[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-deformable-detr":3,"mdc--p1c9ip-key":34,"related-repo-nvidia-tao-train-deformable-detr":2564,"related-org-nvidia-tao-train-deformable-detr":2668},{"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-deformable-detr","train Deformable DETR object detection models","Deformable DETR for 2D object detection. Uses deformable attention for efficient multi-scale feature processing, lighter than DINO with competitive accuracy. Use when training, evaluating, exporting, quantizing, or running inference for a TAO Deformable-DETR model. Trigger phrases include \"train deformable-detr\", \"Deformable DETR object detection\", \"lightweight DETR detector\".",{"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","2026-07-14T05:27:46.552933","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-deformable-detr","---\nname: tao-train-deformable-detr\ndescription: Deformable DETR for 2D object detection. Uses deformable attention for efficient multi-scale feature processing,\n  lighter than DINO with competitive accuracy. Use when training, evaluating, exporting, quantizing, or running inference for\n  a TAO Deformable-DETR model. Trigger phrases include \"train deformable-detr\", \"Deformable DETR object detection\",\n  \"lightweight DETR detector\".\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit.\nmetadata:\n  version: \"0.1.0\"\n  author: NVIDIA Corporation\nallowed-tools: Read Bash\ntags:\n- object\n- detection\n---\n\n# Deformable DETR\n\nDeformable DETR for 2D object detection. Uses deformable attention for efficient multi-scale feature processing. Lighter than DINO with competitive accuracy.\n\nUses pretrained weights. Set `model.pretrained_backbone_path` for backbone-only\nloading or `train.pretrained_model_path` for full model initialization.\n\nSupported parent model actions are `train`, `evaluate`, `inference`, `export`, and `quantize`. The PyT model container does not support a native `gen_trt_engine` subtask for this network. The `gen_trt_engine` action declared in `references\u002Fskill_info.yaml` must run with the TAO Deploy container. 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 `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:** object_detection\n- **Formats:** coco, coco_raw\n- **Monitoring metric:** val_mAP50 for AP50; `val_mAP` for COCO\u002Fpaper-style benchmark comparisons.\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| evaluate | dataset.test_data_sources.image_dir | eval_dataset | images.tar.gz | No |\n| evaluate | dataset.test_data_sources.json_file | eval_dataset | annotations.json | No |\n| export | dataset.train_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | Yes |\n| export | dataset.val_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | Yes |\n| inference | dataset.infer_data_sources.image_dir | inference_dataset | images.tar.gz | Yes |\n| inference | dataset.infer_data_sources.classmap | inference_dataset | label_map.txt | No |\n| quantize | dataset.train_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | Yes |\n| quantize | dataset.val_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | Yes |\n| quantize | dataset.quant_calibration_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | No |\n| train | dataset.train_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | Yes |\n| train | dataset.val_data_sources | train_datasets | image_dir: images.tar.gz, json_file: annotations.json | Yes |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action** — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`.\n\n```python\nS3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\nS3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n```\n\n**train (mandatory data sources):**\n```python\n{\n    \"train.num_epochs\": 10,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"train.gpu_ids\": [0],\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n}\n```\n\n**evaluate (mandatory data sources):**\n```python\n{\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n    \"dataset.test_data_sources.image_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.test_data_sources.json_file\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n```\n\nIf the train or AutoML run changed architecture-affecting fields such as\n`model.enc_layers`, `model.dec_layers`, `model.num_queries`, or\n`model.num_select`, carry the same values into evaluate, export, inference, and\ndeploy actions with the selected checkpoint. In addition to the fields above,\ncarry `model.num_feature_levels`, `model.dim_feedforward`, input image\ndimensions, and dataset class metadata when they were changed. Loading a\ncheckpoint into the default architecture can fail with tensor shape mismatches,\nespecially when smoke-test runs shrink the transformer for speed.\n\n**export (mandatory data sources):**\n```python\n{\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n}\n```\n\n**TensorRT engine generation:**\n\nUse the deploy spec templates after `export`. Do not call `deformable_detr\ngen_trt_engine` from the parent PyT model container; that CLI advertises\n`convert`, `evaluate`, `export`, `inference`, `quantize`, `train`, and\n`default_specs`, but not `gen_trt_engine`. The model action metadata selects the\nTAO Deploy container for engine generation.\n\nDeploy engine generation needs the exported ONNX file as input and creates the\nengine at `gen_trt_engine.trt_engine`.