[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-mask-auto-label":3,"mdc--gbfb1j-key":31,"related-org-nvidia-tao-train-mask-auto-label":1691,"related-repo-nvidia-tao-train-mask-auto-label":1849},{"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-mask-auto-label","train segmentation models with Mask Auto-Label","MAL (Mask Auto-Label) for weakly-supervised segmentation. Produces segmentation masks from minimal annotations (point or box annotations) using a ViT-MAE backbone. Use when training, evaluating, or running inference for a TAO MAL model. Trigger phrases include \"train MAL\", \"Mask Auto-Label\", \"weakly-supervised segmentation\", \"box-prompted segmentation\", \"minimal-annotation mask prediction\".",{"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},"Machine Learning","machine-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","2026-07-14T05:27:01.367353","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-mask-auto-label","---\nname: tao-train-mask-auto-label\ndescription: MAL (Mask Auto-Label) for weakly-supervised segmentation. Produces segmentation masks from minimal annotations\n  (point or box annotations) using a ViT-MAE backbone. Use when training, evaluating, or running inference for a TAO MAL\n  model. Trigger phrases include \"train MAL\", \"Mask Auto-Label\", \"weakly-supervised segmentation\", \"box-prompted\n  segmentation\", \"minimal-annotation mask prediction\".\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- segmentation\n---\n\n# MAL\n\nMAL (Mask Auto-Label) for weakly-supervised segmentation. Produces segmentation masks from minimal annotations (e.g., point or box annotations). Uses ViT-MAE backbone.\n\nSet train.pretrained_model_path for ViT-MAE pretrained weights.\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\n- **Dataset type:** segmentation\n- **Formats:** default\n- **Monitoring metric:** mIoU\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| evaluate | dataset.val_img_dir | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.val_ann_path | eval_dataset | annotations.json | No |\n| inference | inference.img_dir | inference_dataset | images.tar.gz | No |\n| inference | inference.ann_path | inference_dataset | annotations.json | No |\n| train | dataset.train_img_dir | train_datasets | images.tar.gz | No |\n| train | dataset.train_ann_path | train_datasets | annotations.json | No |\n| train | dataset.val_img_dir | eval_dataset | images.tar.gz | No |\n| train | dataset.val_ann_path | eval_dataset | annotations.json | No |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action** — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`.\nMAL expects COCO-style annotation JSON plus image paths that match the JSON\n`file_name` entries after the data source is prepared. Archive-only CSV\u002Fimage\ndatasets are not compatible unless they are converted to this format first.\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.gpu_ids\": [\n        0\n    ],\n    \"train.num_epochs\": 5,\n    \"train.checkpoint_interval\": 5,\n    \"train.validation_interval\": 5,\n    \"dataset.train_img_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.train_ann_path\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.val_img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val_ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n```\n\n**evaluate (mandatory data sources):**\n```python\n{\n    \"evaluate.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"dataset.val_img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val_ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n```\n\n**inference (mandatory data sources):**\n```python\n{\n    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"inference.img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"inference.ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n```\n\nFor checkpoint-dependent actions, use the model resolver declared in\n`references\u002Fskill_info.yaml`. Select the exact epoch\u002Fstep checkpoint requested\nby the user or the best checkpoint when a best-checkpoint action is requested.\nThe `mal_model_latest.pth` symlink is only appropriate when the user explicitly\nasks for the latest checkpoint.\n\n## Eval Dataset\n\nOptional. Val images and annotations configured alongside train paths.\n\n## Important Parameters\n\n- **model.arch**: ViT-MAE backbone variant. Default vit-mae-base\u002F16.\n  Avoid `vit-deit-tiny\u002F16`; the current runtime rejects tiny ViT variants.\n- **train.lr**: Learning rate. Default 1e-6 (very low — fine-tuning ViT).\n- **dataset.crop_size**: Training crop size. Default 512. Use this key, not\n  `model.crop_size`.\n- **train.warmup_epochs**: Warmup epochs before full learning rate.\n- **model.load_mask**: Whether to load pre-computed masks.\n\n## AutoML \u002F HPO Notes\n\nFor MAL AutoML launches, keep the default smoke search space narrow and pass\n`automl_hyperparameters=[\"train.lr\", \"train.wd\"]`. Use conservative Bayesian\nranges around the ViT-MAE fine-tuning defaults, for example\n`train.lr` from `1e-7` to `1e-5` and `train.wd` from `1e-5` to `1e-2`.\nThe packaged train schema marks these two parameters as the default AutoML\nparameters; pass them explicitly when using a runtime that still derives MAL\nsearch metadata from its bundled config module.\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- Multi-GPU strategy: `ddp_find_unused_parameters_true`\n- No fsdp support\n- **LR auto-scaling:** `lr = lr * num_devices * batch_size` (learning rate is scaled automatically by device count and batch size)\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 1 GPU(s), recommended 2 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. ViT-MAE backbone at crop_size=512 needs 24GB+ GPU memory.\n\n## Error Patterns\n\n**CUDA out of memory**: Reduce `dataset.crop_size` (512 -> 384 -> 256) or use a smaller ViT-MAE variant (base vs large).\n\n**Key `crop_size` not in `MALModelConfig`**: The crop-size override was placed\nunder `model.crop_size`. Move it to `dataset.crop_size`.\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\nInference mappings from TAO Core `mal.config.json`:\n\n| Action | Spec Field | Inference Function | Meaning |\n|---|---|---|---|\n| evaluate | `evaluate.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `results_dir` | `output_dir` | current job results directory |\n| inference | `inference.