval\": 1,\n    \"train_num_gpus\": 1,\n}\n","python","",[713],{"type":45,"tag":65,"props":714,"children":715},{"__ignoreMap":711},[716,727,736,745,754,763,772,781,790,799,808,817,826],{"type":45,"tag":717,"props":718,"children":721},"span",{"class":719,"line":720},"line",1,[722],{"type":45,"tag":717,"props":723,"children":724},{},[725],{"type":51,"value":726},"DINO_AUTOML_PROFILE = {\n",{"type":45,"tag":717,"props":728,"children":730},{"class":719,"line":729},2,[731],{"type":45,"tag":717,"props":732,"children":733},{},[734],{"type":51,"value":735},"    \"train_dataset_uri\": \"s3:\u002F\u002Fnvcf-storage-handling\u002Fdata\u002Ftao_od_synthetic_subset_train_no_convert\",\n",{"type":45,"tag":717,"props":737,"children":739},{"class":719,"line":738},3,[740],{"type":45,"tag":717,"props":741,"children":742},{},[743],{"type":51,"value":744},"    \"validation_dataset_uri\": \"s3:\u002F\u002Fnvcf-storage-handling\u002Fdata\u002Ftao_od_synthetic_subset_val_no_convert\",\n",{"type":45,"tag":717,"props":746,"children":748},{"class":719,"line":747},4,[749],{"type":45,"tag":717,"props":750,"children":751},{},[752],{"type":51,"value":753},"    \"object_classes\": 4,\n",{"type":45,"tag":717,"props":755,"children":757},{"class":719,"line":756},5,[758],{"type":45,"tag":717,"props":759,"children":760},{},[761],{"type":51,"value":762},"    \"dataset_num_classes\": 5,\n",{"type":45,"tag":717,"props":764,"children":766},{"class":719,"line":765},6,[767],{"type":45,"tag":717,"props":768,"children":769},{},[770],{"type":51,"value":771},"    \"image_archive\": \"images.tar.gz\",\n",{"type":45,"tag":717,"props":773,"children":775},{"class":719,"line":774},7,[776],{"type":45,"tag":717,"props":777,"children":778},{},[779],{"type":51,"value":780},"    \"annotation_file\": \"annotations.json\",\n",{"type":45,"tag":717,"props":782,"children":784},{"class":719,"line":783},8,[785],{"type":45,"tag":717,"props":786,"children":787},{},[788],{"type":51,"value":789},"    \"max_recommendations\": 10,\n",{"type":45,"tag":717,"props":791,"children":793},{"class":719,"line":792},9,[794],{"type":45,"tag":717,"props":795,"children":796},{},[797],{"type":51,"value":798},"    \"train_num_epochs\": 10,\n",{"type":45,"tag":717,"props":800,"children":802},{"class":719,"line":801},10,[803],{"type":45,"tag":717,"props":804,"children":805},{},[806],{"type":51,"value":807},"    \"train_checkpoint_interval\": 10,\n",{"type":45,"tag":717,"props":809,"children":811},{"class":719,"line":810},11,[812],{"type":45,"tag":717,"props":813,"children":814},{},[815],{"type":51,"value":816},"    \"train_validation_interval\": 1,\n",{"type":45,"tag":717,"props":818,"children":820},{"class":719,"line":819},12,[821],{"type":45,"tag":717,"props":822,"children":823},{},[824],{"type":51,"value":825},"    \"train_num_gpus\": 1,\n",{"type":45,"tag":717,"props":827,"children":829},{"class":719,"line":828},13,[830],{"type":45,"tag":717,"props":831,"children":832},{},[833],{"type":51,"value":834},"}\n",{"type":45,"tag":54,"props":836,"children":837},{},[838],{"type":51,"value":839},"If the user supplies any dataset URI or class-count value, prefer the user value\nand ask for any remaining required DINO value. Do not partially mix a user's\ncustom dataset with this profile's class count unless the user confirms it.",{"type":45,"tag":54,"props":841,"children":842},{},[843,848,850,856,858,864,866,871,873,879,881,886,888,894],{"type":45,"tag":486,"props":844,"children":845},{},[846],{"type":51,"value":847},"Do not prompt for image layout for the standard DINO dataset.",{"type":51,"value":849}," The standard\nTAO DINO dataset artifact is ",{"type":45,"tag":65,"props":851,"children":853},{"className":852},[],[854],{"type":51,"value":855},"images.tar.gz",{"type":51,"value":857}," plus ",{"type":45,"tag":65,"props":859,"children":861},{"className":860},[],[862],{"type":51,"value":863},"annotations.json",{"type":51,"value":865},". Use\n",{"type":45,"tag":65,"props":867,"children":869},{"className":868},[],[870],{"type":51,"value":855},{"type":51,"value":872}," in the remote ",{"type":45,"tag":65,"props":874,"children":876},{"className":875},[],[877],{"type":51,"value":878},"image_dir",{"type":51,"value":880}," spec override. The SDK downloads the\narchive and rewrites the runtime spec to the extracted folder named after the\narchive stem (",{"type":45,"tag":65,"props":882,"children":884},{"className":883},[],[885],{"type":51,"value":855},{"type":51,"value":887}," -> ",{"type":45,"tag":65,"props":889,"children":891},{"className":890},[],[892],{"type":51,"value":893},"images",{"type":51,"value":895},"). Only deviate if the user explicitly\nprovides a different image artifact name.",{"type":45,"tag":82,"props":897,"children":899},{"id":898},"core-workflow",[900],{"type":51,"value":901},"Core Workflow",{"type":45,"tag":54,"props":903,"children":904},{},[905,907,912,914,920,922,928,930,936],{"type":51,"value":906},"DINO supports train, evaluate, export, distill, quantize, and inference. Data-source\noverrides are ",{"type":45,"tag":486,"props":908,"children":909},{},[910],{"type":51,"value":911},"mandatory for every action",{"type":51,"value":913}," — DINO's ",{"type":45,"tag":65,"props":915,"children":917},{"className":916},[],[918],{"type":51,"value":919},"config.json",{"type":51,"value":921}," has empty\n",{"type":45,"tag":65,"props":923,"children":925},{"className":924},[],[926],{"type":51,"value":927},"data_sources",{"type":51,"value":929}," because the runner cannot auto-resolve array-of-objects spec keys.\nThe agent MUST construct data source paths and include them in ",{"type":45,"tag":65,"props":931,"children":933},{"className":932},[],[934],{"type":51,"value":935},"spec_overrides",{"type":51,"value":686},{"type":45,"tag":54,"props":938,"children":939},{},[940,942,947,949,954,956,961,963,968,970,976,978,984],{"type":51,"value":941},"See ",{"type":45,"tag":65,"props":943,"children":945},{"className":944},[],[946],{"type":51,"value":165},{"type":51,"value":948}," for the per-action dataset requirements table,\nthe standard dataset artifact (",{"type":45,"tag":65,"props":950,"children":952},{"className":951},[],[953],{"type":51,"value":855},{"type":51,"value":955}," + ",{"type":45,"tag":65,"props":957,"children":959},{"className":958},[],[960],{"type":51,"value":863},{"type":51,"value":962},") and runtime\nfolder rewrite rules, and the complete per-action ",{"type":45,"tag":65,"props":964,"children":966},{"className":965},[],[967],{"type":51,"value":935},{"type":51,"value":969}," examples for\ntrain, evaluate, export, deploy\u002Fgen_trt_engine, inference, quantize, and distill —\nincluding checkpoint inference via ",{"type":45,"tag":65,"props":971,"children":973},{"className":972},[],[974],{"type":51,"value":975},"parent_model",{"type":51,"value":977},", the ",{"type":45,"tag":65,"props":979,"children":981},{"className":980},[],[982],{"type":51,"value":983},"results_dir\u002Ftrain\u002F",{"type":51,"value":985},"\ncheckpoint location, and the distillation FAN-teacher \u002F student rules.",{"type":45,"tag":82,"props":987,"children":989},{"id":988},"important-parameters-and-defaults",[990],{"type":51,"value":991},"Important Parameters And Defaults",{"type":45,"tag":54,"props":993,"children":994},{},[995,997,1003,1004,1010,1011,1017,1019,1025,1026,1032,1033,1039],{"type":51,"value":996},"Key defaults: ",{"type":45,"tag":65,"props":998,"children":1000},{"className":999},[],[1001],{"type":51,"value":1002},"num_epochs=10",{"type":51,"value":304},{"type":45,"tag":65,"props":1005,"children":1007},{"className":1006},[],[1008],{"type":51,"value":1009},"batch_size=4",{"type":51,"value":304},{"type":45,"tag":65,"props":1012,"children":1014},{"className":1013},[],[1015],{"type":51,"value":1016},"learning_rate=2e-4",{"type":51,"value":1018},",\n",{"type":45,"tag":65,"props":1020,"children":1022},{"className":1021},[],[1023],{"type":51,"value":1024},"lr_backbone=2e-5",{"type":51,"value":304},{"type":45,"tag":65,"props":1027,"children":1029},{"className":1028},[],[1030],{"type":51,"value":1031},"num_classes=91",{"type":51,"value":304},{"type":45,"tag":65,"props":1034,"children":1036},{"className":1035},[],[1037],{"type":51,"value":1038},"backbone=resnet_50",{"type":51,"value":686},{"type":45,"tag":153,"props":1041,"children":1042},{},[1043,1074],{"type":45,"tag":157,"props":1044,"children":1045},{},[1046,1051,1053,1058,1059,1064,1066,1072],{"type":45,"tag":486,"props":1047,"children":1048},{},[1049],{"type":51,"value":1050},"dataset.num_classes",{"type":51,"value":1052},": Default 91 (COCO). Must be >= ",{"type":45,"tag":65,"props":1054,"children":1056},{"className":1055},[],[1057],{"type":51,"value":676},{"type":51,"value":678},{"type":45,"tag":65,"props":1060,"children":1062},{"className":1061},[],[1063],{"type":51,"value":684},{"type":51,"value":1065},". Set as ",{"type":45,"tag":65,"props":1067,"children":1069},{"className":1068},[],[1070],{"type":51,"value":1071},"\u003Cnum_classes> + 1",{"type":51,"value":1073}," in spec overrides.",{"type":45,"tag":157,"props":1075,"children":1076},{},[1077,1082],{"type":45,"tag":486,"props":1078,"children":1079},{},[1080],{"type":51,"value":1081},"num_epochs",{"type":51,"value":1083},": default 10 (quick iteration); real datasets typically need 30-50+ epochs for good mAP.",{"type":45,"tag":54,"props":1085,"children":1086},{},[1087,1088,1093,1095,1101,1103,1109,1110,1116,1117,1123],{"type":51,"value":941},{"type":45,"tag":65,"props":1089,"children":1091},{"className":1090},[],[1092],{"type":51,"value":176},{"type":51,"value":1094}," for the full parameter list (backbone\noptions, ",{"type":45,"tag":65,"props":1096,"children":1098},{"className":1097},[],[1099],{"type":51,"value":1100},"train.optim.lr",{"type":51,"value":1102},"\u002F",{"type":45,"tag":65,"props":1104,"children":1106},{"className":1105},[],[1107],{"type":51,"value":1108},"lr_steps",{"type":51,"value":304},{"type":45,"tag":65,"props":1111,"children":1113},{"className":1112},[],[1114],{"type":51,"value":1115},"model.num_queries",{"type":51,"value":304},{"type":45,"tag":65,"props":1118,"children":1120},{"className":1119},[],[1121],{"type":51,"value":1122},"batch_size",{"type":51,"value":1124},"),\ndefault values, evaluate defaults, export defaults (input 960x544, opset 17,\nTRT data types, workspace 1024 MB), and hardware requirements.",{"type":45,"tag":82,"props":1126,"children":1128},{"id":1127},"multi-gpu-and-automl-hpo",[1129],{"type":51,"value":1130},"Multi-GPU And AutoML \u002F HPO",{"type":45,"tag":54,"props":1132,"children":1133},{},[1134,1136,1142,1144,1150],{"type":51,"value":1135},"When increasing ",{"type":45,"tag":65,"props":1137,"children":1139},{"className":1138},[],[1140],{"type":51,"value":1141},"train.num_gpus",{"type":51,"value":1143},", also set ",{"type":45,"tag":65,"props":1145,"children":1147},{"className":1146},[],[1148],{"type":51,"value":1149},"train.gpu_ids",{"type":51,"value":1151}," to the same visible\ndevice range, or distributed startup can be inconsistent.",{"type":45,"tag":54,"props":1153,"children":1154},{},[1155,1157,1161,1163,1169,1171,1177,1179,1184,1186,1192,1194,1200],{"type":51,"value":1156},"AutoML