\n\n```python\n{\n    \"gen_trt_engine.tensorrt.data_type\": \"FP16\",\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n}\n```\n\n**inference (mandatory data sources):**\n```python\n{\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.infer_data_sources.image_dir\": [f\"{S3_EVAL}\u002Fimages.tar.gz\"],\n    \"dataset.infer_data_sources.classmap\": f\"{S3_EVAL}\u002Flabel_map.txt\",\n}\n```\n\n**quantize (mandatory data sources):**\n```python\n{\n    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.quant_calibration_data_sources\": {\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"},\n}\n```\n## Eval Dataset\n\nOptional. If provided, validation mAP is computed at each checkpoint interval.\n\n## Checkpoint Handling\n\nTraining emits epoch-and-step checkpoints using the pattern\n`model_epoch_\u003Cepoch>_step_\u003Cstep>.pth`, plus a `dd_model_latest.pth` symlink. For\ndependent actions, use the model-specific or SDK-provided checkpoint resolver to\nselect the intended artifact. Evaluation, inference, export, and quantize should\nreceive the selected exact checkpoint path, not the `dd_model_latest.pth` symlink,\nunless the user explicitly asked for latest. Resume\u002Fretrain should set\n`train.resume_training_checkpoint_path` to the exact checkpoint being resumed\nfrom.\n\n## Important Parameters\n\n- **dataset.num_classes**: Number of object classes plus the background class. Default 91 (COCO). Must match annotations.\n- **dataset.eval_class_ids**: Foreground category ids to include in COCO metrics. Set this to every object category id in custom datasets; the template default evaluates class id 1 only.\n- **model.backbone**: Default resnet_50. Supported: resnet_50, gcvit_tiny, gcvit_small, gcvit_base, gcvit_large, gcvit_large_384 (more limited than DINO).\n- **train.optim.lr**: Learning rate. Default 2e-4 (AdamW). lr_backbone is 2e-5.\n- **train.optim.lr_steps**: MultiStep LR schedule. Default [40]. For short runs, set to match ~80% of total epochs.\n- **model.num_queries**: Number of object queries. Default 300. Valid range 100-900.\n- **model.dropout_ratio**: Dropout in transformer layers. Default 0.3 (higher than DINO's 0.0). Reduce for large datasets, increase for small datasets.\n- **model.dim_feedforward**: FFN hidden dim. Default 1024 (vs DINO's 2048). Increasing improves capacity but costs memory.\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| `train.distributed_strategy` | `ddp` or `fsdp` | `ddp` |\n\nSame DDP\u002FFSDP behavior as DINO. Multi-node requires `WORLD_SIZE`, `NODE_RANK`, `MASTER_ADDR`, `MASTER_PORT` env vars set by orchestrator.\n\nWhen increasing `train.num_gpus`, also set `train.gpu_ids` to the same visible\ndevice range. For example, an 8-GPU single-node Slurm run must include both\n`\"train.num_gpus\": 8` and `\"train.gpu_ids\": [0, 1, 2, 3, 4, 5, 6, 7]`.\n\n## Export \u002F TRT Defaults\n\n- Export input: 640x640, opset 17\n- TRT data types: FP32, FP16, INT8\n- TRT workspace: 1024 MB\n- TRT max_batch_size: 1\n\n## Hardware\n\nMinimum 1 GPU(s), recommended 4 GPU(s). 16GB+ (V100 or A100) VRAM per GPU. Slightly lighter than DINO due to smaller FFN. batch_size=4 fits on most 16GB+ GPUs.\n\n## Error Patterns\n\n**CUDA out of memory**: Reduce batch_size (4 -> 2 -> 1).\n\n**num_select must be \u003C num_queries * num_classes**: Same constraint as DINO.\n\n**return_interm_indices length must match num_feature_levels**: Default [1,2,3,4] with num_feature_levels=4.\n\n**Dataset size smaller than total batch size**: Reduce batch_size or num_gpus.\n\n**AutoML metric extraction**: Deformable DETR emits detection metrics in structured training status and logs. For COCO\u002Fpaper-style benchmark comparisons, optimize `val_mAP` with `direction: maximize`; for explicit AP50 workflows, optimize `val_mAP50`. Prefer `results_dir\u002Ftrain\u002Fstatus.json` or AutoML result state before parsing raw logs. Do not optimize `val_loss` for default detection model invocations.\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 `deformable_detr.config.json`:\n\n| Action | Spec Field | Inference Function | Meaning |\n|---|---|---|---|\n| evaluate | `encryption_key` | `key` | encryption key |\n| evaluate | `evaluate.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `evaluate.trt_engine` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `results_dir` | `output_dir` | current job results directory |\n| export | `encryption_key` | `key` | encryption key |\n| export | `export.checkpoint` | `parent_model` | model file inferred 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` | model file inferred from the parent job results folder |\n| inference | `inference.trt_engine` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `results_dir` | `output_dir` | current job results directory |\n| quantize | `encryption_key` | `key` | encryption key |\n| quantize | `quantize.model_path` | `parent_model` | model file inferred from the parent job results folder |\n| quantize | `results_dir` | `output_dir` | current job results directory |\n| train | `encryption_key` | `key` | encryption key |\n| train | `model.pretrained_backbone_path` | `ptm_if_no_resume_model` | PTM when no resume checkpoint exists |\n| train | `results_dir` | `output_dir` | current job results directory |\n| train | `train.pretrained_model_path` | `ptm_if_no_resume_model` | full model PTM when no resume checkpoint exists |\n| train | `train.resume_training_checkpoint_path` | `resume_model` | model file inferred 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-deformable-detr](references\u002Ftao-deploy-deformable-detr.md)\n",{"data":35,"body":44},{"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],"object","detection",{"type":45,"children":46},"root",[47,56,62,84,166,173,264,270,379,410,416,460,467,779,785,805,836,844,948,956,1007,1057,1065,1114,1122,1195,1207,1251,1259,1303,1311,1354,1360,1365,1371,1407,1413,1501,1507,1524,1655,1689,1722,1728,1751,1757,1762,1768,1778,1788,1805,1815,1864,1870,1899,1912,2505,2539,2545,2558],{"type":48,"tag":49,"props":50,"children":52},"element","h1",{"id":51},"deformable-detr",[53],{"type":54,"value":55},"text","Deformable DETR",{"type":48,"tag":57,"props":58,"children":59},"p",{},[60],{"type":54,"value":61},"Deformable DETR for 2D object detection. Uses deformable attention for efficient multi-scale feature processing. Lighter than DINO with competitive accuracy.",{"type":48,"tag":57,"props":63,"children":64},{},[65,67,74,76,82],{"type":54,"value":66},"Uses pretrained weights. Set ",{"type":48,"tag":68,"props":69,"children":71},"code",{"className":70},[],[72],{"type":54,"value":73},"model.pretrained_backbone_path",{"type":54,"value":75}," for backbone-only\nloading or ",{"type":48,"tag":68,"props":77,"children":79},{"className":78},[],[80],{"type":54,"value":81},"train.pretrained_model_path",{"type":54,"value":83}," for full model initialization.",{"type":48,"tag":57,"props":85,"children":86},{},[87,89,95,97,103,104,110,111,117,119,125,127,133,135,140,142,148,150,156,158,164],{"type":54,"value":88},"Supported parent model actions are ",{"type":48,"tag":68,"props":90,"children":92},{"className":91},[],[93],{"type":54,"value":94},"train",{"type":54,"value":96},", ",{"type":48,"tag":68,"props":98,"children":100},{"className":99},[],[101],{"type":54,"value":102},"evaluate",{"type":54,"value":96},{"type":48,"tag":68,"props":105,"children":107},{"className":106},[],[108],{"type":54,"value":109},"inference",{"type":54,"value":96},{"type":48,"tag":68,"props":112,"children":114},{"className":113},[],[115],{"type":54,"value":116},"export",{"type":54,"value":118},", and ",{"type":48,"tag":68,"props":120,"children":122},{"className":121},[],[123],{"type":54,"value":124},"quantize",{"type":54,"value":126},". The PyT model container does not support a native ",{"type":48,"tag":68,"props":128,"children":130},{"className":129},[],[131],{"type":54,"value":132},"gen_trt_engine",{"type":54,"value":134}," subtask for this network. The ",{"type":48,"tag":68,"props":136,"children":138},{"className":137},[],[139],{"type":54,"value":132},{"type":54,"value":141}," action declared in ",{"type":48,"tag":68,"props":143,"children":145},{"className":144},[],[146],{"type":54,"value":147},"references\u002Fskill_info.yaml",{"type":54,"value":149}," must run with the TAO Deploy container. Deploy spec templates live in this skill's ",{"type":48,"tag":68,"props":151,"children":153},{"className":152},[],[154],{"type":54,"value":155},"references\u002F",{"type":54,"value":157}," folder with the ",{"type":48,"tag":68,"props":159,"children":161},{"className":160},[],[162],{"type":54,"value":163},"spec_template_deploy_*.yaml",{"type":54,"value":165}," prefix.",{"type":48,"tag":167,"props":168,"children":170},"h2",{"id":169},"dataclass-schemas",[171],{"type":54,"value":172},"Dataclass Schemas",{"type":48,"tag":57,"props":174,"children":175},{},[176,178,184,186,192,194,200,202,208,210,215,217,223,225,231,233,239,241,247,248,254,256,262],{"type":54,"value":177},"Generated TAO Core schemas are packaged in ",{"type":48,"tag":68,"props":179,"children":181},{"className":180},[],[182],{"type":54,"value":183},"schemas\u002F\u003Caction>.schema.json",{"type":54,"value":185},", with ",{"type":48,"tag":68,"props":187,"children":189},{"className":188},[],[190],{"type":54,"value":191},"schemas\u002Fmanifest.json",{"type":54,"value":193}," listing available actions. Each generated schema also emits ",{"type":48,"tag":68,"props":195,"children":197},{"className":196},[],[198],{"type":54,"value":199},"references\u002Fspec_template_\u003Caction>.yaml",{"type":54,"value":201}," from the schema top-level ",{"type":48,"tag":68,"props":203,"children":205},{"className":204},[],[206],{"type":54,"value":207},"default",{"type":54,"value":209}," field. AutoML enablement is declared at the model layer in ",{"type":48,"tag":68,"props":211,"children":213},{"className":212},[],[214],{"type":54,"value":147},{"type":54,"value":216}," via ",{"type":48,"tag":68,"props":218,"children":220},{"className":219},[],[221],{"type":54,"value":222},"automl_enabled",{"type":54,"value":224},". Runnable AutoML still requires ",{"type":48,"tag":68,"props":226,"children":228},{"className":227},[],[229],{"type":54,"value":230},"schemas\u002Ftrain.schema.json",{"type":54,"value":232}," and ",{"type":48,"tag":68,"props":234,"children":236},{"className":235},[],[237],{"type":54,"value":238},"references\u002Fspec_template_train.yaml",{"type":54,"value":240}," to exist and parse. Use the packaged train schema for ",{"type":48,"tag":68,"props":242,"children":244},{"className":243},[],[245],{"type":54,"value":246},"automl_default_parameters",{"type":54,"value":96},{"type":48,"tag":68,"props":249,"children":251},{"className":250},[],[252],{"type":54,"value":253},"automl_disabled_parameters",{"type":54,"value":255},", defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ",{"type":48,"tag":68,"props":257,"children":259},{"className":258},[],[260],{"type":54,"value":261},"~\u002Ftao-core",{"type":54,"value":263}," at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.",{"type":48,"tag":167,"props":265,"children":267},{"id":266},"train-action-policy",[268],{"type":54,"value":269},"Train Action Policy",{"type":48,"tag":57,"props":271,"children":272},{},[273,275,280,282,288,290,296,298,304,306,312,314,320,322,327,328,334,336,341,342,347,349,355,357,363,365,370,372,377],{"type":54,"value":274},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":48,"tag":68,"props":276,"children":278},{"className":277},[],[279],{"type":54,"value":147},{"type":54,"value":281}," and resolve the run override from either an explicit ",{"type":48,"tag":68,"props":283,"children":285},{"className":284},[],[286],{"type":54,"value":287},"automl_policy",{"type":54,"value":289}," value or the user's workflow request. Use ",{"type":48,"tag":68,"props":291,"children":293},{"className":292},[],[294],{"type":54,"value":295},"automl_policy: on",{"type":54,"value":297}," by default and only expose ",{"type":48,"tag":68,"props":299,"children":301},{"className":300},[],[302],{"type":54,"value":303},"on",{"type":54,"value":305}," \u002F ",{"type":48,"tag":68,"props":307,"children":309},{"className":308},[],[310],{"type":54,"value":311},"off",{"type":54,"value":313}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":48,"tag":68,"props":315,"children":317},{"className":316},[],[318],{"type":54,"value":319},"automl_policy: off",{"type":54,"value":321}," for this run only. When ",{"type":48,"tag":68,"props":323,"children":325},{"className":324},[],[326],{"type":54,"value":295},{"type":54,"value":96},{"type":48,"tag":68,"props":329,"children":331},{"className":330},[],[332],{"type":54,"value":333},"automl_enabled: true",{"type":54,"value":335},", and both ",{"type":48,"tag":68,"props":337,"children":339},{"className":338},[],[340],{"type":54,"value":230},{"type":54,"value":232},{"type":48,"tag":68,"props":343,"children":345},{"className":344},[],[346],{"type":54,"value":238},{"type":54,"value":348}," are packaged, route the train action through ",{"type":48,"tag":68,"props":350,"children":352},{"className":351},[],[353],{"type":54,"value":354},"tao-skill-bank:tao-run-automl",{"type":54,"value":356}," by default with this model's ",{"type":48,"tag":68,"props":358,"children":360},{"className":359},[],[361],{"type":54,"value":362},"skill_dir",{"type":54,"value":364},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":48,"tag":68,"props":366,"children":368},{"className":367},[],[369],{"type":54,"value":287},{"type":54,"value":371},". Use direct model training only when ",{"type":48,"tag":68,"props":373,"children":375},{"className":374},[],[376],{"type":54,"value":319},{"type":54,"value":378}," 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":48,"tag":57,"props":380,"children":381},{},[382,384,389,390,395,396,401,403,408],{"type":54,"value":383},"Non-train actions such as ",{"type":48,"tag":68,"props":385,"children":387},{"className":386},[],[388],{"type":54,"value":102},{"type":54,"value":96},{"type":48,"tag":68,"props":391,"children":393},{"className":392},[],[394],{"type":54,"value":109},{"type":54,"value":96},{"type":48,"tag":68,"props":397,"children":399},{"className":398},[],[400],{"type":54,"value":116},{"type":54,"value":402},", and deploy flows stay in this model skill. The per-run ",{"type":48,"tag":68,"props":404,"children":406},{"className":405},[],[407],{"type":54,"value":287},{"type":54,"value":409}," override does not change model metadata.",{"type":48,"tag":167,"props":411,"children":413},{"id":412},"training-requirements",[414],{"type":54,"value":415},"Training Requirements",{"type":48,"tag":417,"props":418,"children":419},"ul",{},[420,432,442],{"type":48,"tag":421,"props":422,"children":423},"li",{},[424,430],{"type":48,"tag":425,"props":426,"children":427},"strong",{},[428],{"type":54,"value":429},"Dataset type:",{"type":54,"value":431}," object_detection",{"type":48,"tag":421,"props":433,"children":434},{},[435,440],{"type":48,"tag":425,"props":436,"children":437},{},[438],{"type":54,"value":439},"Formats:",{"type":54,"value":441}," coco, coco_raw",{"type":48,"tag":421,"props":443,"children":444},{},[445,450,452,458],{"type":48,"tag":425,"props":446,"children":447},{},[448],{"type":54,"value":449},"Monitoring metric:",{"type":54,"value":451}," val_mAP50 for AP50; ",{"type":48,"tag":68,"props":453,"children":455},{"className":454},[],[456],{"type":54,"value":457},"val_mAP",{"type":54,"value":459}," for COCO\u002Fpaper-style benchmark comparisons.",{"type":48,"tag":461,"props":462,"children":464},"h3",{"id":463},"per-action-dataset-requirements",[465],{"type":54,"value":466},"Per-Action Dataset Requirements",{"type":48,"tag":468,"props":469,"children":470},"table",{},[471,505],{"type":48,"tag":472,"props":473,"children":474},"thead",{},[475],{"type":48,"tag":476,"props":477,"children":478},"tr",{},[479,485,490,495,500],{"type":48,"tag":480,"props":481,"children":482},"th",{},[483],{"type":54,"value":484},"Action",{"type":48,"tag":480,"props":486,"children":487},{},[488],{"type":54,"value":489},"Spec