[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-nvdinov2":3,"mdc--25kstx-key":34,"related-org-nvidia-tao-train-nvdinov2":1859,"related-repo-nvidia-tao-train-nvdinov2":2015},{"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-nvdinov2","train NVDINOv2 vision transformers","NVDINOv2 for self-supervised visual representation learning. Trains vision transformers via self-distillation (teacher-student) without labels and produces general-purpose visual features. Use when training, exporting, or running inference for a TAO NVDINOv2 backbone. Trigger phrases include \"train NVDINOv2\", \"self-supervised ViT pretraining\", \"DINOv2 backbone\", \"visual representation learning\".",{"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},"Machine Learning","machine-learning",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:28:01.846777","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-nvdinov2","---\nname: tao-train-nvdinov2\ndescription: NVDINOv2 for self-supervised visual representation learning. Trains vision transformers via self-distillation\n  (teacher-student) without labels and produces general-purpose visual features. Use when training, exporting, or\n  running inference for a TAO NVDINOv2 backbone. Trigger phrases include \"train NVDINOv2\", \"self-supervised ViT pretraining\",\n  \"DINOv2 backbone\", \"visual representation learning\".\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- self\n- supervised\n- learning\n---\n\n# NVDINOv2\n\nNVDINOv2 for self-supervised visual representation learning. Trains vision transformers via self-distillation (teacher-student) without labels. Produces general-purpose visual features.\n\nSet train.pretrained_model_path for pretrained ViT weights.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`), read `references\u002Ftao-deploy-nvdinov2.md` first. Deploy spec templates live in this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml` prefix.\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 `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\n- **Dataset type:** image_classification\n- **Formats:** ssl\n- **Monitoring metric:** train_loss\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| inference | dataset.test_dataset.images_dir | inference_dataset | images_test.tar.gz | No |\n| train | dataset.train_dataset.images_dir | train_datasets | images_train.tar.gz | No |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for train and inference** — 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 (mandatory data sources):**\n```python\n{\n    \"train.num_gpus\": 1,\n    \"train.num_epochs\": 10,\n    \"train.checkpoint_interval\": 10,\n    \"dataset.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages_train.tar.gz\",\n}\n```\n\n**local AutoML validation \u002F smoke run:**\nUse this shape when the goal is to confirm Bayesian launch, metric selection,\nbest-model choice, and checkpoint persistence on local Docker. It keeps the\nrun representative while avoiding the much slower ViT-Large default.\n\n```python\n{\n    \"wandb.enable\": False,\n    \"model.backbone.teacher_type\": \"vit_s\",\n    \"model.backbone.student_type\": \"vit_s\",\n    \"model.backbone.img_size\": 224,\n    \"dataset.batch_size\": 8,\n    \"dataset.workers\": 2,\n    \"train.num_epochs\": 1,\n    \"train.checkpoint_interval\": 1,\n    \"train.num_prototypes\": 1024,\n    \"train.precision\": \"32-true\",\n    \"train.use_custom_attention\": False,\n    \"train.num_gpus\": 1,\n    \"dataset.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages_train.tar.gz\",\n}\n```\n\n**export (mandatory checkpoint handoff):**\n```python\n{\n    \"export.checkpoint\": \"\u003Cselected train\u002FAutoML student_epoch_* checkpoint>\",\n    \"export.onnx_file\": \"\u002Fpath\u002Fto\u002Fresults\u002Fnvdinov2.onnx\",\n}\n```\n\n**inference (mandatory data sources):**\n```python\n{\n    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML student_epoch_* checkpoint>\",\n    \"model.backbone.teacher_type\": \"\u003Csame value used for train>\",\n    \"model.backbone.student_type\": \"\u003Csame value used for train>\",\n    \"model.backbone.img_size\": \"\u003Csame value used for train>\",\n    \"train.use_custom_attention\": \"\u003Csame value used for train>\",\n    \"dataset.test_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages_test.tar.gz\",\n}\n```\n## Eval Dataset\n\nOptional. SSL training does not use labels. Evaluation is downstream task-specific.\n\n## Important Parameters\n\n- **model.backbone.teacher_type**: Teacher ViT variant. Default vit_l (ViT-Large).\n- **model.backbone.student_type**: Student ViT variant. Default vit_l. Typically matches teacher.\n- **model.backbone.img_size**: Input image size. Default 518. Higher resolution produces better features but costs more memory.\n- **model.backbone.patch_size**: ViT patch size. Default 14.\n- **dataset.batch_size**: Per-GPU batch size. Default 4. SSL training is memory-intensive due to dual (teacher+student) forward passes.\n- **train.layerwise_decay**: Layer-wise learning rate decay. Important for ViT fine-tuning.\n- **train.clip_grad_norm**: Gradient clipping. Important for stable SSL training.\n\n## Multi-GPU \u002F Multi-Node\n\n**Launch method:** Lightning-managed (single `python` process, Lightning spawns workers).\n\n| Spec Key | Description | Default |\n|----------|-------------|---------|\n| `train.num_gpus` | Number of GPUs | 1 |\n| `train.gpu_ids` | GPU device indices | [0] |\n| `train.num_nodes` | Number of nodes | 1 |\n\n- Strategy: `auto` (Lightning picks best strategy automatically)\n- `sync_batchnorm` is always enabled — critical for SSL training with teacher-student framework\n- Multi-GPU strongly recommended (4-8 GPUs) for meaningful SSL training\n\n**Multi-node env vars** (set by orchestrator): `WORLD_SIZE`, `NODE_RANK`, `MASTER_ADDR`, `MASTER_PORT`, `NUM_GPU_PER_NODE`.\n\n## Hardware\n\nMinimum 4 GPU(s), recommended 8 GPU(s). 40GB+ (A100 recommended) VRAM per GPU. SSL with ViT-Large teacher+student is very memory-intensive. Requires A100 40GB+ GPUs. Multi-GPU strongly recommended.\n\n## Error Patterns\n\n**CUDA out of memory**: ViT-Large teacher+student with img_size=518 requires 40GB+ GPU memory. Reduce batch_size, img_size, or use smaller ViT variant.\n\n**Inference checkpoint has unexpected Lightning keys**: For downstream\n`inference`, pass the selected AutoML run's `student_epoch_*.pth` checkpoint,\nnot `nvdinov2_model_latest.pth`. The latest file is a training checkpoint and\nthe inference loader reports unexpected keys such as `state_dict`, optimizer\nstate, and scheduler state.\n\n**Export checkpoint has unexpected Lightning keys**: Export also consumes the\nselected `student_epoch_*.pth` checkpoint. Use the full `model_epoch_*.pth`\ncheckpoint only for resume\u002Fretrain via `train.resume_training_checkpoint_path`.\n\n**TensorRT engine passed to PyT inference**: The packaged PyT `nvdinov2 inference`\nimplementation only loads `.pth` or `.tlt` model paths. TAO Deploy\n`gen_trt_engine` builds a TensorRT engine for downstream consumers, but the PyT\ninference action does not run on that engine.\n\n**Separate distill action not available**: The current TAO PyT CLI exposes\n`export`, `inference`, `train`, and `default_specs` for NvDINOv2. Do not launch\nor advertise a standalone `nvdinov2 distill` action.\n\n**AutoML metric not found**: TAO's status KPI reports the final training scalar\nas `train_loss`. Use `train_loss` with minimize direction for AutoML selection.\nSome Lightning progress lines also render the same scalar as\n`train_loss_epoch`; treat that as a fallback alias only, not the primary\nmonitoring metric.