runs training — all ",{"type":45,"tag":486,"props":1158,"children":1159},{},[1160],{"type":51,"value":473},{"type":51,"value":1162}," above apply. For no-input\nlocal smoke runs, use ",{"type":45,"tag":65,"props":1164,"children":1166},{"className":1165},[],[1167],{"type":51,"value":1168},"DINO_AUTOML_PROFILE",{"type":51,"value":1170},". Recommended metric is ",{"type":45,"tag":65,"props":1172,"children":1174},{"className":1173},[],[1175],{"type":51,"value":1176},"mAP50",{"type":51,"value":1178},"\n(",{"type":45,"tag":65,"props":1180,"children":1182},{"className":1181},[],[1183],{"type":51,"value":528},{"type":51,"value":1185}," for benchmark comparisons) with ",{"type":45,"tag":65,"props":1187,"children":1189},{"className":1188},[],[1190],{"type":51,"value":1191},"direction=\"maximize\"",{"type":51,"value":1193}," and a custom\n",{"type":45,"tag":65,"props":1195,"children":1197},{"className":1196},[],[1198],{"type":51,"value":1199},"metric_extractor",{"type":51,"value":686},{"type":45,"tag":54,"props":1202,"children":1203},{},[1204,1205,1210,1212,1217,1219,1225,1227,1232],{"type":51,"value":941},{"type":45,"tag":65,"props":1206,"children":1208},{"className":1207},[],[1209],{"type":51,"value":187},{"type":51,"value":1211}," for the full multi-GPU spec-consistency\nrule (8-GPU example, NCCL timeout note) and the full AutoML\u002FHPO notes (metric\nselection, ",{"type":45,"tag":65,"props":1213,"children":1215},{"className":1214},[],[1216],{"type":51,"value":1199},{"type":51,"value":1218},", recommended hyperparameters, ",{"type":45,"tag":65,"props":1220,"children":1222},{"className":1221},[],[1223],{"type":51,"value":1224},"weight_decay",{"type":51,"value":1226},"\nbehavior, dense-dataset resume guidance). See ",{"type":45,"tag":65,"props":1228,"children":1230},{"className":1229},[],[1231],{"type":51,"value":198},{"type":51,"value":1233}," for\nAutoML metric extractor code, SDK orchestration internals, and parent-model\ninference mappings.",{"type":45,"tag":82,"props":1235,"children":1237},{"id":1236},"error-patterns",[1238],{"type":51,"value":1239},"Error Patterns",{"type":45,"tag":54,"props":1241,"children":1242},{},[1243,1245,1250,1252,1257,1258,1263,1265,1270,1272,1278,1280,1286,1287,1293],{"type":51,"value":1244},"Common failures include CUDA OOM (reduce ",{"type":45,"tag":65,"props":1246,"children":1248},{"className":1247},[],[1249],{"type":51,"value":1122},{"type":51,"value":1251},"), missing ",{"type":45,"tag":65,"props":1253,"children":1255},{"className":1254},[],[1256],{"type":51,"value":616},{"type":51,"value":1178},{"type":45,"tag":65,"props":1259,"children":1261},{"className":1260},[],[1262],{"type":51,"value":624},{"type":51,"value":1264}," at startup — always supply val), ",{"type":45,"tag":65,"props":1266,"children":1268},{"className":1267},[],[1269],{"type":51,"value":668},{"type":51,"value":1271}," too low (",{"type":45,"tag":65,"props":1273,"children":1275},{"className":1274},[],[1276],{"type":51,"value":1277},"CUDA device-side assert",{"type":51,"value":1279},"), and the parent ",{"type":45,"tag":65,"props":1281,"children":1283},{"className":1282},[],[1284],{"type":51,"value":1285},"dino gen_trt_engine",{"type":51,"value":362},{"type":45,"tag":65,"props":1288,"children":1290},{"className":1289},[],[1291],{"type":51,"value":1292},"dino convert",{"type":51,"value":1294}," PyT-CLI\nrestrictions.",{"type":45,"tag":54,"props":1296,"children":1297},{},[1298,1299,1304],{"type":51,"value":941},{"type":45,"tag":65,"props":1300,"children":1302},{"className":1301},[],[1303],{"type":51,"value":176},{"type":51,"value":1305}," for the complete error-pattern catalog\nwith diagnostics and fixes.",{"type":45,"tag":82,"props":1307,"children":1309},{"id":1308},"spec-param-parent-model-inference",[1310],{"type":51,"value":1311},"Spec Param \u002F Parent Model Inference",{"type":45,"tag":54,"props":1313,"children":1314},{},[1315,1317,1322,1324,1330,1332,1337,1338,1344,1346,1352],{"type":51,"value":1316},"Model-specific inference mappings belong in this MD file, not in ",{"type":45,"tag":65,"props":1318,"children":1320},{"className":1319},[],[1321],{"type":51,"value":919},{"type":51,"value":1323},".