Key",{"type":48,"tag":480,"props":491,"children":492},{},[493],{"type":54,"value":494},"Source",{"type":48,"tag":480,"props":496,"children":497},{},[498],{"type":54,"value":499},"Files",{"type":48,"tag":480,"props":501,"children":502},{},[503],{"type":54,"value":504},"List?",{"type":48,"tag":506,"props":507,"children":508},"tbody",{},[509,537,562,589,613,638,663,686,709,733,756],{"type":48,"tag":476,"props":510,"children":511},{},[512,517,522,527,532],{"type":48,"tag":513,"props":514,"children":515},"td",{},[516],{"type":54,"value":102},{"type":48,"tag":513,"props":518,"children":519},{},[520],{"type":54,"value":521},"dataset.test_data_sources.image_dir",{"type":48,"tag":513,"props":523,"children":524},{},[525],{"type":54,"value":526},"eval_dataset",{"type":48,"tag":513,"props":528,"children":529},{},[530],{"type":54,"value":531},"images.tar.gz",{"type":48,"tag":513,"props":533,"children":534},{},[535],{"type":54,"value":536},"No",{"type":48,"tag":476,"props":538,"children":539},{},[540,544,549,553,558],{"type":48,"tag":513,"props":541,"children":542},{},[543],{"type":54,"value":102},{"type":48,"tag":513,"props":545,"children":546},{},[547],{"type":54,"value":548},"dataset.test_data_sources.json_file",{"type":48,"tag":513,"props":550,"children":551},{},[552],{"type":54,"value":526},{"type":48,"tag":513,"props":554,"children":555},{},[556],{"type":54,"value":557},"annotations.json",{"type":48,"tag":513,"props":559,"children":560},{},[561],{"type":54,"value":536},{"type":48,"tag":476,"props":563,"children":564},{},[565,569,574,579,584],{"type":48,"tag":513,"props":566,"children":567},{},[568],{"type":54,"value":116},{"type":48,"tag":513,"props":570,"children":571},{},[572],{"type":54,"value":573},"dataset.train_data_sources",{"type":48,"tag":513,"props":575,"children":576},{},[577],{"type":54,"value":578},"train_datasets",{"type":48,"tag":513,"props":580,"children":581},{},[582],{"type":54,"value":583},"image_dir: images.tar.gz, json_file: annotations.json",{"type":48,"tag":513,"props":585,"children":586},{},[587],{"type":54,"value":588},"Yes",{"type":48,"tag":476,"props":590,"children":591},{},[592,596,601,605,609],{"type":48,"tag":513,"props":593,"children":594},{},[595],{"type":54,"value":116},{"type":48,"tag":513,"props":597,"children":598},{},[599],{"type":54,"value":600},"dataset.val_data_sources",{"type":48,"tag":513,"props":602,"children":603},{},[604],{"type":54,"value":578},{"type":48,"tag":513,"props":606,"children":607},{},[608],{"type":54,"value":583},{"type":48,"tag":513,"props":610,"children":611},{},[612],{"type":54,"value":588},{"type":48,"tag":476,"props":614,"children":615},{},[616,620,625,630,634],{"type":48,"tag":513,"props":617,"children":618},{},[619],{"type":54,"value":109},{"type":48,"tag":513,"props":621,"children":622},{},[623],{"type":54,"value":624},"dataset.infer_data_sources.image_dir",{"type":48,"tag":513,"props":626,"children":627},{},[628],{"type":54,"value":629},"inference_dataset",{"type":48,"tag":513,"props":631,"children":632},{},[633],{"type":54,"value":531},{"type":48,"tag":513,"props":635,"children":636},{},[637],{"type":54,"value":588},{"type":48,"tag":476,"props":639,"children":640},{},[641,645,650,654,659],{"type":48,"tag":513,"props":642,"children":643},{},[644],{"type":54,"value":109},{"type":48,"tag":513,"props":646,"children":647},{},[648],{"type":54,"value":649},"dataset.infer_data_sources.classmap",{"type":48,"tag":513,"props":651,"children":652},{},[653],{"type":54,"value":629},{"type":48,"tag":513,"props":655,"children":656},{},[657],{"type":54,"value":658},"label_map.txt",{"type":48,"tag":513,"props":660,"children":661},{},[662],{"type":54,"value":536},{"type":48,"tag":476,"props":664,"children":665},{},[666,670,674,678,682],{"type":48,"tag":513,"props":667,"children":668},{},[669],{"type":54,"value":124},{"type":48,"tag":513,"props":671,"children":672},{},[673],{"type":54,"value":573},{"type":48,"tag":513,"props":675,"children":676},{},[677],{"type":54,"value":578},{"type":48,"tag":513,"props":679,"children":680},{},[681],{"type":54,"value":583},{"type":48,"tag":513,"props":683,"children":684},{},[685],{"type":54,"value":588},{"type":48,"tag":476,"props":687,"children":688},{},[689,693,697,701,705],{"type":48,"tag":513,"props":690,"children":691},{},[692],{"type":54,"value":124},{"type":48,"tag":513,"props":694,"children":695},{},[696],{"type":54,"value":600},{"type":48,"tag":513,"props":698,"children":699},{},[700],{"type":54,"value":578},{"type":48,"tag":513,"props":702,"children":703},{},[704],{"type":54,"value":583},{"type":48,"tag":513,"props":706,"children":707},{},[708],{"type":54,"value":588},{"type":48,"tag":476,"props":710,"children":711},{},[712,716,721,725,729],{"type":48,"tag":513,"props":713,"children":714},{},[715],{"type":54,"value":124},{"type":48,"tag":513,"props":717,"children":718},{},[719],{"type":54,"value":720},"dataset.quant_calibration_data_sources",{"type":48,"tag":513,"props":722,"children":723},{},[724],{"type":54,"value":578},{"type":48,"tag":513,"props":726,"children":727},{},[728],{"type":54,"value":583},{"type":48,"tag":513,"props":730,"children":731},{},[732],{"type":54,"value":536},{"type":48,"tag":476,"props":734,"children":735},{},[736,740,744,748,752],{"type":48,"tag":513,"props":737,"children":738},{},[739],{"type":54,"value":94},{"type":48,"tag":513,"props":741,"children":742},{},[743],{"type":54,"value":573},{"type":48,"tag":513,"props":745,"children":746},{},[747],{"type":54,"value":578},{"type":48,"tag":513,"props":749,"children":750},{},[751],{"type":54,"value":583},{"type":48,"tag":513,"props":753,"children":754},{},[755],{"type":54,"value":588},{"type":48,"tag":476,"props":757,"children":758},{},[759,763,767,771,775],{"type":48,"tag":513,"props":760,"children":761},{},[762],{"type":54,"value":94},{"type":48,"tag":513,"props":764,"children":765},{},[766],{"type":54,"value":600},{"type":48,"tag":513,"props":768,"children":769},{},[770],{"type":54,"value":578},{"type":48,"tag":513,"props":772,"children":773},{},[774],{"type":54,"value":583},{"type":48,"tag":513,"props":776,"children":777},{},[778],{"type":54,"value":588},{"type":48,"tag":461,"props":780,"children":782},{"id":781},"typical-spec-overrides",[783],{"type":54,"value":784},"Typical