\n\n**Slow convergence**: SSL needs many epochs. Default 10 is for quick testing; production runs typically use 100+ epochs.\n\n## Spec Param \u002F Parent Model Inference\n\nModel-specific inference mappings belong in this MD file, not in `config.json`. Generated runners should read this section and apply the mappings with SDK helpers before `create_job()`. This mirrors the old microservices `infer_params.py` flow.\n\nModel-specific handoff mappings:\n\n| Action | Spec Field | Inference Function | Meaning |\n|---|---|---|---|\n| export | `encryption_key` | `key` | encryption key |\n| export | `export.checkpoint` | `parent_model` | selected `student_epoch_*.pth` checkpoint from the parent job results folder |\n| export | `export.onnx_file` | `create_onnx_file` | output ONNX path |\n| export | `results_dir` | `output_dir` | current job results directory |\n| inference | `encryption_key` | `key` | encryption key |\n| inference | `inference.checkpoint` | `parent_model` | selected `student_epoch_*.pth` checkpoint from the parent job results folder |\n| inference | `results_dir` | `output_dir` | current job results directory |\n| train | `encryption_key` | `key` | encryption key |\n| train | `results_dir` | `output_dir` | current job results directory |\n| train | `train.pretrained_model_path` | `ptm_if_no_resume_model` | PTM when no resume checkpoint exists |\n| train | `train.resume_training_checkpoint_path` | `resume_model` | selected full `model_epoch_*.pth` training checkpoint from the current job results folder |\n\nFor `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. Do not add these mappings back to `config.json` and do not patch generated runner scripts to guess checkpoint paths.\n\n## Deployment\n\n- [tao-deploy-nvdinov2](references\u002Ftao-deploy-nvdinov2.md)\n",{"data":35,"body":45},{"name":4,"description":6,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"tags":41},"Requires docker + nvidia-container-toolkit.",{"version":38,"author":39},"0.1.0","NVIDIA Corporation","Read Bash",[42,43,44],"self","supervised","learning",{"type":46,"children":47},"root",[48,57,63,68,106,113,206,212,321,348,354,390,397,494,500,520,551,559,618,628,760,768,805,813,882,888,893,899,972,978,995,1089,1121,1166,1172,1177,1183,1193,1234,1266,1307,1352,1385,1395,1401,1430,1435,1801,1835,1841,1853],{"type":49,"tag":50,"props":51,"children":53},"element","h1",{"id":52},"nvdinov2",[54],{"type":55,"value":56},"text","NVDINOv2",{"type":49,"tag":58,"props":59,"children":60},"p",{},[61],{"type":55,"value":62},"NVDINOv2 for self-supervised visual representation learning. Trains vision transformers via self-distillation (teacher-student) without labels. Produces general-purpose visual features.",{"type":49,"tag":58,"props":64,"children":65},{},[66],{"type":55,"value":67},"Set train.pretrained_model_path for pretrained ViT weights.",{"type":49,"tag":58,"props":69,"children":70},{},[71,73,80,82,88,90,96,98,104],{"type":55,"value":72},"For TAO Deploy TensorRT actions (",{"type":49,"tag":74,"props":75,"children":77},"code",{"className":76},[],[78],{"type":55,"value":79},"gen_trt_engine",{"type":55,"value":81},"), read ",{"type":49,"tag":74,"props":83,"children":85},{"className":84},[],[86],{"type":55,"value":87},"references\u002Ftao-deploy-nvdinov2.md",{"type":55,"value":89}," first. Deploy spec templates live in this skill's ",{"type":49,"tag":74,"props":91,"children":93},{"className":92},[],[94],{"type":55,"value":95},"references\u002F",{"type":55,"value":97}," folder with the ",{"type":49,"tag":74,"props":99,"children":101},{"className":100},[],[102],{"type":55,"value":103},"spec_template_deploy_*.yaml",{"type":55,"value":105}," prefix.",{"type":49,"tag":107,"props":108,"children":110},"h2",{"id":109},"dataclass-schemas",[111],{"type":55,"value":112},"Dataclass Schemas",{"type":49,"tag":58,"props":114,"children":115},{},[116,118,124,126,132,134,140,142,148,150,156,158,164,166,172,174,180,182,188,190,196,198,204],{"type":55,"value":117},"Generated TAO Core schemas are packaged in ",{"type":49,"tag":74,"props":119,"children":121},{"className":120},[],[122],{"type":55,"value":123},"schemas\u002F\u003Caction>.schema.json",{"type":55,"value":125},", with ",{"type":49,"tag":74,"props":127,"children":129},{"className":128},[],[130],{"type":55,"value":131},"schemas\u002Fmanifest.json",{"type":55,"value":133}," listing available actions. Each generated schema also emits ",{"type":49,"tag":74,"props":135,"children":137},{"className":136},[],[138],{"type":55,"value":139},"references\u002Fspec_template_\u003Caction>.yaml",{"type":55,"value":141}," from the schema top-level ",{"type":49,"tag":74,"props":143,"children":145},{"className":144},[],[146],{"type":55,"value":147},"default",{"type":55,"value":149}," field. AutoML enablement is declared at the model layer in ",{"type":49,"tag":74,"props":151,"children":153},{"className":152},[],[154],{"type":55,"value":155},"references\u002Fskill_info.yaml",{"type":55,"value":157}," via ",{"type":49,"tag":74,"props":159,"children":161},{"className":160},[],[162],{"type":55,"value":163},"automl_enabled",{"type":55,"value":165},". Runnable AutoML still requires ",{"type":49,"tag":74,"props":167,"children":169},{"className":168},[],[170],{"type":55,"value":171},"schemas\u002Ftrain.schema.json",{"type":55,"value":173}," and ",{"type":49,"tag":74,"props":175,"children":177},{"className":176},[],[178],{"type":55,"value":179},"references\u002Fspec_template_train.yaml",{"type":55,"value":181}," to exist and parse. Use the packaged train schema for ",{"type":49,"tag":74,"props":183,"children":185},{"className":184},[],[186],{"type":55,"value":187},"automl_default_parameters",{"type":55,"value":189},", ",{"type":49,"tag":74,"props":191,"children":193},{"className":192},[],[194],{"type":55,"value":195},"automl_disabled_parameters",{"type":55,"value":197},", defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ",{"type":49,"tag":74,"props":199,"children":201},{"className":200},[],[202],{"type":55,"value":203},"~\u002Ftao-core",{"type":55,"value":205}," at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.",{"type":49,"tag":107,"props":207,"children":209},{"id":208},"train-action-policy",[210],{"type":55,"value":211},"Train Action Policy",{"type":49,"tag":58,"props":213,"children":214},{},[215,217,222,224,230,232,238,240,246,248,254,256,262,264,269,270,276,278,283,284,289,291,297,299,305,307,312,314,319],{"type":55,"value":216},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":49,"tag":74,"props":218,"children":220},{"className":219},[],[221],{"type":55,"value":155},{"type":55,"value":223}," and resolve the run override from either an explicit ",{"type":49,"tag":74,"props":225,"children":227},{"className":226},[],[228],{"type":55,"value":229},"automl_policy",{"type":55,"value":231}," value or the user's workflow request. Use ",{"type":49,"tag":74,"props":233,"children":235},{"className":234},[],[236],{"type":55,"value":237},"automl_policy: on",{"type":55,"value":239}," by default and only expose ",{"type":49,"tag":74,"props":241,"children":243},{"className":242},[],[244],{"type":55,"value":245},"on",{"type":55,"value":247}," \u002F ",{"type":49,"tag":74,"props":249,"children":251},{"className":250},[],[252],{"type":55,"value":253},"off",{"type":55,"value":255}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":49,"tag":74,"props":257,"children":259},{"className":258},[],[260],{"type":55,"value":261},"automl_policy: off",{"type":55,"value":263}," for this run only. When ",{"type":49,"tag":74,"props":265,"children":267},{"className":266},[],[268],{"type":55,"value":237},{"type":55,"value":189},{"type":49,"tag":74,"props":271,"children":273},{"className":272},[],[274],{"type":55,"value":275},"automl_enabled: true",{"type":55,"value":277},", and both ",{"type":49,"tag":74,"props":279,"children":281},{"className":280},[],[282],{"type":55,"value":171},{"type":55,"value":173},{"type":49,"tag":74,"props":285,"children":287},{"className":286},[],[288],{"type":55,"value":179},{"type":55,"value":290}," are packaged, route the train action through ",{"type":49,"tag":74,"props":292,"children":294},{"className":293},[],[295],{"type":55,"value":296},"tao-skill-bank:tao-run-automl",{"type":55,"value":298}," by default with this model's ",{"type":49,"tag":74,"props":300,"children":302},{"className":301},[],[303],{"type":55,"value":304},"skill_dir",{"type":55,"value":306},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":49,"tag":74,"props":308,"children":310},{"className":309},[],[311],{"type":55,"value":229},{"type":55,"value":313},". Use direct model training only when ",{"type":49,"tag":74,"props":315,"children":317},{"className":316},[],[318],{"type":55,"value":261},{"type":55,"value":320}," 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":49,"tag":58,"props":322,"children":323},{},[324,326,332,333,339,341,346],{"type":55,"value":325},"Non-train actions such as ",{"type":49,"tag":74,"props":327,"children":329},{"className":328},[],[330],{"type":55,"value":331},"inference",{"type":55,"value":189},{"type":49,"tag":74,"props":334,"children":336},{"className":335},[],[337],{"type":55,"value":338},"export",{"type":55,"value":340},", and deploy flows stay in this model skill. The per-run ",{"type":49,"tag":74,"props":342,"children":344},{"className":343},[],[345],{"type":55,"value":229},{"type":55,"value":347}," override does not change model metadata.",{"type":49,"tag":107,"props":349,"children":351},{"id":350},"training-requirements",[352],{"type":55,"value":353},"Training Requirements",{"type":49,"tag":355,"props":356,"children":357},"ul",{},[358,370,380],{"type":49,"tag":359,"props":360,"children":361},"li",{},[362,368],{"type":49,"tag":363,"props":364,"children":365},"strong",{},[366],{"type":55,"value":367},"Dataset type:",{"type":55,"value":369}," image_classification",{"type":49,"tag":359,"props":371,"children":372},{},[373,378],{"type":49,"tag":363,"props":374,"children":375},{},[376],{"type":55,"value":377},"Formats:",{"type":55,"value":379}," ssl",{"type":49,"tag":359,"props":381,"children":382},{},[383,388],{"type":49,"tag":363,"props":384,"children":385},{},[386],{"type":55,"value":387},"Monitoring metric:",{"type":55,"value":389}," train_loss",{"type":49,"tag":391,"props":392,"children":394},"h3",{"id":393},"per-action-dataset-requirements",[395],{"type":55,"value":396},"Per-Action Dataset Requirements",{"type":49,"tag":398,"props":399,"children":400},"table",{},[401,435],{"type":49,"tag":402,"props":403,"children":404},"thead",{},[405],{"type":49,"tag":406,"props":407,"children":408},"tr",{},[409,415,420,425,430],{"type":49,"tag":410,"props":411,"children":412},"th",{},[413],{"type":55,"value":414},"Action",{"type":49,"tag":410,"props":416,"children":417},{},[418],{"type":55,"value":419},"Spec Key",{"type":49,"tag":410,"props":421,"children":422},{},[423],{"type":55,"value":424},"Source",{"type":49,"tag":410,"props":426,"children":427},{},[428],{"type":55,"value":429},"Files",{"type":49,"tag":410,"props":431,"children":432},{},[433],{"type":55,"value":434},"List?",{"type":49,"tag":436,"props":437,"children":438},"tbody",{},[439,467],{"type":49,"tag":406,"props":440,"children":441},{},[442,447,452,457,462],{"type":49,"tag":443,"props":444,"children":445},"td",{},[446],{"type":55,"value":331},{"type":49,"tag":443,"props":448,"children":449},{},[450],{"type":55,"value":451},"dataset.test_dataset.images_dir",{"type":49,"tag":443,"props":453,"children":454},{},[455],{"type":55,"value":456},"inference_dataset",{"type":49,"tag":443,"props":458,"children":459},{},[460],{"type":55,"value":461},"images_test.tar.gz",{"type":49,"tag":443,"props":463,"children":464},{},[465],{"type":55,"value":466},"No",{"type":49,"tag":406,"props":468,"children":469},{},[470,475,480,485,490],{"type":49,"tag":443,"props":471,"children":472},{},[473],{"type":55,"value":474},"train",{"type":49,"tag":443,"props":476,"children":477},{},[478],{"type":55,"value":479},"dataset.train_dataset.images_dir",{"type":49,"tag":443,"props":481,"children":482},{},[483],{"type":55,"value":484},"train_datasets",{"type":49,"tag":443,"props":486,"children":487},{},[488],{"type":55,"value":489},"images_train.tar.gz",{"type":49,"tag":443,"props":491,"children":492},{},[493],{"type":55,"value":466},{"type":49,"tag":391,"props":495,"children":497},{"id":496},"typical-spec-overrides",[498],{"type":55,"value":499},"Typical Spec Overrides",{"type":49,"tag":58,"props":501,"children":502},{},[503,505,510,512,518],{"type":55,"value":504},"Data source overrides are ",{"type":49,"tag":363,"props":506,"children":507},{},[508],{"type":55,"value":509},"mandatory for train and inference",{"type":55,"value":511}," — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in ",{"type":49,"tag":74,"props":513,"children":515},{"className":514},[],[516],{"type":55,"value":517},"spec_overrides",{"type":55,"value":519},".",{"type":49,"tag":521,"props":522,"children":527},"pre",{"className":523,"code":524,"language":525,"meta":526,"style":526},"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","",[528],{"type":49,"tag":74,"props":529,"children":530},{"__ignoreMap":526},[531,542],{"type":49,"tag":532,"props":533,"children":536},"span",{"class":534,"line":535},"line",1,[537],{"type":49,"tag":532,"props":538,"children":539},{},[540],{"type":55,"value":541},"S3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\n",{"type":49,"tag":532,"props":543,"children":545},{"class":534,"line":544},2,[546],{"type":49,"tag":532,"props":547,"children":548},{},[549],{"type":55,"value":550},"S3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n",{"type":49,"tag":58,"props":552,"children":553},{},[554],{"type":49,"tag":363,"props":555,"children":556},{},[557],{"type":55,"value":558},"train (mandatory data sources):",{"type":49,"tag":521,"props":560,"children":562},{"className":523,"code":561,"language":525,"meta":526,"style":526},"{\n    \"train.num_gpus\": 1,\n    \"train.num_epochs\": 10,\n    \"train.checkpoint_interval\": 10,\n    \"dataset.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages_train.tar.gz\",\n}\n",[563],{"type":49,"tag":74,"props":564,"children":565},{"__ignoreMap":526},[566,574,582,591,600,609],{"type":49,"tag":532,"props":567,"children":568},{"class":534,"line":535},[569],{"type":49,"tag":532,"props":570,"children":571},{},[572],{"type":55,"value":573},"{\n",{"type":49,"tag":532,"props":575,"children":576},{"class":534,"line":544},[577],{"type":49,"tag":532,"props":578,"children":579},{},[580],{"type":55,"value":581},"    \"train.num_gpus\": 1,\n",{"type":49,"tag":532,"props":583,"children":585},{"class":534,"line":584},3,[586],{"type":49,"tag":532,"props":587,"children":588},{},[589],{"type":55,"value":590},"    \"train.num_epochs\": 10,\n",{"type":49,"tag":532,"props":592,"children":594},{"class":534,"line":593},4,[595],{"type":49,"tag":532,"props":596,"children":597},{},[598],{"type":55,"value":599},"    \"train.checkpoint_interval\": 10,\n",{"type":49,"tag":532,"props":601,"children":603},{"class":534,"line":602},5,[604],{"type":49,"tag":532,"props":605,"children":606},{},[607],{"type":55,"value":608},"    \"dataset.