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `inference.label_dump_path` | `create_inference_result_file_mal` | MAL inference JSON path |\n| inference | `results_dir` | `output_dir` | current job results directory |\n| train | `train.pretrained_model_path` | `ptm_if_no_resume_model` | optional pretrained model when not resuming |\n| train | `train.resume_training_checkpoint_path` | `resume_model` | exact checkpoint for resume runs |\n| train | `results_dir` | `output_dir` | current job results directory |\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",{"data":32,"body":40},{"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],"segmentation",{"type":41,"children":42},"root",[43,52,58,63,70,164,170,279,313,319,355,362,602,608,636,667,675,797,805,848,856,901,921,927,932,938,1007,1013,1069,1075,1092,1186,1223,1268,1274,1279,1285,1302,1339,1345,1374,1387,1650,1685],{"type":44,"tag":45,"props":46,"children":48},"element","h1",{"id":47},"mal",[49],{"type":50,"value":51},"text","MAL",{"type":44,"tag":53,"props":54,"children":55},"p",{},[56],{"type":50,"value":57},"MAL (Mask Auto-Label) for weakly-supervised segmentation. Produces segmentation masks from minimal annotations (e.g., point or box annotations). Uses ViT-MAE backbone.",{"type":44,"tag":53,"props":59,"children":60},{},[61],{"type":50,"value":62},"Set train.pretrained_model_path for ViT-MAE pretrained weights.",{"type":44,"tag":64,"props":65,"children":67},"h2",{"id":66},"dataclass-schemas",[68],{"type":50,"value":69},"Dataclass Schemas",{"type":44,"tag":53,"props":71,"children":72},{},[73,75,82,84,90,92,98,100,106,108,114,116,122,124,130,132,138,140,146,148,154,156,162],{"type":50,"value":74},"Generated TAO Core schemas are packaged in ",{"type":44,"tag":76,"props":77,"children":79},"code",{"className":78},[],[80],{"type":50,"value":81},"schemas\u002F\u003Caction>.schema.json",{"type":50,"value":83},", with ",{"type":44,"tag":76,"props":85,"children":87},{"className":86},[],[88],{"type":50,"value":89},"schemas\u002Fmanifest.json",{"type":50,"value":91}," listing available actions. Each generated schema also emits ",{"type":44,"tag":76,"props":93,"children":95},{"className":94},[],[96],{"type":50,"value":97},"references\u002Fspec_template_\u003Caction>.yaml",{"type":50,"value":99}," from the schema top-level ",{"type":44,"tag":76,"props":101,"children":103},{"className":102},[],[104],{"type":50,"value":105},"default",{"type":50,"value":107}," field. AutoML enablement is declared at the model layer in ",{"type":44,"tag":76,"props":109,"children":111},{"className":110},[],[112],{"type":50,"value":113},"references\u002Fskill_info.yaml",{"type":50,"value":115}," via ",{"type":44,"tag":76,"props":117,"children":119},{"className":118},[],[120],{"type":50,"value":121},"automl_enabled",{"type":50,"value":123},". Runnable AutoML still requires ",{"type":44,"tag":76,"props":125,"children":127},{"className":126},[],[128],{"type":50,"value":129},"schemas\u002Ftrain.schema.json",{"type":50,"value":131}," and ",{"type":44,"tag":76,"props":133,"children":135},{"className":134},[],[136],{"type":50,"value":137},"references\u002Fspec_template_train.yaml",{"type":50,"value":139}," to exist and parse. Use the packaged train schema for ",{"type":44,"tag":76,"props":141,"children":143},{"className":142},[],[144],{"type":50,"value":145},"automl_default_parameters",{"type":50,"value":147},", ",{"type":44,"tag":76,"props":149,"children":151},{"className":150},[],[152],{"type":50,"value":153},"automl_disabled_parameters",{"type":50,"value":155},", defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ",{"type":44,"tag":76,"props":157,"children":159},{"className":158},[],[160],{"type":50,"value":161},"~\u002Ftao-core",{"type":50,"value":163}," at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.",{"type":44,"tag":64,"props":165,"children":167},{"id":166},"train-action-policy",[168],{"type":50,"value":169},"Train Action Policy",{"type":44,"tag":53,"props":171,"children":172},{},[173,175,180,182,188,190,196,198,204,206,212,214,220,222,227,228,234,236,241,242,247,249,255,257,263,265,270,272,277],{"type":50,"value":174},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":44,"tag":76,"props":176,"children":178},{"className":177},[],[179],{"type":50,"value":113},{"type":50,"value":181}," and resolve the run override from either an explicit ",{"type":44,"tag":76,"props":183,"children":185},{"className":184},[],[186],{"type":50,"value":187},"automl_policy",{"type":50,"value":189}," value or the user's workflow request. Use ",{"type":44,"tag":76,"props":191,"children":193},{"className":192},[],[194],{"type":50,"value":195},"automl_policy: on",{"type":50,"value":197}," by default and only expose ",{"type":44,"tag":76,"props":199,"children":201},{"className":200},[],[202],{"type":50,"value":203},"on",{"type":50,"value":205}," \u002F ",{"type":44,"tag":76,"props":207,"children":209},{"className":208},[],[210],{"type":50,"value":211},"off",{"type":50,"value":213}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":44,"tag":76,"props":215,"children":217},{"className":216},[],[218],{"type":50,"value":219},"automl_policy: off",{"type":50,"value":221}," for this run only. When ",{"type":44,"tag":76,"props":223,"children":225},{"className":224},[],[226],{"type":50,"value":195},{"type":50,"value":147},{"type":44,"tag":76,"props":229,"children":231},{"className":230},[],[232],{"type":50,"value":233},"automl_enabled: true",{"type":50,"value":235},", and both ",{"type":44,"tag":76,"props":237,"children":239},{"className":238},[],[240],{"type":50,"value":129},{"type":50,"value":131},{"type":44,"tag":76,"props":243,"children":245},{"className":244},[],[246],{"type":50,"value":137},{"type":50,"value":248}," are packaged, route the train action through ",{"type":44,"tag":76,"props":250,"children":252},{"className":251},[],[253],{"type":50,"value":254},"tao-skill-bank:tao-run-automl",{"type":50,"value":256}," by default with this model's ",{"type":44,"tag":76,"props":258,"children":260},{"className":259},[],[261],{"type":50,"value":262},"skill_dir",{"type":50,"value":264},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":44,"tag":76,"props":266,"children":268},{"className":267},[],[269],{"type":50,"value":187},{"type":50,"value":271},". Use direct model training only when ",{"type":44,"tag":76,"props":273,"children":275},{"className":274},[],[276],{"type":50,"value":219},{"type":50,"value":278}," 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":44,"tag":53,"props":280,"children":281},{},[282,284,290,291,297,298,304,306,311],{"type":50,"value":283},"Non-train actions such as ",{"type":44,"tag":76,"props":285,"children":287},{"className":286},[],[288],{"type":50,"value":289},"evaluate",{"type":50,"value":147},{"type":44,"tag":76,"props":292,"children":294},{"className":293},[],[295],{"type":50,"value":296},"inference",{"type":50,"value":147},{"type":44,"tag":76,"props":299,"children":301},{"className":300},[],[302],{"type":50,"value":303},"export",{"type":50,"value":305},", and deploy flows stay in this model skill. The per-run ",{"type":44,"tag":76,"props":307,"children":309},{"className":308},[],[310],{"type":50,"value":187},{"type":50,"value":312}," override does not change model metadata.",{"type":44,"tag":64,"props":314,"children":316},{"id":315},"training-requirements",[317],{"type":50,"value":318},"Training Requirements",{"type":44,"tag":320,"props":321,"children":322},"ul",{},[323,335,345],{"type":44,"tag":324,"props":325,"children":326},"li",{},[327,333],{"type":44,"tag":328,"props":329,"children":330},"strong",{},[331],{"type":50,"value":332},"Dataset type:",{"type":50,"value":334}," segmentation",{"type":44,"tag":324,"props":336,"children":337},{},[338,343],{"type":44,"tag":328,"props":339,"children":340},{},[341],{"type":50,"value":342},"Formats:",{"type":50,"value":344}," default",{"type":44,"tag":324,"props":346,"children":347},{},[348,353],{"type":44,"tag":328,"props":349,"children":350},{},[351],{"type":50,"value":352},"Monitoring metric:",{"type":50,"value":354}," mIoU",{"type":44,"tag":356,"props":357,"children":359},"h3",{"id":358},"per-action-dataset-requirements",[360],{"type":50,"value":361},"Per-Action Dataset Requirements",{"type":44,"tag":363,"props":364,"children":365},"table",{},[366,400],{"type":44,"tag":367,"props":368,"children":369},"thead",{},[370],{"type":44,"tag":371,"props":372,"children":373},"tr",{},[374,380,385,390,395],{"type":44,"tag":375,"props":376,"children":377},"th",{},[378],{"type":50,"value":379},"Action",{"type":44,"tag":375,"props":381,"children":382},{},[383],{"type":50,"value":384},"Spec Key",{"type":44,"tag":375,"props":386,"children":387},{},[388],{"type":50,"value":389},"Source",{"type":44,"tag":375,"props":391,"children":392},{},[393],{"type":50,"value":394},"Files",{"type":44,"tag":375,"props":396,"children":397},{},[398],{"type":50,"value":399},"List?",{"type":44,"tag":401,"props":402,"children":403},"tbody",{},[404,432,457,482,506,532,556,579],{"type":44,"tag":371,"props":405,"children":406},{},[407,412,417,422,427],{"type":44,"tag":408,"props":409,"children":410},"td",{},[411],{"type":50,"value":289},{"type":44,"tag":408,"props":413,"children":414},{},[415],{"type":50,"value":416},"dataset.val_img_dir",{"type":44,"tag":408,"props":418,"children":419},{},[420],{"type":50,"value":421},"eval_dataset",{"type":44,"tag":408,"props":423,"children":424},{},[425],{"type":50,"value":426},"images.tar.gz",{"type":44,"tag":408,"props":428,"children":429},{},[430],{"type":50,"value":431},"No",{"type":44,"tag":371,"props":433,"children":434},{},[435,439,444,448,453],{"type":44,"tag":408,"props":436,"children":437},{},[438],{"type":50,"value":289},{"type":44,"tag":408,"props":440,"children":441},{},[442],{"type":50,"value":443},"dataset.val_ann_path",{"type":44,"tag":408,"props":445,"children":446},{},[447],{"type":50,"value":421},{"type":44,"tag":408,"props":449,"children":450},{},[451],{"type":50,"value":452},"annotations.json",{"type":44,"tag":408,"props":454,"children":455},{},[456],{"type":50,"value":431},{"type":44,"tag":371,"props":458,"children":459},{},[460,464,469,474,478],{"type":44,"tag":408,"props":461,"children":462},{},[463],{"type":50,"value":296},{"type":44,"tag":408,"props":465,"children":466},{},[467],{"type":50,"value":468},"inference.img_dir",{"type":44,"tag":408,"props":470,"children":471},{},[472],{"type":50,"value":473},"inference_dataset",{"type":44,"tag":408,"props":475,"children":476},{},[477],{"type":50,"value":426},{"type":44,"tag":408,"props":479,"children":480},{},[481],{"type":50,"value":431},{"type":44,"tag":371,"props":483,"children":484},{},[485,489,494,498,502],{"type":44,"tag":408,"props":486,"children":487},{},[488],{"type":50,"value":296},{"type":44,"tag":408,"props":490,"children":491},{},[492],{"type":50,"value":493},"inference.ann_path",{"type":44,"tag":408,"props":495,"children":496},{},[497],{"type":50,"value":473},{"type":44,"tag":408,"props":499,"children":500},{},[501],{"type":50,"value":452},{"type":44,"tag":408,"props":503,"children":504},{},[505],{"type":50,"value":431},{"type":44,"tag":371,"props":507,"children":508},{},[509,514,519,524,528],{"type":44,"tag":408,"props":510,"children":511},{},[512],{"type":50,"value":513},"train",{"type":44,"tag":408,"props":515,"children":516},{},[517],{"type":50,"value":518},"dataset.train_img_dir",{"type":44,"tag":408,"props":520,"children":521},{},[522],{"type":50,"value":523},"train_datasets",{"type":44,"tag":408,"props":525,"children":526},{},[527],{"type":50,"value":426},{"type":44,"tag":408,"props":529,"children":530},{},[531],{"type":50,"value":431},{"type":44,"tag":371,"props":533,"children":534},{},[535,539,544,548,552],{"type":44,"tag":408,"props":536,"children":537},{},[538],{"type":50,"value":513},{"type":44,"tag":408,"props":540,"children":541},{},[542],{"type":50,"value":543},"dataset.train_ann_path",{"type":44,"tag":408,"props":545,"children":546},{},[547],{"type":50,"value":523},{"type":44,"tag":408,"props":549,"children":550},{},[551],{"type":50,"value":452},{"type":44,"tag":408,"props":553,"children":554},{},[555],{"type":50,"value":431},{"type":44,"tag":371,"props":557,"children":558},{},[559,563,567,571,575],{"type":44,"tag":408,"props":560,"children":561},{},[562],{"type":50,"value":513},{"type":44,"tag":408,"props":564,"children":565},{},[566],{"type":50,"value":416},{"type":44,"tag":408,"props":568,"children":569},{},[570],{"type":50,"value":421},{"type":44,"tag":408,"props":572,"children":573},{},[574],{"type":50,"value":426},{"type":44,"tag":408,"props":576,"children":577},{},[578],{"type":50,"value":431},{"type":44,"tag":371,"props":580,"children":581},{},[582,586,590,594,598],{"type":44,"tag":408,"props":583,"children":584},{},[585],{"type":50,"value":513},{"type":44,"tag":408,"props":587,"children":588},{},[589],{"type":50,"value":443},{"type":44,"tag":408,"props":591,"children":592},{},[593],{"type":50,"value":421},{"type":44,"tag":408,"props":595,"children":596},{},[597],{"type":50,"value":452},{"type":44,"tag":408,"props":599,"children":600},{},[601],{"type":50,"value":431},{"type":44,"tag":356,"props":603,"children":605},{"id":604},"typical-spec-overrides",[606],{"type":50,"value":607},"Typical Spec Overrides",{"type":44,"tag":53,"props":609,"children":610},{},[611,613,618,620,626,628,634],{"type":50,"value":612},"Data source overrides are ",{"type":44,"tag":328,"props":614,"children":615},{},[616],{"type":50,"value":617},"mandatory for every action",{"type":50,"value":619}," — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in ",{"type":44,"tag":76,"props":621,"children":623},{"className":622},[],[624],{"type":50,"value":625},"spec_overrides",{"type":50,"value":627},".