Spec Overrides",{"type":48,"tag":57,"props":786,"children":787},{},[788,790,795,797,803],{"type":54,"value":789},"Data source overrides are ",{"type":48,"tag":425,"props":791,"children":792},{},[793],{"type":54,"value":794},"mandatory for every action",{"type":54,"value":796}," — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in ",{"type":48,"tag":68,"props":798,"children":800},{"className":799},[],[801],{"type":54,"value":802},"spec_overrides",{"type":54,"value":804},".",{"type":48,"tag":806,"props":807,"children":812},"pre",{"className":808,"code":809,"language":810,"meta":811,"style":811},"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","",[813],{"type":48,"tag":68,"props":814,"children":815},{"__ignoreMap":811},[816,827],{"type":48,"tag":817,"props":818,"children":821},"span",{"class":819,"line":820},"line",1,[822],{"type":48,"tag":817,"props":823,"children":824},{},[825],{"type":54,"value":826},"S3_TRAIN = \"s3:\u002F\u002Fbucket\u002Fdata\u002Ftrain\"\n",{"type":48,"tag":817,"props":828,"children":830},{"class":819,"line":829},2,[831],{"type":48,"tag":817,"props":832,"children":833},{},[834],{"type":54,"value":835},"S3_EVAL = \"s3:\u002F\u002Fbucket\u002Fdata\u002Feval\"\n",{"type":48,"tag":57,"props":837,"children":838},{},[839],{"type":48,"tag":425,"props":840,"children":841},{},[842],{"type":54,"value":843},"train (mandatory data sources):",{"type":48,"tag":806,"props":845,"children":847},{"className":808,"code":846,"language":810,"meta":811,"style":811},"{\n    \"train.num_epochs\": 10,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"train.gpu_ids\": [0],\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n}\n",[848],{"type":48,"tag":68,"props":849,"children":850},{"__ignoreMap":811},[851,859,867,876,885,894,903,912,921,930,939],{"type":48,"tag":817,"props":852,"children":853},{"class":819,"line":820},[854],{"type":48,"tag":817,"props":855,"children":856},{},[857],{"type":54,"value":858},"{\n",{"type":48,"tag":817,"props":860,"children":861},{"class":819,"line":829},[862],{"type":48,"tag":817,"props":863,"children":864},{},[865],{"type":54,"value":866},"    \"train.num_epochs\": 10,\n",{"type":48,"tag":817,"props":868,"children":870},{"class":819,"line":869},3,[871],{"type":48,"tag":817,"props":872,"children":873},{},[874],{"type":54,"value":875},"    \"train.checkpoint_interval\": 10,\n",{"type":48,"tag":817,"props":877,"children":879},{"class":819,"line":878},4,[880],{"type":48,"tag":817,"props":881,"children":882},{},[883],{"type":54,"value":884},"    \"train.validation_interval\": 10,\n",{"type":48,"tag":817,"props":886,"children":888},{"class":819,"line":887},5,[889],{"type":48,"tag":817,"props":890,"children":891},{},[892],{"type":54,"value":893},"    \"train.num_gpus\": 1,\n",{"type":48,"tag":817,"props":895,"children":897},{"class":819,"line":896},6,[898],{"type":48,"tag":817,"props":899,"children":900},{},[901],{"type":54,"value":902},"    \"train.gpu_ids\": [0],\n",{"type":48,"tag":817,"props":904,"children":906},{"class":819,"line":905},7,[907],{"type":48,"tag":817,"props":908,"children":909},{},[910],{"type":54,"value":911},"    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n",{"type":48,"tag":817,"props":913,"children":915},{"class":819,"line":914},8,[916],{"type":48,"tag":817,"props":917,"children":918},{},[919],{"type":54,"value":920},"    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n",{"type":48,"tag":817,"props":922,"children":924},{"class":819,"line":923},9,[925],{"type":48,"tag":817,"props":926,"children":927},{},[928],{"type":54,"value":929},"    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n",{"type":48,"tag":817,"props":931,"children":933},{"class":819,"line":932},10,[934],{"type":48,"tag":817,"props":935,"children":936},{},[937],{"type":54,"value":938},"    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n",{"type":48,"tag":817,"props":940,"children":942},{"class":819,"line":941},11,[943],{"type":48,"tag":817,"props":944,"children":945},{},[946],{"type":54,"value":947},"}\n",{"type":48,"tag":57,"props":949,"children":950},{},[951],{"type":48,"tag":425,"props":952,"children":953},{},[954],{"type":54,"value":955},"evaluate (mandatory data sources):",{"type":48,"tag":806,"props":957,"children":959},{"className":808,"code":958,"language":810,"meta":811,"style":811},"{\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n    \"dataset.test_data_sources.image_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n    \"dataset.test_data_sources.json_file\": f\"{S3_EVAL}\u002Fannotations.json\",\n}\n",[960],{"type":48,"tag":68,"props":961,"children":962},{"__ignoreMap":811},[963,970,977,984,992,1000],{"type":48,"tag":817,"props":964,"children":965},{"class":819,"line":820},[966],{"type":48,"tag":817,"props":967,"children":968},{},[969],{"type":54,"value":858},{"type":48,"tag":817,"props":971,"children":972},{"class":819,"line":829},[973],{"type":48,"tag":817,"props":974,"children":975},{},[976],{"type":54,"value":911},{"type":48,"tag":817,"props":978,"children":979},{"class":819,"line":869},[980],{"type":48,"tag":817,"props":981,"children":982},{},[983],{"type":54,"value":920},{"type":48,"tag":817,"props":985,"children":986},{"class":819,"line":878},[987],{"type":48,"tag":817,"props":988,"children":989},{},[990],{"type":54,"value":991},"    \"dataset.test_data_sources.image_dir\": f\"{S3_EVAL}\u002Fimages.tar.gz\",\n",{"type":48,"tag":817,"props":993,"children":994},{"class":819,"line":887},[995],{"type":48,"tag":817,"props":996,"children":997},{},[998],{"type":54,"value":999},"    \"dataset.test_data_sources.json_file\": f\"{S3_EVAL}\u002Fannotations.json\",\n",{"type":48,"tag":817,"props":1001,"children":1002},{"class":819,"line":896},[1003],{"type":48,"tag":817,"props":1004,"children":1005},{},[1006],{"type":54,"value":947},{"type":48,"tag":57,"props":1008,"children":1009},{},[1010,1012,1018,1019,1025,1026,1032,1034,1040,1042,1048,1049,1055],{"type":54,"value":1011},"If the train or AutoML run changed architecture-affecting fields such as\n",{"type":48,"tag":68,"props":1013,"children":1015},{"className":1014},[],[1016],{"type":54,"value":1017},"model.enc_layers",{"type":54,"value":96},{"type":48,"tag":68,"props":1020,"children":1022},{"className":1021},[],[1023],{"type":54,"value":1024},"model.dec_layers",{"type":54,"value":96},{"type":48,"tag":68,"props":1027,"children":1029},{"className":1028},[],[1030],{"type":54,"value":1031},"model.num_queries",{"type":54,"value":1033},", or\n",{"type":48,"tag":68,"props":1035,"children":1037},{"className":1036},[],[1038],{"type":54,"value":1039},"model.num_select",{"type":54,"value":1041},", carry the same values into evaluate, export, inference, and\ndeploy actions with the selected checkpoint. In addition to the fields above,\ncarry ",{"type":48,"tag":68,"props":1043,"children":1045},{"className":1044},[],[1046],{"type":54,"value":1047},"model.num_feature_levels",{"type":54,"value":96},{"type":48,"tag":68,"props":1050,"children":1052},{"className":1051},[],[1053],{"type":54,"value":1054},"model.dim_feedforward",{"type":54,"value":1056},", input image\ndimensions, and dataset class metadata when they were changed. Loading a\ncheckpoint into the default architecture can fail with tensor shape mismatches,\nespecially when smoke-test runs shrink the transformer for speed.",{"type":48,"tag":57,"props":1058,"children":1059},{},[1060],{"type":48,"tag":425,"props":1061,"children":1062},{},[1063],{"type":54,"value":1064},"export (mandatory data sources):",{"type":48,"tag":806,"props":1066,"children":1068},{"className":808,"code":1067,"language":810,"meta":811,"style":811},"{\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.eval_class_ids\": [1, 2, \"...\"],\n    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n}\n",[1069],{"type":48,"tag":68,"props":1070,"children":1071},{"__ignoreMap":811},[1072,1079,1086,1093,1100,1107],{"type":48,"tag":817,"props":1073,"children":1074},{"class":819,"line":820},[1075],{"type":48,"tag":817,"props":1076,"children":1077},{},[1078],{"type":54,"value":858},{"type":48,"tag":817,"props":1080,"children":1081},{"class":819,"line":829},[1082],{"type":48,"tag":817,"props":1083,"children":1084},{},[1085],{"type":54,"value":911},{"type":48,"tag":817,"props":1087,"children":1088},{"class":819,"line":869},[1089],{"type":48,"tag":817,"props":1090,"children":1091},{},[1092],{"type":54,"value":920},{"type":48,"tag":817,"props":1094,"children":1095},{"class":819,"line":878},[1096],{"type":48,"tag":817,"props":1097,"children":1098},{},[1099],{"type":54,"value":929},{"type":48,"tag":817,"props":1101,"children":1102},{"class":819,"line":887},[1103],{"type":48,"tag":817,"props":1104,"children":1105},{},[1106],{"type":54,"value":938},{"type":48,"tag":817,"props":1108,"children":1109},{"class":819,"line":896},[1110],{"type":48,"tag":817,"props":1111,"children":1112},{},[1113],{"type":54,"value":947},{"type":48,"tag":57,"props":1115,"children":1116},{},[1117],{"type":48,"tag":425,"props":1118,"children":1119},{},[1120],{"type":54,"value":1121},"TensorRT engine generation:",{"type":48,"tag":57,"props":1123,"children":1124},{},[1125,1127,1132,1134,1140,1142,1148,1149,1154,1155,1160,1161,1166,1167,1172,1173,1178,1180,1186,1188,1193],{"type":54,"value":1126},"Use the deploy spec templates after ",{"type":48,"tag":68,"props":1128,"children":1130},{"className":1129},[],[1131],{"type":54,"value":116},{"type":54,"value":1133},". Do not call ",{"type":48,"tag":68,"props":1135,"children":1137},{"className":1136},[],[1138],{"type":54,"value":1139},"deformable_detr gen_trt_engine",{"type":54,"value":1141}," from the parent PyT model container; that CLI advertises\n",{"type":48,"tag":68,"props":1143,"children":1145},{"className":1144},[],[1146],{"type":54,"value":1147},"convert",{"type":54,"value":96},{"type":48,"tag":68,"props":1150,"children":1152},{"className":1151},[],[1153],{"type":54,"value":102},{"type":54,"value":96},{"type":48,"tag":68,"props":1156,"children":1158},{"className":1157},[],[1159],{"type":54,"value":116},{"type":54,"value":96},{"type":48,"tag":68,"props":1162,"children":1164},{"className":1163},[],[1165],{"type":54,"value":109},{"type":54,"value":96},{"type":48,"tag":68,"props":1168,"children":1170},{"className":1169},[],[1171],{"type":54,"value":124},{"type":54,"value":96},{"type":48,"tag":68,"props":1174,"children":1176},{"className":1175},[],[1177],{"type":54,"value":94},{"type":54,"value":1179},", and\n",{"type":48,"tag":68,"props":1181,"children":1183},{"className":1182},[],[1184],{"type":54,"value":1185},"default_specs",{"type":54,"value":1187},", but