\nGenerated runners read the mappings and apply them with SDK helpers before\n",{"type":45,"tag":65,"props":1325,"children":1327},{"className":1326},[],[1328],{"type":51,"value":1329},"create_job()",{"type":51,"value":1331},". For ",{"type":45,"tag":65,"props":1333,"children":1335},{"className":1334},[],[1336],{"type":51,"value":975},{"type":51,"value":1102},{"type":45,"tag":65,"props":1339,"children":1341},{"className":1340},[],[1342],{"type":51,"value":1343},"parent_model_folder",{"type":51,"value":1345},", pass the upstream\ntrain\u002Fexport\u002FAutoML child job id as ",{"type":45,"tag":65,"props":1347,"children":1349},{"className":1348},[],[1350],{"type":51,"value":1351},"parent_job_id",{"type":51,"value":1353},"; the SDK lists the parent\nresult folder, filters checkpoint artifacts, and returns the selected model.",{"type":45,"tag":54,"props":1355,"children":1356},{},[1357,1358,1363,1365,1370,1371,1377,1378,1384,1385,1391,1392,1398,1399,1405],{"type":51,"value":941},{"type":45,"tag":65,"props":1359,"children":1361},{"className":1360},[],[1362],{"type":51,"value":198},{"type":51,"value":1364}," for the full inference-mapping table (per\naction: ",{"type":45,"tag":65,"props":1366,"children":1368},{"className":1367},[],[1369],{"type":51,"value":975},{"type":51,"value":304},{"type":45,"tag":65,"props":1372,"children":1374},{"className":1373},[],[1375],{"type":51,"value":1376},"key",{"type":51,"value":304},{"type":45,"tag":65,"props":1379,"children":1381},{"className":1380},[],[1382],{"type":51,"value":1383},"output_dir",{"type":51,"value":304},{"type":45,"tag":65,"props":1386,"children":1388},{"className":1387},[],[1389],{"type":51,"value":1390},"ptm_if_no_resume_model",{"type":51,"value":1018},{"type":45,"tag":65,"props":1393,"children":1395},{"className":1394},[],[1396],{"type":51,"value":1397},"resume_model",{"type":51,"value":304},{"type":45,"tag":65,"props":1400,"children":1402},{"className":1401},[],[1403],{"type":51,"value":1404},"create_onnx_file",{"type":51,"value":1406},") and the TensorRT-mapping note. TensorRT\nmappings live in the deploy workflow, not the PyT model skill.",{"type":45,"tag":82,"props":1408,"children":1410},{"id":1409},"optional-running-via-the-tao-sdk",[1411],{"type":51,"value":1412},"Optional: running via the TAO SDK",{"type":45,"tag":54,"props":1414,"children":1415},{},[1416,1418,1424,1426,1431,1433,1438,1440,1446,1448,1454,1456,1462],{"type":51,"value":1417},"When running DINO through the TAO SDK (",{"type":45,"tag":65,"props":1419,"children":1421},{"className":1420},[],[1422],{"type":51,"value":1423},"script_runner",{"type":51,"value":1425}," orchestration, S3 I\u002FO\nwrapping, AutoML), skills read ",{"type":45,"tag":65,"props":1427,"children":1429},{"className":1428},[],[1430],{"type":51,"value":270},{"type":51,"value":1432}," for input and\nspec-param mappings. See ",{"type":45,"tag":65,"props":1434,"children":1436},{"className":1435},[],[1437],{"type":51,"value":198},{"type":51,"value":1439}," for SDK orchestration\ninternals, including the data-sources gap and the ",{"type":45,"tag":65,"props":1441,"children":1443},{"className":1442},[],[1444],{"type":51,"value":1445},"[0]",{"type":51,"value":1447},"-indexed ",{"type":45,"tag":65,"props":1449,"children":1451},{"className":1450},[],[1452],{"type":51,"value":1453},"inputs",{"type":51,"value":1455},"\ndeclarations. Skip this when running locally with ",{"type":45,"tag":65,"props":1457,"children":1459},{"className":1458},[],[1460],{"type":51,"value":1461},"docker run",{"type":51,"value":686},{"type":45,"tag":82,"props":1464,"children":1466},{"id":1465},"deployment",[1467],{"type":51,"value":1468},"Deployment",{"type":45,"tag":153,"props":1470,"children":1471},{},[1472],{"type":45,"tag":157,"props":1473,"children":1474},{},[1475],{"type":45,"tag":1476,"props":1477,"children":1478},"a",{"href":127},[1479],{"type":51,"value":1480},"tao-deploy-dino",{"type":45,"tag":1482,"props":1483,"children":1484},"style",{},[1485],{"type":51,"value":1486},"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":1488,"total":1590},[1489,1506,1518,1532,1544,1561,1576],{"slug":1490,"name":1490,"fn":1491,"description":1492,"org":1493,"tags":1494,"stars":20,"repoUrl":21,"updatedAt":1505},"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},[1495,1498,1501,1502],{"name":1496,"slug":1497,"type":15},"Data Analysis","data-analysis",{"name":1499,"slug":1500,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1503,"slug":1504,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1507,"name":1507,"fn":1508,"description":1509,"org":1510,"tags":1511,"stars":20,"repoUrl":21,"updatedAt":1517},"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},[1512,1513,1516],{"name":1468,"slug":1465,"type":15},{"name":1514,"slug":1515,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1519,"name":1519,"fn":1520,"description":1521,"org":1522,"tags":1523,"stars":20,"repoUrl":21,"updatedAt":1531},"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},[1524,1527,1528],{"name":1525,"slug":1526,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1529,"slug":1530,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1533,"name":1533,"fn":1534,"description":1535,"org":1536,"tags":1537,"stars":20,"repoUrl":21,"updatedAt":1543},"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},[1538,1539,1540],{"name":1496,"slug":1497,"type":15},{"name":9,"slug":8,"type":15},{"name":1541,"slug":1542,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1545,"name":1545,"fn":1546,"description":1547,"org":1548,"tags":1549,"stars":20,"repoUrl":21,"updatedAt":1560},"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},[1550,1553,1556,1557],{"name":1551,"slug":1552,"type":15},"Automation","automation",{"name":1554,"slug":1555,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1558,"slug":1559,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1562,"name":1562,"fn":1563,"description":1564,"org":1565,"tags":1566,"stars":20,"repoUrl":21,"updatedAt":1575},"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},[1567,1568,1571,1572],{"name":1468,"slug":1465,"type":15},{"name":1569,"slug":1570,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1573,"slug":1574,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1577,"name":1577,"fn":1578,"description":1579,"org":1580,"tags":1581,"stars":20,"repoUrl":21,"updatedAt":1589},"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},[1582,1583,1586],{"name":9,"slug":8,"type":15},{"name":1584,"slug":1585,"type":15},"Quantum