\nMAL expects COCO-style annotation JSON plus image paths that match the JSON\n",{"type":44,"tag":76,"props":629,"children":631},{"className":630},[],[632],{"type":50,"value":633},"file_name",{"type":50,"value":635}," entries after the data source is prepared. Archive-only CSV\u002Fimage\ndatasets are not compatible unless they are converted to this format first.",{"type":44,"tag":637,"props":638,"children":643},"pre",{"className":639,"code":640,"language":641,"meta":642,"style":642},"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","",[644],{"type":44,"tag":76,"props":645,"children":646},{"__ignoreMap":642},[647,658],{"type":44,"tag":648,"props":649,"children":652},"span",{"class":650,"line":651},"line",1,[653],{"type":44,"tag":648,"props":654,"children":655},{},[656],{"type":50,"value":657},"S3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\n",{"type":44,"tag":648,"props":659,"children":661},{"class":650,"line":660},2,[662],{"type":44,"tag":648,"props":663,"children":664},{},[665],{"type":50,"value":666},"S3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n",{"type":44,"tag":53,"props":668,"children":669},{},[670],{"type":44,"tag":328,"props":671,"children":672},{},[673],{"type":50,"value":674},"train (mandatory data sources):",{"type":44,"tag":637,"props":676,"children":678},{"className":639,"code":677,"language":641,"meta":642,"style":642},"{\n    \"train.num_gpus\": 1,\n    \"train.gpu_ids\": [\n        0\n    ],\n    \"train.num_epochs\": 5,\n    \"train.checkpoint_interval\": 5,\n    \"train.validation_interval\": 5,\n    \"dataset.train_img_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n    \"dataset.train_ann_path\": f\"{S3_TRAIN}\u002Fannotations.json\",\n    \"dataset.val_img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val_ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n",[679],{"type":44,"tag":76,"props":680,"children":681},{"__ignoreMap":642},[682,690,698,707,716,725,734,743,752,761,770,779,788],{"type":44,"tag":648,"props":683,"children":684},{"class":650,"line":651},[685],{"type":44,"tag":648,"props":686,"children":687},{},[688],{"type":50,"value":689},"{\n",{"type":44,"tag":648,"props":691,"children":692},{"class":650,"line":660},[693],{"type":44,"tag":648,"props":694,"children":695},{},[696],{"type":50,"value":697},"    \"train.num_gpus\": 1,\n",{"type":44,"tag":648,"props":699,"children":701},{"class":650,"line":700},3,[702],{"type":44,"tag":648,"props":703,"children":704},{},[705],{"type":50,"value":706},"    \"train.gpu_ids\": [\n",{"type":44,"tag":648,"props":708,"children":710},{"class":650,"line":709},4,[711],{"type":44,"tag":648,"props":712,"children":713},{},[714],{"type":50,"value":715},"        0\n",{"type":44,"tag":648,"props":717,"children":719},{"class":650,"line":718},5,[720],{"type":44,"tag":648,"props":721,"children":722},{},[723],{"type":50,"value":724},"    ],\n",{"type":44,"tag":648,"props":726,"children":728},{"class":650,"line":727},6,[729],{"type":44,"tag":648,"props":730,"children":731},{},[732],{"type":50,"value":733},"    \"train.num_epochs\": 5,\n",{"type":44,"tag":648,"props":735,"children":737},{"class":650,"line":736},7,[738],{"type":44,"tag":648,"props":739,"children":740},{},[741],{"type":50,"value":742},"    \"train.checkpoint_interval\": 5,\n",{"type":44,"tag":648,"props":744,"children":746},{"class":650,"line":745},8,[747],{"type":44,"tag":648,"props":748,"children":749},{},[750],{"type":50,"value":751},"    \"train.validation_interval\": 5,\n",{"type":44,"tag":648,"props":753,"children":755},{"class":650,"line":754},9,[756],{"type":44,"tag":648,"props":757,"children":758},{},[759],{"type":50,"value":760},"    \"dataset.train_img_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\",\n",{"type":44,"tag":648,"props":762,"children":764},{"class":650,"line":763},10,[765],{"type":44,"tag":648,"props":766,"children":767},{},[768],{"type":50,"value":769},"    \"dataset.train_ann_path\": f\"{S3_TRAIN}\u002Fannotations.json\",\n",{"type":44,"tag":648,"props":771,"children":773},{"class":650,"line":772},11,[774],{"type":44,"tag":648,"props":775,"children":776},{},[777],{"type":50,"value":778},"    \"dataset.val_img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n",{"type":44,"tag":648,"props":780,"children":782},{"class":650,"line":781},12,[783],{"type":44,"tag":648,"props":784,"children":785},{},[786],{"type":50,"value":787},"    \"dataset.val_ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n",{"type":44,"tag":648,"props":789,"children":791},{"class":650,"line":790},13,[792],{"type":44,"tag":648,"props":793,"children":794},{},[795],{"type":50,"value":796},"}\n",{"type":44,"tag":53,"props":798,"children":799},{},[800],{"type":44,"tag":328,"props":801,"children":802},{},[803],{"type":50,"value":804},"evaluate (mandatory data sources):",{"type":44,"tag":637,"props":806,"children":808},{"className":639,"code":807,"language":641,"meta":642,"style":642},"{\n    \"evaluate.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"dataset.val_img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.val_ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n",[809],{"type":44,"tag":76,"props":810,"children":811},{"__ignoreMap":642},[812,819,827,834,841],{"type":44,"tag":648,"props":813,"children":814},{"class":650,"line":651},[815],{"type":44,"tag":648,"props":816,"children":817},{},[818],{"type":50,"value":689},{"type":44,"tag":648,"props":820,"children":821},{"class":650,"line":660},[822],{"type":44,"tag":648,"props":823,"children":824},{},[825],{"type":50,"value":826},"    \"evaluate.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n",{"type":44,"tag":648,"props":828,"children":829},{"class":650,"line":700},[830],{"type":44,"tag":648,"props":831,"children":832},{},[833],{"type":50,"value":778},{"type":44,"tag":648,"props":835,"children":836},{"class":650,"line":709},[837],{"type":44,"tag":648,"props":838,"children":839},{},[840],{"type":50,"value":787},{"type":44,"tag":648,"props":842,"children":843},{"class":650,"line":718},[844],{"type":44,"tag":648,"props":845,"children":846},{},[847],{"type":50,"value":796},{"type":44,"tag":53,"props":849,"children":850},{},[851],{"type":44,"tag":328,"props":852,"children":853},{},[854],{"type":50,"value":855},"inference (mandatory data sources):",{"type":44,"tag":637,"props":857,"children":859},{"className":639,"code":858,"language":641,"meta":642,"style":642},"{\n    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n    \"inference.img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"inference.ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n",[860],{"type":44,"tag":76,"props":861,"children":862},{"__ignoreMap":642},[863,870,878,886,894],{"type":44,"tag":648,"props":864,"children":865},{"class":650,"line":651},[866],{"type":44,"tag":648,"props":867,"children":868},{},[869],{"type":50,"value":689},{"type":44,"tag":648,"props":871,"children":872},{"class":650,"line":660},[873],{"type":44,"tag":648,"props":874,"children":875},{},[876],{"type":50,"value":877},"    \"inference.checkpoint\": \"\u003Cselected train\u002FAutoML checkpoint>\",\n",{"type":44,"tag":648,"props":879,"children":880},{"class":650,"line":700},[881],{"type":44,"tag":648,"props":882,"children":883},{},[884],{"type":50,"value":885},"    \"inference.img_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n",{"type":44,"tag":648,"props":887,"children":888},{"class":650,"line":709},[889],{"type":44,"tag":648,"props":890,"children":891},{},[892],{"type":50,"value":893},"    \"inference.ann_path\": f\"{S3_EVAL}\u002Fannotations.json\",\n",{"type":44,"tag":648,"props":895,"children":896},{"class":650,"line":718},[897],{"type":44,"tag":648,"props":898,"children":899},{},[900],{"type":50,"value":796},{"type":44,"tag":53,"props":902,"children":903},{},[904,906,911,913,919],{"type":50,"value":905},"For checkpoint-dependent actions, use the model resolver declared in\n",{"type":44,"tag":76,"props":907,"children":909},{"className":908},[],[910],{"type":50,"value":113},{"type":50,"value":912},". Select the exact epoch\u002Fstep checkpoint requested\nby the user or the best checkpoint when a best-checkpoint action is requested.\nThe ",{"type":44,"tag":76,"props":914,"children":916},{"className":915},[],[917],{"type":50,"value":918},"mal_model_latest.pth",{"type":50,"value":920}," symlink is only appropriate when the user explicitly\nasks for the latest checkpoint.",{"type":44,"tag":64,"props":922,"children":924},{"id":923},"eval-dataset",[925],{"type":50,"value":926},"Eval Dataset",{"type":44,"tag":53,"props":928,"children":929},{},[930],{"type":50,"value":931},"Optional. Val images and annotations configured alongside train paths.",{"type":44,"tag":64,"props":933,"children":935},{"id":934},"important-parameters",[936],{"type":50,"value":937},"Important Parameters",{"type":44,"tag":320,"props":939,"children":940},{},[941,959,969,987,997],{"type":44,"tag":324,"props":942,"children":943},{},[944,949,951,957],{"type":44,"tag":328,"props":945,"children":946},{},[947],{"type":50,"value":948},"model.arch",{"type":50,"value":950},": ViT-MAE backbone variant. Default vit-mae-base\u002F16.\nAvoid ",{"type":44,"tag":76,"props":952,"children":954},{"className":953},[],[955],{"type":50,"value":956},"vit-deit-tiny\u002F16",{"type":50,"value":958},"; the current runtime rejects tiny ViT variants.",{"type":44,"tag":324,"props":960,"children":961},{},[962,967],{"type":44,"tag":328,"props":963,"children":964},{},[965],{"type":50,"value":966},"train.lr",{"type":50,"value":968},": Learning rate. Default 1e-6 (very low — fine-tuning ViT).",{"type":44,"tag":324,"props":970,"children":971},{},[972,977,979,985],{"type":44,"tag":328,"props":973,"children":974},{},[975],{"type":50,"value":976},"dataset.crop_size",{"type":50,"value":978},": Training crop size. Default 512. Use this key, not\n",{"type":44,"tag":76,"props":980,"children":982},{"className":981},[],[983],{"type":50,"value":984},"model.crop_size",{"type":50,"value":986},".",{"type":44,"tag":324,"props":988,"children":989},{},[990,995],{"type":44,"tag":328,"props":991,"children":992},{},[993],{"type":50,"value":994},"train.warmup_epochs",{"type":50,"value":996},": Warmup epochs before full learning rate.",{"type":44,"tag":324,"props":998,"children":999},{},[1000,1005],{"type":44,"tag":328,"props":1001,"children":1002},{},[1003],{"type":50,"value":1004},"model.load_mask",{"type":50,"value":1006},": Whether to load pre-computed masks.",{"type":44,"tag":64,"props":1008,"children":1010},{"id":1009},"automl-hpo-notes",[1011],{"type":50,"value":1012},"AutoML \u002F HPO Notes",{"type":44,"tag":53,"props":1014,"children":1015},{},[1016,1018,1024,1026,1031,1033,1039,1041,1047,1048,1054,1055,1060,1061,1067],{"type":50,"value":1017},"For MAL AutoML launches, keep the default smoke search space narrow and pass\n",{"type":44,"tag":76,"props":1019,"children":1021},{"className":1020},[],[1022],{"type":50,"value":1023},"automl_hyperparameters=[\"train.lr\", \"train.wd\"]",{"type":50,"value":1025},". Use conservative Bayesian\nranges around the ViT-MAE fine-tuning defaults, for example\n",{"type":44,"tag":76,"props":1027,"children":1029},{"className":1028},[],[1030],{"type":50,"value":966},{"type":50,"value":1032}," from ",{"type":44,"tag":76,"props":1034,"children":1036},{"className":1035},[],[1037],{"type":50,"value":1038},"1e-7",{"type":50,"value":1040}," to ",{"type":44,"tag":76,"props":1042,"children":1044},{"className":1043},[],[1045],{"type":50,"value":1046},"1e-5",{"type":50,"value":131},{"type":44,"tag":76,"props":1049,"children":1051},{"className":1050},[],[1052],{"type":50,"value":1053},"train.wd",{"type":50,"value":1032},{"type":44,"tag":76,"props":1056,"children":1058},{"className":1057},[],[1059],{"type":50,"value":1046},{"type":50,"value":1040},{"type":44,"tag":76,"props":1062,"children":1064},{"className":1063},[],[1065],{"type":50,"value":1066},"1e-2",{"type":50,"value":1068},".\nThe packaged train schema marks these two parameters as the default AutoML\nparameters; pass them explicitly when using a runtime that still derives MAL\nsearch metadata from its bundled config module.",{"type":44,"tag":64,"props":1070,"children":1072},{"id":1071},"multi-gpu-multi-node",[1073],{"type":50,"value":1074},"Multi-GPU \u002F Multi-Node",{"type":44,"tag":53,"props":1076,"children":1077},{},[1078,1083,1085,1090],{"type":44,"tag":328,"props":1079,"children":1080},{},[1081],{"type":50,"value":1082},"Launch method:",{"type":50,"value":1084}," Lightning-managed (single ",{"type":44,"tag":76,"props":1086,"children":1088},{"className":1087},[],[1089],{"type":50,"value":641},{"type":50,"value":1091}," process, Lightning spawns