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages_train.tar.gz\",\n",{"type":49,"tag":532,"props":610,"children":612},{"class":534,"line":611},6,[613],{"type":49,"tag":532,"props":614,"children":615},{},[616],{"type":55,"value":617},"}\n",{"type":49,"tag":58,"props":619,"children":620},{},[621,626],{"type":49,"tag":363,"props":622,"children":623},{},[624],{"type":55,"value":625},"local AutoML validation \u002F smoke run:",{"type":55,"value":627},"\nUse this shape when the goal is to confirm Bayesian launch, metric selection,\nbest-model choice, and checkpoint persistence on local Docker. It keeps the\nrun representative while avoiding the much slower ViT-Large default.",{"type":49,"tag":521,"props":629,"children":631},{"className":523,"code":630,"language":525,"meta":526,"style":526},"{\n    \"wandb.enable\": False,\n    \"model.backbone.teacher_type\": \"vit_s\",\n    \"model.backbone.student_type\": \"vit_s\",\n    \"model.backbone.img_size\": 224,\n    \"dataset.batch_size\": 8,\n    \"dataset.workers\": 2,\n    \"train.num_epochs\": 1,\n    \"train.checkpoint_interval\": 1,\n    \"train.num_prototypes\": 1024,\n    \"train.precision\": \"32-true\",\n    \"train.use_custom_attention\": False,\n    \"train.num_gpus\": 1,\n    \"dataset.train_dataset.images_dir\": f\"{S3_TRAIN}\u002Fimages_train.tar.gz\",\n}\n",[632],{"type":49,"tag":74,"props":633,"children":634},{"__ignoreMap":526},[635,642,650,658,666,674,682,691,700,709,718,727,736,744,752],{"type":49,"tag":532,"props":636,"children":637},{"class":534,"line":535},[638],{"type":49,"tag":532,"props":639,"children":640},{},[641],{"type":55,"value":573},{"type":49,"tag":532,"props":643,"children":644},{"class":534,"line":544},[645],{"type":49,"tag":532,"props":646,"children":647},{},[648],{"type":55,"value":649},"    \"wandb.enable\": False,\n",{"type":49,"tag":532,"props":651,"children":652},{"class":534,"line":584},[653],{"type":49,"tag":532,"props":654,"children":655},{},[656],{"type":55,"value":657},"    \"model.backbone.teacher_type\": \"vit_s\",\n",{"type":49,"tag":532,"props":659,"children":660},{"class":534,"line":593},[661],{"type":49,"tag":532,"props":662,"children":663},{},[664],{"type":55,"value":665},"    \"model.backbone.student_type\": \"vit_s\",\n",{"type":49,"tag":532,"props":667,"children":668},{"class":534,"line":602},[669],{"type":49,"tag":532,"props":670,"children":671},{},[672],{"type":55,"value":673},"    \"model.backbone.img_size\": 224,\n",{"type":49,"tag":532,"props":675,"children":676},{"class":534,"line":611},[677],{"type":49,"tag":532,"props":678,"children":679},{},[680],{"type":55,"value":681},"    \"dataset.batch_size\": 8,\n",{"type":49,"tag":532,"props":683,"children":685},{"class":534,"line":684},7,[686],{"type":49,"tag":532,"props":687,"children":688},{},[689],{"type":55,"value":690},"    \"dataset.workers\": 2,\n",{"type":49,"tag":532,"props":692,"children":694},{"class":534,"line":693},8,[695],{"type":49,"tag":532,"props":696,"children":697},{},[698],{"type":55,"value":699},"    \"train.num_epochs\": 1,\n",{"type":49,"tag":532,"props":701,"children":703},{"class":534,"line":702},9,[704],{"type":49,"tag":532,"props":705,"children":706},{},[707],{"type":55,"value":708},"    \"train.checkpoint_interval\": 1,\n",{"type":49,"tag":532,"props":710,"children":712},{"class":534,"line":711},10,[713],{"type":49,"tag":532,"props":714,"children":715},{},[716],{"type":55,"value":717},"    \"train.num_prototypes\": 1024,\n",{"type":49,"tag":532,"props":719,"children":721},{"class":534,"line":720},11,[722],{"type":49,"tag":532,"props":723,"children":724},{},[725],{"type":55,"value":726},"    \"train.precision\": \"32-true\",\n",{"type":49,"tag":532,"props":728,"children":730},{"class":534,"line":729},12,[731],{"type":49,"tag":532,"props":732,"children":733},{},[734],{"type":55,"value":735},"    \"train.use_custom_attention\": False,\n",{"type":49,"tag":532,"props":737,"children":739},{"class":534,"line":738},13,[740],{"type":49,"tag":532,"props":741,"children":742},{},[743],{"type":55,"value":581},{"type":49,"tag":532,"props":745,"children":747},{"class":534,"line":746},14,[748],{"type":49,"tag":532,"props":749,"children":750},{},[751],{"type":55,"value":608},{"type":49,"tag":532,"props":753,"children":755},{"class":534,"line":754},15,[756],{"type":49,"tag":532,"props":757,"children":758},{},[759],{"type":55,"value":617},{"type":49,"tag":58,"props":761,"children":762},{},[763],{"type":49,"tag":363,"props":764,"children":765},{},[766],{"type":55,"value":767},"export (mandatory checkpoint handoff):",{"type":49,"tag":521,"props":769,"children":771},{"className":523,"code":770,"language":525,"meta":526,"style":526},"{\n    \"export.checkpoint\": \"\u003Cselected train\u002FAutoML student_epoch_* checkpoint>\",\n    \"export.onnx_file\": \"\u002Fpath\u002Fto\u002Fresults\u002Fnvdinov2.onnx\",\n}\n",[772],{"type":49,"tag":74,"props":773,"children":774},{"__ignoreMap":526},[775,782,790,798],{"type":49,"tag":532,"props":776,"children":777},{"class":534,"line":535},[778],{"type":49,"tag":532,"props":779,"children":780},{},[781],{"type":55,"value":573},{"type":49,"tag":532,"props":783,"children":784},{"class":534,"line":544},[785],{"type":49,"tag":532,"props":786,"children":787},{},[788],{"type":55,"value":789},"    \"export.checkpoint\": \"\u003Cselected train\u002FAutoML student_epoch_* checkpoint>\",\n",{"type":49,"tag":532,"props":791,"children":792},{"class":534,"line":584},[793],{"type":49,"tag":532,"props":794,"children":795},{},[796],{"type":55,"value":797},"    \"export.onnx_file\": \"\u002Fpath\u002Fto\u002Fresults\u002Fnvdinov2.onnx\",\n",{"type":49,"tag":532,"props":799,"children":800},{"class":534,"line":593},[801],{"type":49,"tag":532,"props":802,"children":803},{},[804],{"type":55,"value":617},{"type":49,"tag":58,"props":806,"children":807},{},[808],{"type":49,"tag":363,"props":809,"children":810},{},[811],{"type":55,"value":812},"inference (mandatory data sources):",{"type":49,"tag":521,"props":814,"children":816},{"className":523,"code":815,"language":525,"meta":526,"style":526},"{\n    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML student_epoch_* checkpoint>\",\n    \"model.backbone.teacher_type\": \"\u003Csame value used for train>\",\n    \"model.backbone.student_type\": \"\u003Csame value used for train>\",\n    \"model.backbone.img_size\": \"\u003Csame value used for train>\",\n    \"train.use_custom_attention\": \"\u003Csame value used for train>\",\n    \"dataset.test_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages_test.tar.gz\",\n}\n",[817],{"type":49,"tag":74,"props":818,"children":819},{"__ignoreMap":526},[820,827,835,843,851,859,867,875],{"type":49,"tag":532,"props":821,"children":822},{"class":534,"line":535},[823],{"type":49,"tag":532,"props":824,"children":825},{},[826],{"type":55,"value":573},{"type":49,"tag":532,"props":828,"children":829},{"class":534,"line":544},[830],{"type":49,"tag":532,"props":831,"children":832},{},[833],{"type":55,"value":834},"    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML student_epoch_* checkpoint>\",\n",{"type":49,"tag":532,"props":836,"children":837},{"class":534,"line":584},[838],{"type":49,"tag":532,"props":839,"children":840},{},[841],{"type":55,"value":842},"    \"model.backbone.teacher_type\": \"\u003Csame value used for train>\",\n",{"type":49,"tag":532,"props":844,"children":845},{"class":534,"line":593},[846],{"type":49,"tag":532,"props":847,"children":848},{},[849],{"type":55,"value":850},"    \"model.backbone.student_type\": \"\u003Csame value used for train>\",\n",{"type":49,"tag":532,"props":852,"children":853},{"class":534,"line":602},[854],{"type":49,"tag":532,"props":855,"children":856},{},[857],{"type":55,"value":858},"    \"model.backbone.img_size\": \"\u003Csame value used for train>\",\n",{"type":49,"tag":532,"props":860,"children":861},{"class":534,"line":611},[862],{"type":49,"tag":532,"props":863,"children":864},{},[865],{"type":55,"value":866},"    \"train.use_custom_attention\": \"\u003Csame value used for train>\",\n",{"type":49,"tag":532,"props":868,"children":869},{"class":534,"line":684},[870],{"type":49,"tag":532,"props":871,"children":872},{},[873],{"type":55,"value":874},"    \"dataset.test_dataset.images_dir\": f\"{S3_EVAL}\u002Fimages_test.tar.gz\",\n",{"type":49,"tag":532,"props":876,"children":877},{"class":534,"line":693},[878],{"type":49,"tag":532,"props":879,"children":880},{},[881],{"type":55,"value":617},{"type":49,"tag":107,"props":883,"children":885},{"id":884},"eval-dataset",[886],{"type":55,"value":887},"Eval Dataset",{"type":49,"tag":58,"props":889,"children":890},{},[891],{"type":55,"value":892},"Optional. SSL training does not use labels. Evaluation is downstream task-specific.",{"type":49,"tag":107,"props":894,"children":896},{"id":895},"important-parameters",[897],{"type":55,"value":898},"Important Parameters",{"type":49,"tag":355,"props":900,"children":901},{},[902,912,922,932,942,952,962],{"type":49,"tag":359,"props":903,"children":904},{},[905,910],{"type":49,"tag":363,"props":906,"children":907},{},[908],{"type":55,"value":909},"model.backbone.teacher_type",{"type":55,"value":911},": Teacher ViT variant. Default vit_l (ViT-Large).",{"type":49,"tag":359,"props":913,"children":914},{},[915,920],{"type":49,"tag":363,"props":916,"children":917},{},[918],{"type":55,"value":919},"model.backbone.student_type",{"type":55,"value":921},": Student ViT variant. Default vit_l. Typically matches teacher.",{"type":49,"tag":359,"props":923,"children":924},{},[925,930],{"type":49,"tag":363,"props":926,"children":927},{},[928],{"type":55,"value":929},"model.backbone.img_size",{"type":55,"value":931},": Input image size. Default 518. Higher resolution produces better features but costs more memory.",{"type":49,"tag":359,"props":933,"children":934},{},[935,940],{"type":49,"tag":363,"props":936,"children":937},{},[938],{"type":55,"value":939},"model.backbone.patch_size",{"type":55,"value":941},": ViT patch size. Default 14.",{"type":49,"tag":359,"props":943,"children":944},{},[945,950],{"type":49,"tag":363,"props":946,"children":947},{},[948],{"type":55,"value":949},"dataset.batch_size",{"type":55,"value":951},": Per-GPU batch size. Default 4. SSL training is memory-intensive due to dual (teacher+student) forward passes.",{"type":49,"tag":359,"props":953,"children":954},{},[955,960],{"type":49,"tag":363,"props":956,"children":957},{},[958],{"type":55,"value":959},"train.layerwise_decay",{"type":55,"value":961},": Layer-wise learning rate decay. Important for ViT fine-tuning.",{"type":49,"tag":359,"props":963,"children":964},{},[965,970],{"type":49,"tag":363,"props":966,"children":967},{},[968],{"type":55,"value":969},"train.clip_grad_norm",{"type":55,"value":971},": Gradient clipping. Important for stable SSL training.",{"type":49,"tag":107,"props":973,"children":975},{"id":974},"multi-gpu-multi-node",[976],{"type":55,"value":977},"Multi-GPU \u002F Multi-Node",{"type":49,"tag":58,"props":979,"children":980},{},[981,986,988,993],{"type":49,"tag":363,"props":982,"children":983},{},[984],{"type":55,"value":985},"Launch method:",{"type":55,"value":987}," Lightning-managed (single ",{"type":49,"tag":74,"props":989,"children":991},{"className":990},[],[992],{"type":55,"value":525},{"type":55,"value":994}," process, Lightning spawns workers).",{"type":49,"tag":398,"props":996,"children":997},{},[998,1018],{"type":49,"tag":402,"props":999,"children":1000},{},[1001],{"type":49,"tag":406,"props":1002,"children":1003},{},[1004,1008,1013],{"type":49,"tag":410,"props":1005,"children":1006},{},[1007],{"type":55,"value":419},{"type":49,"tag":410,"props":1009,"children":1010},{},[1011],{"type":55,"value":1012},"Description",{"type":49,"tag":410,"props":1014,"children":1015},{},[1016],{"type":55,"value":1017},"Default",{"type":49,"tag":436,"props":1019,"children":1020},{},[1021,1043,1068],{"type":49,"tag":406,"props":1022,"children":1023},{},[1024,1033,1038],{"type":49,"tag":443,"props":1025,"children":1026},{},[1027],{"type":49,"tag":74,"props":1028,"children":1030},{"className":1029},[],[1031],{"type":55,"value":1032},"train.num_gpus",{"type":49,"tag":443,"props":1034,"children":1035},{},[1036],{"type":55,"value":1037},"Number of GPUs",{"type":49,"tag":443,"props":1039,"children":1040},{},[1041],{"type":55,"value":1042},"1",{"type":49,"tag":406,"props":1044,"children":1045},{},[1046,1055,1060],{"type":49,"tag":443,"props":1047,"children":1048},{},[1049],{"type":49,"tag":74,"props":1050,"children":1052},{"className":1051},[],[1053],{"type":55,"value":1054},"train.gpu_ids",{"type":49,"tag":443,"props":1056,"children":1057},{},[1058],{"type":55,"value":1059},"GPU device indices",{"type":49,"tag":443,"props":1061,"children":1062},{},[1063],{"type":49,"tag":532,"props":1064,"children":1065},{},[1066],{"type":55,"value":1067},"0",{"type":49,"tag":406,"props":1069,"children":1070},{},[1071,1080,1085],{"type":49,"tag":443,"props":1072,"children":1073},{},[1074],{"type":49,"tag":74,"props":1075,"children":1077},{"className":1076},[],[1078],{"type":55,"value":1079},"train.num_nodes",{"type":49,"tag":443,"props":1081,"children":1082},{},[1083],{"type":55,"value":1084},"Number of nodes",{"type":49,"tag":443,"props":1086,"children":1087},{},[1088],{"type":55,"value":1042},{"type":49,"tag":355,"props":1090,"children":1091},{},[1092,1105,1116],{"type":49,"tag":359,"props":1093,"children":1094},{},[1095,1097,1103],{"type":55,"value":1096},"Strategy: ",{"type":49,"tag":74,"props":1098,"children":1100},{"className":1099},[],[1101],{"type":55,"value":1102},"auto",{"type":55,"value":1104}," (Lightning picks best strategy automatically)",{"type":49,"tag":359,"props":1106,"children":1107},{},[1108,1114],{"type":49,"tag":74,"props":1109,"children":1111},{"className":1110},[],[1112],{"type":55,"value":1113},"sync_batchnorm",{"type":55,"value":1115}," is always enabled — critical for SSL training with teacher-student framework",{"type":49,"tag":359,"props":1117,"children":1118},{},[1119],{"type":55,"value":1120},"Multi-GPU strongly recommended (4-8 GPUs) for meaningful SSL training",{"type":49,"tag":58,"props":1122,"children":1123},{},[1124,1129,1131,1137,1138,1144,1145,1151,1152,1158,1159,1165],{"type":49,"tag":363,"props":1125,"children":1126},{},[1127],{"type":55,"value":1128},"Multi-node env vars",{"type":55,"value":1130}," (set by orchestrator): ",{"type":49,"tag":74,"props":1132,"children":1134},{"className":1133},[],[1135],{"type":55,"value":1136},"WORLD_SIZE",{"type":55,"value":189},{"type":49,"tag":74,"props":1139,"children":1141},{"className":1140},[],[1142],{"type":55,"value":1143},"NODE_RANK",{"type":55,"value":189},{"type":49,"tag":74,"props":1146,"children":1148},{"className":1147},[],[1149],{"type":55,"value":1150},"MASTER_ADDR",{"type":55,"value":189},{"type":49,"tag":74,"props":1153,"children":1155},{"className":1154},[],[1156],{"type":55,"value":1157},"MASTER_PORT",{"type":55,"value":189},{"type":49,"tag":74,"props":1160,"children":1162},{"className":1161},[],[1163],{"type":55,"value":1164},"NUM_GPU_PER_NODE",{"type":55,"value":519},{"type":49,"tag":107,"props":1167,"children":1169},{"id":1168},"hardware",[1170],{"type":55,"value":1171},"Hardware",{"type":49,"tag":58,"props":1173,"children":1174},{},[1175],{"type":55,"value":1176},"Minimum 4 GPU(s), recommended 8 GPU(s). 