Computing","quantum-computing",{"name":1587,"slug":1588,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1592,"total":1742},[1593,1611,1626,1637,1649,1663,1676,1690,1701,1710,1724,1733],{"slug":1594,"name":1594,"fn":1595,"description":1596,"org":1597,"tags":1598,"stars":1608,"repoUrl":1609,"updatedAt":1610},"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},[1599,1602,1605],{"name":1600,"slug":1601,"type":15},"Documentation","documentation",{"name":1603,"slug":1604,"type":15},"MCP","mcp",{"name":1606,"slug":1607,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1612,"name":1612,"fn":1613,"description":1614,"org":1615,"tags":1616,"stars":1623,"repoUrl":1624,"updatedAt":1625},"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},[1617,1620,1621],{"name":1618,"slug":1619,"type":15},"Containers","containers",{"name":1468,"slug":1465,"type":15},{"name":1622,"slug":710,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1627,"name":1627,"fn":1628,"description":1629,"org":1630,"tags":1631,"stars":1623,"repoUrl":1624,"updatedAt":1636},"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},[1632,1635],{"name":1633,"slug":1634,"type":15},"CI\u002FCD","ci-cd",{"name":1468,"slug":1465,"type":15},"2026-07-14T05:25:59.97109",{"slug":1638,"name":1638,"fn":1639,"description":1640,"org":1641,"tags":1642,"stars":1623,"repoUrl":1624,"updatedAt":1648},"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},[1643,1644,1645],{"name":1633,"slug":1634,"type":15},{"name":1468,"slug":1465,"type":15},{"name":1646,"slug":1647,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1650,"name":1650,"fn":1651,"description":1652,"org":1653,"tags":1654,"stars":1623,"repoUrl":1624,"updatedAt":1662},"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},[1655,1658,1659],{"name":1656,"slug":1657,"type":15},"Debugging","debugging",{"name":1646,"slug":1647,"type":15},{"name":1660,"slug":1661,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1664,"name":1664,"fn":1665,"description":1666,"org":1667,"tags":1668,"stars":1623,"repoUrl":1624,"updatedAt":1675},"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},[1669,1672],{"name":1670,"slug":1671,"type":15},"Best Practices","best-practices",{"name":1673,"slug":1674,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1677,"name":1677,"fn":1678,"description":1679,"org":1680,"tags":1681,"stars":1623,"repoUrl":1624,"updatedAt":1689},"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},[1682,1685,1688],{"name":1683,"slug":1684,"type":15},"Machine Learning","machine-learning",{"name":1686,"slug":1687,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1691,"name":1691,"fn":1692,"description":1693,"org":1694,"tags":1695,"stars":1623,"repoUrl":1624,"updatedAt":1700},"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},[1696,1699],{"name":1697,"slug":1698,"type":15},"QA","qa",{"name":1541,"slug":1542,"type":15},"2026-07-14T05:25:53.673039",{"slug":1702,"name":1702,"fn":1703,"description":1704,"org":1705,"tags":1706,"stars":1623,"repoUrl":1624,"updatedAt":1709},"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},[1707,1708],{"name":1468,"slug":1465,"type":15},{"name":1514,"slug":1515,"type":15},"2026-07-14T05:25:49.362534",{"slug":1711,"name":1711,"fn":1712,"description":1713,"org":1714,"tags":1715,"stars":1623,"repoUrl":1624,"updatedAt":1723},"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},[1716,1719,1720],{"name":1717,"slug":1718,"type":15},"Code Review","code-review",{"name":1646,"slug":1647,"type":15},{"name":1721,"slug":1722,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1725,"name":1725,"fn":1726,"description":1727,"org":1728,"tags":1729,"stars":1623,"repoUrl":1624,"updatedAt":1732},"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},[1730,1731],{"name":1697,"slug":1698,"type":15},{"name":1541,"slug":1542,"type":15},"2026-07-14T05:25:54.928983",{"slug":1734,"name":1734,"fn":1735,"description":1736,"org":1737,"tags":1738,"stars":1623,"repoUrl":1624,"updatedAt":1741},"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},[1739,1740],{"name":1551,"slug":1552,"type":15},{"name":1633,"slug":1634,"type":15},"2026-07-30T05:29:03.275638",496]