workers).",{"type":44,"tag":363,"props":1093,"children":1094},{},[1095,1115],{"type":44,"tag":367,"props":1096,"children":1097},{},[1098],{"type":44,"tag":371,"props":1099,"children":1100},{},[1101,1105,1110],{"type":44,"tag":375,"props":1102,"children":1103},{},[1104],{"type":50,"value":384},{"type":44,"tag":375,"props":1106,"children":1107},{},[1108],{"type":50,"value":1109},"Description",{"type":44,"tag":375,"props":1111,"children":1112},{},[1113],{"type":50,"value":1114},"Default",{"type":44,"tag":401,"props":1116,"children":1117},{},[1118,1140,1165],{"type":44,"tag":371,"props":1119,"children":1120},{},[1121,1130,1135],{"type":44,"tag":408,"props":1122,"children":1123},{},[1124],{"type":44,"tag":76,"props":1125,"children":1127},{"className":1126},[],[1128],{"type":50,"value":1129},"train.num_gpus",{"type":44,"tag":408,"props":1131,"children":1132},{},[1133],{"type":50,"value":1134},"Number of GPUs",{"type":44,"tag":408,"props":1136,"children":1137},{},[1138],{"type":50,"value":1139},"1",{"type":44,"tag":371,"props":1141,"children":1142},{},[1143,1152,1157],{"type":44,"tag":408,"props":1144,"children":1145},{},[1146],{"type":44,"tag":76,"props":1147,"children":1149},{"className":1148},[],[1150],{"type":50,"value":1151},"train.gpu_ids",{"type":44,"tag":408,"props":1153,"children":1154},{},[1155],{"type":50,"value":1156},"GPU device indices",{"type":44,"tag":408,"props":1158,"children":1159},{},[1160],{"type":44,"tag":648,"props":1161,"children":1162},{},[1163],{"type":50,"value":1164},"0",{"type":44,"tag":371,"props":1166,"children":1167},{},[1168,1177,1182],{"type":44,"tag":408,"props":1169,"children":1170},{},[1171],{"type":44,"tag":76,"props":1172,"children":1174},{"className":1173},[],[1175],{"type":50,"value":1176},"train.num_nodes",{"type":44,"tag":408,"props":1178,"children":1179},{},[1180],{"type":50,"value":1181},"Number of nodes",{"type":44,"tag":408,"props":1183,"children":1184},{},[1185],{"type":50,"value":1139},{"type":44,"tag":320,"props":1187,"children":1188},{},[1189,1200,1205],{"type":44,"tag":324,"props":1190,"children":1191},{},[1192,1194],{"type":50,"value":1193},"Multi-GPU strategy: ",{"type":44,"tag":76,"props":1195,"children":1197},{"className":1196},[],[1198],{"type":50,"value":1199},"ddp_find_unused_parameters_true",{"type":44,"tag":324,"props":1201,"children":1202},{},[1203],{"type":50,"value":1204},"No fsdp support",{"type":44,"tag":324,"props":1206,"children":1207},{},[1208,1213,1215,1221],{"type":44,"tag":328,"props":1209,"children":1210},{},[1211],{"type":50,"value":1212},"LR auto-scaling:",{"type":50,"value":1214}," ",{"type":44,"tag":76,"props":1216,"children":1218},{"className":1217},[],[1219],{"type":50,"value":1220},"lr = lr * num_devices * batch_size",{"type":50,"value":1222}," (learning rate is scaled automatically by device count and batch size)",{"type":44,"tag":53,"props":1224,"children":1225},{},[1226,1231,1233,1239,1240,1246,1247,1253,1254,1260,1261,1267],{"type":44,"tag":328,"props":1227,"children":1228},{},[1229],{"type":50,"value":1230},"Multi-node env vars",{"type":50,"value":1232}," (set by orchestrator): ",{"type":44,"tag":76,"props":1234,"children":1236},{"className":1235},[],[1237],{"type":50,"value":1238},"WORLD_SIZE",{"type":50,"value":147},{"type":44,"tag":76,"props":1241,"children":1243},{"className":1242},[],[1244],{"type":50,"value":1245},"NODE_RANK",{"type":50,"value":147},{"type":44,"tag":76,"props":1248,"children":1250},{"className":1249},[],[1251],{"type":50,"value":1252},"MASTER_ADDR",{"type":50,"value":147},{"type":44,"tag":76,"props":1255,"children":1257},{"className":1256},[],[1258],{"type":50,"value":1259},"MASTER_PORT",{"type":50,"value":147},{"type":44,"tag":76,"props":1262,"children":1264},{"className":1263},[],[1265],{"type":50,"value":1266},"NUM_GPU_PER_NODE",{"type":50,"value":986},{"type":44,"tag":64,"props":1269,"children":1271},{"id":1270},"hardware",[1272],{"type":50,"value":1273},"Hardware",{"type":44,"tag":53,"props":1275,"children":1276},{},[1277],{"type":50,"value":1278},"Minimum 1 GPU(s), recommended 2 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. ViT-MAE backbone at crop_size=512 needs 24GB+ GPU memory.",{"type":44,"tag":64,"props":1280,"children":1282},{"id":1281},"error-patterns",[1283],{"type":50,"value":1284},"Error Patterns",{"type":44,"tag":53,"props":1286,"children":1287},{},[1288,1293,1295,1300],{"type":44,"tag":328,"props":1289,"children":1290},{},[1291],{"type":50,"value":1292},"CUDA out of memory",{"type":50,"value":1294},": Reduce ",{"type":44,"tag":76,"props":1296,"children":1298},{"className":1297},[],[1299],{"type":50,"value":976},{"type":50,"value":1301}," (512 -> 384 -> 256) or use a smaller ViT-MAE variant (base vs large).",{"type":44,"tag":53,"props":1303,"children":1304},{},[1305,1324,1326,1331,1333,1338],{"type":44,"tag":328,"props":1306,"children":1307},{},[1308,1310,1316,1318],{"type":50,"value":1309},"Key ",{"type":44,"tag":76,"props":1311,"children":1313},{"className":1312},[],[1314],{"type":50,"value":1315},"crop_size",{"type":50,"value":1317}," not in ",{"type":44,"tag":76,"props":1319,"children":1321},{"className":1320},[],[1322],{"type":50,"value":1323},"MALModelConfig",{"type":50,"value":1325},": The crop-size override was placed\nunder ",{"type":44,"tag":76,"props":1327,"children":1329},{"className":1328},[],[1330],{"type":50,"value":984},{"type":50,"value":1332},". Move it to ",{"type":44,"tag":76,"props":1334,"children":1336},{"className":1335},[],[1337],{"type":50,"value":976},{"type":50,"value":986},{"type":44,"tag":64,"props":1340,"children":1342},{"id":1341},"spec-param-parent-model-inference",[1343],{"type":50,"value":1344},"Spec Param \u002F Parent Model Inference",{"type":44,"tag":53,"props":1346,"children":1347},{},[1348,1350,1356,1358,1364,1366,1372],{"type":50,"value":1349},"Model-specific inference mappings belong in this MD file, not in ",{"type":44,"tag":76,"props":1351,"children":1353},{"className":1352},[],[1354],{"type":50,"value":1355},"config.json",{"type":50,"value":1357},". Generated runners should read this section and apply the mappings with SDK helpers before ",{"type":44,"tag":76,"props":1359,"children":1361},{"className":1360},[],[1362],{"type":50,"value":1363},"create_job()",{"type":50,"value":1365},". This mirrors the old microservices ",{"type":44,"tag":76,"props":1367,"children":1369},{"className":1368},[],[1370],{"type":50,"value":1371},"infer_params.py",{"type":50,"value":1373}," flow.",{"type":44,"tag":53,"props":1375,"children":1376},{},[1377,1379,1385],{"type":50,"value":1378},"Inference mappings from TAO Core ",{"type":44,"tag":76,"props":1380,"children":1382},{"className":1381},[],[1383],{"type":50,"value":1384},"mal.config.json",{"type":50,"value":1386},":",{"type":44,"tag":363,"props":1388,"children":1389},{},[1390,1415],{"type":44,"tag":367,"props":1391,"children":1392},{},[1393],{"type":44,"tag":371,"props":1394,"children":1395},{},[1396,1400,1405,1410],{"type":44,"tag":375,"props":1397,"children":1398},{},[1399],{"type":50,"value":379},{"type":44,"tag":375,"props":1401,"children":1402},{},[1403],{"type":50,"value":1404},"Spec