40GB+ (A100 recommended) VRAM per GPU. SSL with ViT-Large teacher+student is very memory-intensive. Requires A100 40GB+ GPUs. Multi-GPU strongly recommended.",{"type":49,"tag":107,"props":1178,"children":1180},{"id":1179},"error-patterns",[1181],{"type":55,"value":1182},"Error Patterns",{"type":49,"tag":58,"props":1184,"children":1185},{},[1186,1191],{"type":49,"tag":363,"props":1187,"children":1188},{},[1189],{"type":55,"value":1190},"CUDA out of memory",{"type":55,"value":1192},": ViT-Large teacher+student with img_size=518 requires 40GB+ GPU memory. Reduce batch_size, img_size, or use smaller ViT variant.",{"type":49,"tag":58,"props":1194,"children":1195},{},[1196,1201,1203,1208,1210,1216,1218,1224,1226,1232],{"type":49,"tag":363,"props":1197,"children":1198},{},[1199],{"type":55,"value":1200},"Inference checkpoint has unexpected Lightning keys",{"type":55,"value":1202},": For downstream\n",{"type":49,"tag":74,"props":1204,"children":1206},{"className":1205},[],[1207],{"type":55,"value":331},{"type":55,"value":1209},", pass the selected AutoML run's ",{"type":49,"tag":74,"props":1211,"children":1213},{"className":1212},[],[1214],{"type":55,"value":1215},"student_epoch_*.pth",{"type":55,"value":1217}," checkpoint,\nnot ",{"type":49,"tag":74,"props":1219,"children":1221},{"className":1220},[],[1222],{"type":55,"value":1223},"nvdinov2_model_latest.pth",{"type":55,"value":1225},". The latest file is a training checkpoint and\nthe inference loader reports unexpected keys such as ",{"type":49,"tag":74,"props":1227,"children":1229},{"className":1228},[],[1230],{"type":55,"value":1231},"state_dict",{"type":55,"value":1233},", optimizer\nstate, and scheduler state.",{"type":49,"tag":58,"props":1235,"children":1236},{},[1237,1242,1244,1249,1251,1257,1259,1265],{"type":49,"tag":363,"props":1238,"children":1239},{},[1240],{"type":55,"value":1241},"Export checkpoint has unexpected Lightning keys",{"type":55,"value":1243},": Export also consumes the\nselected ",{"type":49,"tag":74,"props":1245,"children":1247},{"className":1246},[],[1248],{"type":55,"value":1215},{"type":55,"value":1250}," checkpoint. Use the full ",{"type":49,"tag":74,"props":1252,"children":1254},{"className":1253},[],[1255],{"type":55,"value":1256},"model_epoch_*.pth",{"type":55,"value":1258},"\ncheckpoint only for resume\u002Fretrain via ",{"type":49,"tag":74,"props":1260,"children":1262},{"className":1261},[],[1263],{"type":55,"value":1264},"train.resume_training_checkpoint_path",{"type":55,"value":519},{"type":49,"tag":58,"props":1267,"children":1268},{},[1269,1274,1276,1282,1284,1290,1292,1298,1300,1305],{"type":49,"tag":363,"props":1270,"children":1271},{},[1272],{"type":55,"value":1273},"TensorRT engine passed to PyT inference",{"type":55,"value":1275},": The packaged PyT ",{"type":49,"tag":74,"props":1277,"children":1279},{"className":1278},[],[1280],{"type":55,"value":1281},"nvdinov2 inference",{"type":55,"value":1283},"\nimplementation only loads ",{"type":49,"tag":74,"props":1285,"children":1287},{"className":1286},[],[1288],{"type":55,"value":1289},".pth",{"type":55,"value":1291}," or ",{"type":49,"tag":74,"props":1293,"children":1295},{"className":1294},[],[1296],{"type":55,"value":1297},".tlt",{"type":55,"value":1299}," model paths. TAO Deploy\n",{"type":49,"tag":74,"props":1301,"children":1303},{"className":1302},[],[1304],{"type":55,"value":79},{"type":55,"value":1306}," builds a TensorRT engine for downstream consumers, but the PyT\ninference action does not run on that engine.",{"type":49,"tag":58,"props":1308,"children":1309},{},[1310,1315,1317,1322,1323,1328,1329,1334,1336,1342,1344,1350],{"type":49,"tag":363,"props":1311,"children":1312},{},[1313],{"type":55,"value":1314},"Separate distill action not available",{"type":55,"value":1316},": The current TAO PyT CLI exposes\n",{"type":49,"tag":74,"props":1318,"children":1320},{"className":1319},[],[1321],{"type":55,"value":338},{"type":55,"value":189},{"type":49,"tag":74,"props":1324,"children":1326},{"className":1325},[],[1327],{"type":55,"value":331},{"type":55,"value":189},{"type":49,"tag":74,"props":1330,"children":1332},{"className":1331},[],[1333],{"type":55,"value":474},{"type":55,"value":1335},", and ",{"type":49,"tag":74,"props":1337,"children":1339},{"className":1338},[],[1340],{"type":55,"value":1341},"default_specs",{"type":55,"value":1343}," for NvDINOv2. Do not launch\nor advertise a standalone ",{"type":49,"tag":74,"props":1345,"children":1347},{"className":1346},[],[1348],{"type":55,"value":1349},"nvdinov2 distill",{"type":55,"value":1351}," action.",{"type":49,"tag":58,"props":1353,"children":1354},{},[1355,1360,1362,1368,1370,1375,1377,1383],{"type":49,"tag":363,"props":1356,"children":1357},{},[1358],{"type":55,"value":1359},"AutoML metric not found",{"type":55,"value":1361},": TAO's status KPI reports the final training scalar\nas ",{"type":49,"tag":74,"props":1363,"children":1365},{"className":1364},[],[1366],{"type":55,"value":1367},"train_loss",{"type":55,"value":1369},". Use ",{"type":49,"tag":74,"props":1371,"children":1373},{"className":1372},[],[1374],{"type":55,"value":1367},{"type":55,"value":1376}," with minimize direction for AutoML selection.\nSome Lightning progress lines also render the same scalar as\n",{"type":49,"tag":74,"props":1378,"children":1380},{"className":1379},[],[1381],{"type":55,"value":1382},"train_loss_epoch",{"type":55,"value":1384},"; treat that as a fallback alias only, not the primary\nmonitoring metric.",{"type":49,"tag":58,"props":1386,"children":1387},{},[1388,1393],{"type":49,"tag":363,"props":1389,"children":1390},{},[1391],{"type":55,"value":1392},"Slow convergence",{"type":55,"value":1394},": SSL needs many epochs. Default 10 is for quick testing; production runs typically use 100+ epochs.",{"type":49,"tag":107,"props":1396,"children":1398},{"id":1397},"spec-param-parent-model-inference",[1399],{"type":55,"value":1400},"Spec Param \u002F Parent Model Inference",{"type":49,"tag":58,"props":1402,"children":1403},{},[1404,1406,1412,1414,1420,1422,1428],{"type":55,"value":1405},"Model-specific inference mappings belong in this MD file, not in ",{"type":49,"tag":74,"props":1407,"children":1409},{"className":1408},[],[1410],{"type":55,"value":1411},"config.json",{"type":55,"value":1413},". Generated runners should read this section and apply the mappings with SDK helpers before ",{"type":49,"tag":74,"props":1415,"children":1417},{"className":1416},[],[1418],{"type":55,"value":1419},"create_job()",{"type":55,"value":1421},". This mirrors the old microservices ",{"type":49,"tag":74,"props":1423,"children":1425},{"className":1424},[],[1426],{"type":55,"value":1427},"infer_params.py",{"type":55,"value":1429}," flow.",{"type":49,"tag":58,"props":1431,"children":1432},{},[1433],{"type":55,"value":1434},"Model-specific handoff mappings:",{"type":49,"tag":398,"props":1436,"children":1437},{},[1438,1463],{"type":49,"tag":402,"props":1439,"children":1440},{},[1441],{"type":49,"tag":406,"props":1442,"children":1443},{},[1444,1448,1453,1458],{"type":49,"tag":410,"props":1445,"children":1446},{},[1447],{"type":55,"value":414},{"type":49,"tag":410,"props":1449,"children":1450},{},[1451],{"type":55,"value":1452},"Spec Field",{"type":49,"tag":410,"props":1454,"children":1455},{},[1456],{"type":55,"value":1457},"Inference Function",{"type":49,"tag":410,"props":1459,"children":1460},{},[1461],{"type":55,"value":1462},"Meaning",{"type":49,"tag":436,"props":1464,"children":1465},{},[1466,1496,1533,1563,1593,1620,1654,1681,1708,1735,1765],{"type":49,"tag":406,"props":1467,"children":1468},{},[1469,1473,1482,1491],{"type":49,"tag":443,"props":1470,"children":1471},{},[1472],{"type":55,"value":338},{"type":49,"tag":443,"props":1474,"children":1475},{},[1476],{"type":49,"tag":74,"props":1477,"children":1479},{"className":1478},[],[1480],{"type":55,"value":1481},"encryption_key",{"type":49,"tag":443,"props":1483,"children":1484},{},[1485],{"type":49,"tag":74,"props":1486,"children":1488},{"className":1487},[],[1489],{"type":55,"value":1490},"key",{"type":49,"tag":443,"props":1492,"children":1493},{},[1494],{"type":55,"value":1495},"encryption key",{"type":49,"tag":406,"props":1497,"children":1498},{},[1499,1503,1512,1521],{"type":49,"tag":443,"props":1500,"children":1501},{},[1502],{"type":55,"value":338},{"type":49,"tag":443,"props":1504,"children":1505},{},[1506],{"type":49,"tag":74,"props":1507,"children":1509},{"className":1508},[],[1510],{"type":55,"value":1511},"export.checkpoint",{"type":49,"tag":443,"props":1513,"children":1514},{},[1515],{"type":49,"tag":74,"props":1516,"children":1518},{"className":1517},[],[1519],{"type":55,"value":1520},"parent_model",{"type":49,"tag":443,"props":1522,"children":1523},{},[1524,1526,1531],{"type":55,"value":1525},"selected ",{"type":49,"tag":74,"props":1527,"children":1529},{"className":1528},[],[1530],{"type":55,"value":1215},{"type":55,"value":1532}," checkpoint from the parent job results folder",{"type":49,"tag":406,"props":1534,"children":1535},{},[1536,1540,1549,1558],{"type":49,"tag":443,"props":1537,"children":1538},{},[1539],{"type":55,"value":338},{"type":49,"tag":443,"props":1541,"children":1542},{},[1543],{"type":49,"tag":74,"props":1544,"children":1546},{"className":1545},[],[1547],{"type":55,"value":1548},"export.onnx_file",{"type":49,"tag":443,"props":1550,"children":1551},{},[1552],{"type":49,"tag":74,"props":1553,"children":1555},{"className":1554},[],[1556],{"type":55,"value":1557},"create_onnx_file",{"type":49,"tag":443,"props":1559,"children":1560},{},[1561],{"type":55,"value":1562},"output ONNX path",{"type":49,"tag":406,"props":1564,"children":1565},{},[1566,1570,1579,1588],{"type":49,"tag":443,"props":1567,"children":1568},{},[1569],{"type":55,"value":338},{"type":49,"tag":443,"props":1571,"children":1572},{},[1573],{"type":49,"tag":74,"props":1574,"children":1576},{"className":1575},[],[1577],{"type":55,"value":1578},"results_dir",{"type":49,"tag":443,"props":1580,"children":1581},{},[1582],{"type":49,"tag":74,"props":1583,"children":1585},{"className":1584},[],[1586],{"type":55,"value":1587},"output_dir",{"type":49,"tag":443,"props":1589,"children":1590},{},[1591],{"type":55,"value":1592},"current job results directory",{"type":49,"tag":406,"props":1594,"children":1595},{},[1596,1600,1608,1616],{"type":49,"tag":443,"props":1597,"children":1598},{},[1599],{"type":55,"value":331},{"type":49,"tag":443,"props":1601,"children":1602},{},[1603],{"type":49,"tag":74,"props":1604,"children":1606},{"className":1605},[],[1607],{"type":55,"value":1481},{"type":49,"tag":443,"props":1609,"children":1610},{},[1611],{"type":49,"tag":74,"props":1612,"children":1614},{"className":1613},[],[1615],{"type":55,"value":1490},{"type":49,"tag":443,"props":1617,"children":1618},{},[1619],{"type":55,"value":1495},{"type":49,"tag":406,"props":1621,"children":1622},{},[1623,1627,1636,1644],{"type":49,"tag":443,"props":1624,"children":1625},{},[1626],{"type":55,"value":331},{"type":49,"tag":443,"props":1628,"children":1629},{},[1630],{"type":49,"tag":74,"props":1631,"children":1633},{"className":1632},[],[1634],{"type":55,"value":1635},"inference.checkpoint",{"type":49,"tag":443,"props":1637,"children":1638},{},[1639],{"type":49,"tag":74,"props":1640,"children":1642},{"className":1641},[],[1643],{"type":55,"value":1520},{"type":49,"tag":443,"props":1645,"children":1646},{},[1647,1648,1653],{"type":55,"value":1525},{"type":49,"tag":74,"props":1649,"children":1651},{"className":1650},[],[1652],{"type":55,"value":1215},{"type":55,"value":1532},{"type":49,"tag":406,"props":1655,"children":1656},{},[1657,1661,1669,1677],{"type":49,"tag":443,"props":1658,"children":1659},{},[1660],{"type":55,"value":331},{"type":49,"tag":443,"props":1662,"children":1663},{},[1664],{"type":49,"tag":74,"props":1665,"children":1667},{"className":1666},[],[1668],{"type":55,"value":1578},{"type":49,"tag":443,"props":1670,"children":1671},{},[1672],{"type":49,"tag":74,"props":1673,"children":1675},{"className":1674},[],[1676],{"type":55,"value":1587},{"type":49,"tag":443,"props":1678,"children":1679},{},[1680],{"type":55,"value":1592},{"type":49,"tag":406,"props":1682,"children":1683},{},[1684,1688,1696,1704],{"type":49,"tag":443,"props":1685,"children":1686},{},[1687],{"type":55,"value":474},{"type":49,"tag":443,"props":1689,"children":1690},{},[1691],{"type":49,"tag":74,"props":1692,"children":1694},{"className":1693},[],[1695],{"type":55,"value":1481},{"type":49,"tag":443,"props":1697,"children":1698},{},[1699],{"type":49,"tag":74,"props":1700,"children":1702},{"className":1701},[],[1703],{"type":55,"value":1490},{"type":49,"tag":443,"props":1705,"children":1706},{},[1707],{"type":55,"value":1495},{"type":49,"tag":406,"props":1709,"children":1710},{},[1711,1715,1723,1731],{"type":49,"tag":443,"props":1712,"children":1713},{},[1714],{"type":55,"value":474},{"type":49,"tag":443,"props":1716,"children":1717},{},[1718],{"type":49,"tag":74,"props":1719,"children":1721},{"className":1720},[],[1722],{"type":55,"value":1578},{"type":49,"tag":443,"props":1724,"children":1725},{},[1726],{"type":49,"tag":74,"props":1727,"children":1729},{"className":1728},[],[1730],{"type":55,"value":1587},{"type":49,"tag":443,"props":1732,"children":1733},{},[1734],{"type":55,"value":1592},{"type":49,"tag":406,"props":1736,"children":1737},{},[1738,1742,1751,1760],{"type":49,"tag":443,"props":1739,"children":1740},{},[1741],{"type":55,"value":474},{"type":49,"tag":443,"props":1743,"children":1744},{},[1745],{"type":49,"tag":74,"props":1746,"children":1748},{"className":1747},[],[1749],{"type":55,"value":1750},"train.pretrained_model_path",{"type":49,"tag":443,"props":1752,"children":1753},{},[1754],{"type":49,"tag":74,"props":1755,"children":1757},{"className":1756},[],[1758],{"type":55,"value":1759},"ptm_if_no_resume_model",{"type":49,"tag":443,"props":1761,"children":1762},{},[1763],{"type":55,"value":1764},"PTM when no resume checkpoint exists",{"type":49,"tag":406,"props":1766,"children":1767},{},[1768,1772,1780,1789],{"type":49,"tag":443,"props":1769,"children":1770},{},[1771],{"type":55,"value":474},{"type":49,"tag":443,"props":1773,"children":1774},{},[1775],{"type":49,"tag":74,"props":1776,"children":1778},{"className":1777},[],[1779],{"type":55,"value":1264},{"type":49,"tag":443,"props":1781,"children":1782},{},[1783],{"type":49,"tag":74,"props":1784,"children":1786},{"className":1785},[],[1787],{"type":55,"value":1788},"resume_model",{"type":49,"tag":443,"props":1790,"children":1791},{},[1792,1794,1799],{"type":55,"value":1793},"selected full ",{"type":49,"tag":74,"props":1795,"children":1797},{"className":1796},[],[1798],{"type":55,"value":1256},{"type":55,"value":1800}," training checkpoint from the current job results folder",{"type":49,"tag":58,"props":1802,"children":1803},{},[1804,1806,1811,1812,1818,1820,1826,1828,1833],{"type":55,"value":1805},"For ",{"type":49,"tag":74,"props":1807,"children":1809},{"className":1808},[],[1810],{"type":55,"value":1520},{"type":55,"value":1291},{"type":49,"tag":74,"props":1813,"children":1815},{"className":1814},[],[1816],{"type":55,"value":1817},"parent_model_folder",{"type":55,"value":1819},", pass the upstream train\u002Fexport\u002FAutoML child