Field",{"type":44,"tag":375,"props":1406,"children":1407},{},[1408],{"type":50,"value":1409},"Inference Function",{"type":44,"tag":375,"props":1411,"children":1412},{},[1413],{"type":50,"value":1414},"Meaning",{"type":44,"tag":401,"props":1416,"children":1417},{},[1418,1448,1478,1506,1536,1563,1593,1623],{"type":44,"tag":371,"props":1419,"children":1420},{},[1421,1425,1434,1443],{"type":44,"tag":408,"props":1422,"children":1423},{},[1424],{"type":50,"value":289},{"type":44,"tag":408,"props":1426,"children":1427},{},[1428],{"type":44,"tag":76,"props":1429,"children":1431},{"className":1430},[],[1432],{"type":50,"value":1433},"evaluate.checkpoint",{"type":44,"tag":408,"props":1435,"children":1436},{},[1437],{"type":44,"tag":76,"props":1438,"children":1440},{"className":1439},[],[1441],{"type":50,"value":1442},"parent_model",{"type":44,"tag":408,"props":1444,"children":1445},{},[1446],{"type":50,"value":1447},"model file inferred from the parent job results folder",{"type":44,"tag":371,"props":1449,"children":1450},{},[1451,1455,1464,1473],{"type":44,"tag":408,"props":1452,"children":1453},{},[1454],{"type":50,"value":289},{"type":44,"tag":408,"props":1456,"children":1457},{},[1458],{"type":44,"tag":76,"props":1459,"children":1461},{"className":1460},[],[1462],{"type":50,"value":1463},"results_dir",{"type":44,"tag":408,"props":1465,"children":1466},{},[1467],{"type":44,"tag":76,"props":1468,"children":1470},{"className":1469},[],[1471],{"type":50,"value":1472},"output_dir",{"type":44,"tag":408,"props":1474,"children":1475},{},[1476],{"type":50,"value":1477},"current job results directory",{"type":44,"tag":371,"props":1479,"children":1480},{},[1481,1485,1494,1502],{"type":44,"tag":408,"props":1482,"children":1483},{},[1484],{"type":50,"value":296},{"type":44,"tag":408,"props":1486,"children":1487},{},[1488],{"type":44,"tag":76,"props":1489,"children":1491},{"className":1490},[],[1492],{"type":50,"value":1493},"inference.checkpoint",{"type":44,"tag":408,"props":1495,"children":1496},{},[1497],{"type":44,"tag":76,"props":1498,"children":1500},{"className":1499},[],[1501],{"type":50,"value":1442},{"type":44,"tag":408,"props":1503,"children":1504},{},[1505],{"type":50,"value":1447},{"type":44,"tag":371,"props":1507,"children":1508},{},[1509,1513,1522,1531],{"type":44,"tag":408,"props":1510,"children":1511},{},[1512],{"type":50,"value":296},{"type":44,"tag":408,"props":1514,"children":1515},{},[1516],{"type":44,"tag":76,"props":1517,"children":1519},{"className":1518},[],[1520],{"type":50,"value":1521},"inference.label_dump_path",{"type":44,"tag":408,"props":1523,"children":1524},{},[1525],{"type":44,"tag":76,"props":1526,"children":1528},{"className":1527},[],[1529],{"type":50,"value":1530},"create_inference_result_file_mal",{"type":44,"tag":408,"props":1532,"children":1533},{},[1534],{"type":50,"value":1535},"MAL inference JSON path",{"type":44,"tag":371,"props":1537,"children":1538},{},[1539,1543,1551,1559],{"type":44,"tag":408,"props":1540,"children":1541},{},[1542],{"type":50,"value":296},{"type":44,"tag":408,"props":1544,"children":1545},{},[1546],{"type":44,"tag":76,"props":1547,"children":1549},{"className":1548},[],[1550],{"type":50,"value":1463},{"type":44,"tag":408,"props":1552,"children":1553},{},[1554],{"type":44,"tag":76,"props":1555,"children":1557},{"className":1556},[],[1558],{"type":50,"value":1472},{"type":44,"tag":408,"props":1560,"children":1561},{},[1562],{"type":50,"value":1477},{"type":44,"tag":371,"props":1564,"children":1565},{},[1566,1570,1579,1588],{"type":44,"tag":408,"props":1567,"children":1568},{},[1569],{"type":50,"value":513},{"type":44,"tag":408,"props":1571,"children":1572},{},[1573],{"type":44,"tag":76,"props":1574,"children":1576},{"className":1575},[],[1577],{"type":50,"value":1578},"train.pretrained_model_path",{"type":44,"tag":408,"props":1580,"children":1581},{},[1582],{"type":44,"tag":76,"props":1583,"children":1585},{"className":1584},[],[1586],{"type":50,"value":1587},"ptm_if_no_resume_model",{"type":44,"tag":408,"props":1589,"children":1590},{},[1591],{"type":50,"value":1592},"optional pretrained model when not resuming",{"type":44,"tag":371,"props":1594,"children":1595},{},[1596,1600,1609,1618],{"type":44,"tag":408,"props":1597,"children":1598},{},[1599],{"type":50,"value":513},{"type":44,"tag":408,"props":1601,"children":1602},{},[1603],{"type":44,"tag":76,"props":1604,"children":1606},{"className":1605},[],[1607],{"type":50,"value":1608},"train.resume_training_checkpoint_path",{"type":44,"tag":408,"props":1610,"children":1611},{},[1612],{"type":44,"tag":76,"props":1613,"children":1615},{"className":1614},[],[1616],{"type":50,"value":1617},"resume_model",{"type":44,"tag":408,"props":1619,"children":1620},{},[1621],{"type":50,"value":1622},"exact checkpoint for resume runs",{"type":44,"tag":371,"props":1624,"children":1625},{},[1626,1630,1638,1646],{"type":44,"tag":408,"props":1627,"children":1628},{},[1629],{"type":50,"value":513},{"type":44,"tag":408,"props":1631,"children":1632},{},[1633],{"type":44,"tag":76,"props":1634,"children":1636},{"className":1635},[],[1637],{"type":50,"value":1463},{"type":44,"tag":408,"props":1639,"children":1640},{},[1641],{"type":44,"tag":76,"props":1642,"children":1644},{"className":1643},[],[1645],{"type":50,"value":1472},{"type":44,"tag":408,"props":1647,"children":1648},{},[1649],{"type":50,"value":1477},{"type":44,"tag":53,"props":1651,"children":1652},{},[1653,1655,1660,1662,1668,1670,1676,1678,1683],{"type":50,"value":1654},"For ",{"type":44,"tag":76,"props":1656,"children":1658},{"className":1657},[],[1659],{"type":50,"value":1442},{"type":50,"value":1661}," or ",{"type":44,"tag":76,"props":1663,"children":1665},{"className":1664},[],[1666],{"type":50,"value":1667},"parent_model_folder",{"type":50,"value":1669},", pass the upstream train\u002Fexport\u002FAutoML child job id as ",{"type":44,"tag":76,"props":1671,"children":1673},{"className":1672},[],[1674],{"type":50,"value":1675},"parent_job_id",{"type":50,"value":1677},". 