job id as ",{"type":49,"tag":74,"props":1821,"children":1823},{"className":1822},[],[1824],{"type":55,"value":1825},"parent_job_id",{"type":55,"value":1827},". The SDK lists the parent result folder, filters checkpoint artifacts, and returns the selected model file or folder. Do not add these mappings back to ",{"type":49,"tag":74,"props":1829,"children":1831},{"className":1830},[],[1832],{"type":55,"value":1411},{"type":55,"value":1834}," and do not patch generated runner scripts to guess checkpoint paths.",{"type":49,"tag":107,"props":1836,"children":1838},{"id":1837},"deployment",[1839],{"type":55,"value":1840},"Deployment",{"type":49,"tag":355,"props":1842,"children":1843},{},[1844],{"type":49,"tag":359,"props":1845,"children":1846},{},[1847],{"type":49,"tag":1848,"props":1849,"children":1850},"a",{"href":87},[1851],{"type":55,"value":1852},"tao-deploy-nvdinov2",{"type":49,"tag":1854,"props":1855,"children":1856},"style",{},[1857],{"type":55,"value":1858},"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":1860,"total":2014},[1861,1879,1894,1905,1917,1931,1944,1956,1969,1980,1994,2003],{"slug":1862,"name":1862,"fn":1863,"description":1864,"org":1865,"tags":1866,"stars":1876,"repoUrl":1877,"updatedAt":1878},"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},[1867,1870,1873],{"name":1868,"slug":1869,"type":15},"Documentation","documentation",{"name":1871,"slug":1872,"type":15},"MCP","mcp",{"name":1874,"slug":1875,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1880,"name":1880,"fn":1881,"description":1882,"org":1883,"tags":1884,"stars":1891,"repoUrl":1892,"updatedAt":1893},"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},[1885,1888,1889],{"name":1886,"slug":1887,"type":15},"Containers","containers",{"name":1840,"slug":1837,"type":15},{"name":1890,"slug":525,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1895,"name":1895,"fn":1896,"description":1897,"org":1898,"tags":1899,"stars":1891,"repoUrl":1892,"updatedAt":1904},"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},[1900,1903],{"name":1901,"slug":1902,"type":15},"CI\u002FCD","ci-cd",{"name":1840,"slug":1837,"type":15},"2026-07-14T05:25:59.97109",{"slug":1906,"name":1906,"fn":1907,"description":1908,"org":1909,"tags":1910,"stars":1891,"repoUrl":1892,"updatedAt":1916},"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},[1911,1912,1913],{"name":1901,"slug":1902,"type":15},{"name":1840,"slug":1837,"type":15},{"name":1914,"slug":1915,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1918,"name":1918,"fn":1919,"description":1920,"org":1921,"tags":1922,"stars":1891,"repoUrl":1892,"updatedAt":1930},"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},[1923,1926,1927],{"name":1924,"slug":1925,"type":15},"Debugging","debugging",{"name":1914,"slug":1915,"type":15},{"name":1928,"slug":1929,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1932,"name":1932,"fn":1933,"description":1934,"org":1935,"tags":1936,"stars":1891,"repoUrl":1892,"updatedAt":1943},"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},[1937,1940],{"name":1938,"slug":1939,"type":15},"Best Practices","best-practices",{"name":1941,"slug":1942,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1945,"name":1945,"fn":1946,"description":1947,"org":1948,"tags":1949,"stars":1891,"repoUrl":1892,"updatedAt":1955},"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},[1950,1951,1954],{"name":17,"slug":18,"type":15},{"name":1952,"slug":1953,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1957,"name":1957,"fn":1958,"description":1959,"org":1960,"tags":1961,"stars":1891,"repoUrl":1892,"updatedAt":1968},"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},[1962,1965],{"name":1963,"slug":1964,"type":15},"QA","qa",{"name":1966,"slug":1967,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":1970,"name":1970,"fn":1971,"description":1972,"org":1973,"tags":1974,"stars":1891,"repoUrl":1892,"updatedAt":1979},"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},[1975,1976],{"name":1840,"slug":1837,"type":15},{"name":1977,"slug":1978,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":1981,"name":1981,"fn":1982,"description":1983,"org":1984,"tags":1985,"stars":1891,"repoUrl":1892,"updatedAt":1993},"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},[1986,1989,1990],{"name":1987,"slug":1988,"type":15},"Code Review","code-review",{"name":1914,"slug":1915,"type":15},{"name":1991,"slug":1992,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1995,"name":1995,"fn":1996,"description":1997,"org":1998,"tags":1999,"stars":1891,"repoUrl":1892,"updatedAt":2002},"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},[2000,2001],{"name":1963,"slug":1964,"type":15},{"name":1966,"slug":1967,"type":15},"2026-07-14T05:25:54.928983",{"slug":2004,"name":2004,"fn":2005,"description":2006,"org":2007,"tags":2008,"stars":1891,"repoUrl":1892,"updatedAt":2013},"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},[2009,2012],{"name":2010,"slug":2011,"type":15},"Automation","automation",{"name":1901,"slug":1902,"type":15},"2026-07-30T05:29:03.275638",496,{"items":2016,"total":2112},[2017,2034,2044,2058,2068,2083,2098],{"slug":2018,"name":2018,"fn":2019,"description":2020,"org":2021,"tags":2022,"stars":23,"repoUrl":24,"updatedAt":2033},"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},[2023,2026,2029,2030],{"name":2024,"slug":2025,"type":15},"Data Analysis","data-analysis",{"name":2027,"slug":2028,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2031,"slug":2032,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2035,"name":2035,"fn":2036,"description":2037,"org":2038,"tags":2039,"stars":23,"repoUrl":24,"updatedAt":2043},"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},[2040,2041,2042],{"name":1840,"slug":1837,"type":15},{"name":1977,"slug":1978,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2045,"name":2045,"fn":2046,"description":2047,"org":2048,"tags":2049,"stars":23,"repoUrl":24,"updatedAt":2057},"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},[2050,2053,2054],{"name":2051,"slug":2052,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2055,"slug":2056,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2059,"name":2059,"fn":2060,"description":2061,"org":2062,"tags":2063,"stars":23,"repoUrl":24,"updatedAt":2067},"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},[2064,2065,2066],{"name":2024,"slug":2025,"type":15},{"name":9,"slug":8,"type":15},{"name":1966,"slug":1967,"type":15},"2026-07-17T05:29:03.913266",{"slug":2069,"name":2069,"fn":2070,"description":2071,"org":2072,"tags":2073,"stars":23,"repoUrl":24,"updatedAt":2082},"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},[2074,2075,2078,2079],{"name":2010,"slug":2011,"type":15},{"name":2076,"slug":2077,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":2080,"slug":2081,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2084,"name":2084,"fn":2085,"description":2086,"org":2087,"tags":2088,"stars":23,"repoUrl":24,"updatedAt":2097},"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},[2089,2090,2093,2094],{"name":1840,"slug":1837,"type":15},{"name":2091,"slug":2092,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2095,"slug":2096,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2099,"name":2099,"fn":2100,"description":2101,"org":2102,"tags":2103,"stars":23,"repoUrl":24,"updatedAt":2111},"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},[2104,2105,2108],{"name":9,"slug":8,"type":15},{"name":2106,"slug":2107,"type":15},"Quantum Computing","quantum-computing",{"name":2109,"slug":2110,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]