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":44,"tag":76,"props":1679,"children":1681},{"className":1680},[],[1682],{"type":50,"value":1355},{"type":50,"value":1684}," and do not patch generated runner scripts to guess checkpoint paths.",{"type":44,"tag":1686,"props":1687,"children":1688},"style",{},[1689],{"type":50,"value":1690},"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":1692,"total":1848},[1693,1711,1728,1739,1751,1765,1778,1790,1803,1814,1828,1837],{"slug":1694,"name":1694,"fn":1695,"description":1696,"org":1697,"tags":1698,"stars":1708,"repoUrl":1709,"updatedAt":1710},"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},[1699,1702,1705],{"name":1700,"slug":1701,"type":15},"Documentation","documentation",{"name":1703,"slug":1704,"type":15},"MCP","mcp",{"name":1706,"slug":1707,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1712,"name":1712,"fn":1713,"description":1714,"org":1715,"tags":1716,"stars":1725,"repoUrl":1726,"updatedAt":1727},"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},[1717,1720,1723],{"name":1718,"slug":1719,"type":15},"Containers","containers",{"name":1721,"slug":1722,"type":15},"Deployment","deployment",{"name":1724,"slug":641,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1729,"name":1729,"fn":1730,"description":1731,"org":1732,"tags":1733,"stars":1725,"repoUrl":1726,"updatedAt":1738},"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},[1734,1737],{"name":1735,"slug":1736,"type":15},"CI\u002FCD","ci-cd",{"name":1721,"slug":1722,"type":15},"2026-07-14T05:25:59.97109",{"slug":1740,"name":1740,"fn":1741,"description":1742,"org":1743,"tags":1744,"stars":1725,"repoUrl":1726,"updatedAt":1750},"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},[1745,1746,1747],{"name":1735,"slug":1736,"type":15},{"name":1721,"slug":1722,"type":15},{"name":1748,"slug":1749,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1752,"name":1752,"fn":1753,"description":1754,"org":1755,"tags":1756,"stars":1725,"repoUrl":1726,"updatedAt":1764},"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},[1757,1760,1761],{"name":1758,"slug":1759,"type":15},"Debugging","debugging",{"name":1748,"slug":1749,"type":15},{"name":1762,"slug":1763,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1766,"name":1766,"fn":1767,"description":1768,"org":1769,"tags":1770,"stars":1725,"repoUrl":1726,"updatedAt":1777},"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},[1771,1774],{"name":1772,"slug":1773,"type":15},"Best Practices","best-practices",{"name":1775,"slug":1776,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1779,"name":1779,"fn":1780,"description":1781,"org":1782,"tags":1783,"stars":1725,"repoUrl":1726,"updatedAt":1789},"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},[1784,1785,1788],{"name":13,"slug":14,"type":15},{"name":1786,"slug":1787,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1791,"name":1791,"fn":1792,"description":1793,"org":1794,"tags":1795,"stars":1725,"repoUrl":1726,"updatedAt":1802},"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},[1796,1799],{"name":1797,"slug":1798,"type":15},"QA","qa",{"name":1800,"slug":1801,"type":15},"Testing","testing","2026-07-14T05:25:53.673039",{"slug":1804,"name":1804,"fn":1805,"description":1806,"org":1807,"tags":1808,"stars":1725,"repoUrl":1726,"updatedAt":1813},"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},[1809,1810],{"name":1721,"slug":1722,"type":15},{"name":1811,"slug":1812,"type":15},"Infrastructure","infrastructure","2026-07-14T05:25:49.362534",{"slug":1815,"name":1815,"fn":1816,"description":1817,"org":1818,"tags":1819,"stars":1725,"repoUrl":1726,"updatedAt":1827},"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},[1820,1823,1824],{"name":1821,"slug":1822,"type":15},"Code Review","code-review",{"name":1748,"slug":1749,"type":15},{"name":1825,"slug":1826,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1829,"name":1829,"fn":1830,"description":1831,"org":1832,"tags":1833,"stars":1725,"repoUrl":1726,"updatedAt":1836},"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},[1834,1835],{"name":1797,"slug":1798,"type":15},{"name":1800,"slug":1801,"type":15},"2026-07-14T05:25:54.928983",{"slug":1838,"name":1838,"fn":1839,"description":1840,"org":1841,"tags":1842,"stars":1725,"repoUrl":1726,"updatedAt":1847},"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},[1843,1846],{"name":1844,"slug":1845,"type":15},"Automation","automation",{"name":1735,"slug":1736,"type":15},"2026-07-30T05:29:03.275638",496,{"items":1850,"total":1946},[1851,1868,1878,1892,1902,1917,1932],{"slug":1852,"name":1852,"fn":1853,"description":1854,"org":1855,"tags":1856,"stars":20,"repoUrl":21,"updatedAt":1867},"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},[1857,1860,1863,1864],{"name":1858,"slug":1859,"type":15},"Data Analysis","data-analysis",{"name":1861,"slug":1862,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1865,"slug":1866,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1869,"name":1869,"fn":1870,"description":1871,"org":1872,"tags":1873,"stars":20,"repoUrl":21,"updatedAt":1877},"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},[1874,1875,1876],{"name":1721,"slug":1722,"type":15},{"name":1811,"slug":1812,"type":15},{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1879,"name":1879,"fn":1880,"description":1881,"org":1882,"tags":1883,"stars":20,"repoUrl":21,"updatedAt":1891},"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},[1884,1887,1888],{"name":1885,"slug":1886,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1889,"slug":1890,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1893,"name":1893,"fn":1894,"description":1895,"org":1896,"tags":1897,"stars":20,"repoUrl":21,"updatedAt":1901},"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},[1898,1899,1900],{"name":1858,"slug":1859,"type":15},{"name":9,"slug":8,"type":15},{"name":1800,"slug":1801,"type":15},"2026-07-17T05:29:03.913266",{"slug":1903,"name":1903,"fn":1904,"description":1905,"org":1906,"tags":1907,"stars":20,"repoUrl":21,"updatedAt":1916},"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},[1908,1909,1912,1913],{"name":1844,"slug":1845,"type":15},{"name":1910,"slug":1911,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1914,"slug":1915,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1918,"name":1918,"fn":1919,"description":1920,"org":1921,"tags":1922,"stars":20,"repoUrl":21,"updatedAt":1931},"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},[1923,1924,1927,1928],{"name":1721,"slug":1722,"type":15},{"name":1925,"slug":1926,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1929,"slug":1930,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1933,"name":1933,"fn":1934,"description":1935,"org":1936,"tags":1937,"stars":20,"repoUrl":21,"updatedAt":1945},"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},[1938,1939,1942],{"name":9,"slug":8,"type":15},{"name":1940,"slug":1941,"type":15},"Quantum Computing","quantum-computing",{"name":1943,"slug":1944,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305]