not ",{"type":48,"tag":68,"props":1189,"children":1191},{"className":1190},[],[1192],{"type":54,"value":132},{"type":54,"value":1194},". The model action metadata selects the\nTAO Deploy container for engine generation.",{"type":48,"tag":57,"props":1196,"children":1197},{},[1198,1200,1206],{"type":54,"value":1199},"Deploy engine generation needs the exported ONNX file as input and creates the\nengine at ",{"type":48,"tag":68,"props":1201,"children":1203},{"className":1202},[],[1204],{"type":54,"value":1205},"gen_trt_engine.trt_engine",{"type":54,"value":804},{"type":48,"tag":806,"props":1208,"children":1210},{"className":808,"code":1209,"language":810,"meta":811,"style":811},"{\n    \"gen_trt_engine.tensorrt.data_type\": \"FP16\",\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n}\n",[1211],{"type":48,"tag":68,"props":1212,"children":1213},{"__ignoreMap":811},[1214,1221,1229,1236,1244],{"type":48,"tag":817,"props":1215,"children":1216},{"class":819,"line":820},[1217],{"type":48,"tag":817,"props":1218,"children":1219},{},[1220],{"type":54,"value":858},{"type":48,"tag":817,"props":1222,"children":1223},{"class":819,"line":829},[1224],{"type":48,"tag":817,"props":1225,"children":1226},{},[1227],{"type":54,"value":1228},"    \"gen_trt_engine.tensorrt.data_type\": \"FP16\",\n",{"type":48,"tag":817,"props":1230,"children":1231},{"class":819,"line":869},[1232],{"type":48,"tag":817,"props":1233,"children":1234},{},[1235],{"type":54,"value":911},{"type":48,"tag":817,"props":1237,"children":1238},{"class":819,"line":878},[1239],{"type":48,"tag":817,"props":1240,"children":1241},{},[1242],{"type":54,"value":1243},"    \"gen_trt_engine.tensorrt.calibration.cal_image_dir\": [f\"{S3_TRAIN}\u002Fimages.tar.gz\"],\n",{"type":48,"tag":817,"props":1245,"children":1246},{"class":819,"line":887},[1247],{"type":48,"tag":817,"props":1248,"children":1249},{},[1250],{"type":54,"value":947},{"type":48,"tag":57,"props":1252,"children":1253},{},[1254],{"type":48,"tag":425,"props":1255,"children":1256},{},[1257],{"type":54,"value":1258},"inference (mandatory data sources):",{"type":48,"tag":806,"props":1260,"children":1262},{"className":808,"code":1261,"language":810,"meta":811,"style":811},"{\n    \"dataset.num_classes\": \"\u003Cobject classes> + 1\",\n    \"dataset.infer_data_sources.image_dir\": [f\"{S3_EVAL}\u002Fimages.tar.gz\"],\n    \"dataset.infer_data_sources.classmap\": f\"{S3_EVAL}\u002Flabel_map.txt\",\n}\n",[1263],{"type":48,"tag":68,"props":1264,"children":1265},{"__ignoreMap":811},[1266,1273,1280,1288,1296],{"type":48,"tag":817,"props":1267,"children":1268},{"class":819,"line":820},[1269],{"type":48,"tag":817,"props":1270,"children":1271},{},[1272],{"type":54,"value":858},{"type":48,"tag":817,"props":1274,"children":1275},{"class":819,"line":829},[1276],{"type":48,"tag":817,"props":1277,"children":1278},{},[1279],{"type":54,"value":911},{"type":48,"tag":817,"props":1281,"children":1282},{"class":819,"line":869},[1283],{"type":48,"tag":817,"props":1284,"children":1285},{},[1286],{"type":54,"value":1287},"    \"dataset.infer_data_sources.image_dir\": [f\"{S3_EVAL}\u002Fimages.tar.gz\"],\n",{"type":48,"tag":817,"props":1289,"children":1290},{"class":819,"line":878},[1291],{"type":48,"tag":817,"props":1292,"children":1293},{},[1294],{"type":54,"value":1295},"    \"dataset.infer_data_sources.classmap\": f\"{S3_EVAL}\u002Flabel_map.txt\",\n",{"type":48,"tag":817,"props":1297,"children":1298},{"class":819,"line":887},[1299],{"type":48,"tag":817,"props":1300,"children":1301},{},[1302],{"type":54,"value":947},{"type":48,"tag":57,"props":1304,"children":1305},{},[1306],{"type":48,"tag":425,"props":1307,"children":1308},{},[1309],{"type":54,"value":1310},"quantize (mandatory data sources):",{"type":48,"tag":806,"props":1312,"children":1314},{"className":808,"code":1313,"language":810,"meta":811,"style":811},"{\n    \"dataset.train_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.val_data_sources\": [{\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"}],\n    \"dataset.quant_calibration_data_sources\": {\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"},\n}\n",[1315],{"type":48,"tag":68,"props":1316,"children":1317},{"__ignoreMap":811},[1318,1325,1332,1339,1347],{"type":48,"tag":817,"props":1319,"children":1320},{"class":819,"line":820},[1321],{"type":48,"tag":817,"props":1322,"children":1323},{},[1324],{"type":54,"value":858},{"type":48,"tag":817,"props":1326,"children":1327},{"class":819,"line":829},[1328],{"type":48,"tag":817,"props":1329,"children":1330},{},[1331],{"type":54,"value":929},{"type":48,"tag":817,"props":1333,"children":1334},{"class":819,"line":869},[1335],{"type":48,"tag":817,"props":1336,"children":1337},{},[1338],{"type":54,"value":938},{"type":48,"tag":817,"props":1340,"children":1341},{"class":819,"line":878},[1342],{"type":48,"tag":817,"props":1343,"children":1344},{},[1345],{"type":54,"value":1346},"    \"dataset.quant_calibration_data_sources\": {\"image_dir\": f\"{S3_TRAIN}\u002Fimages.tar.gz\", \"json_file\": f\"{S3_TRAIN}\u002Fannotations.json\"},\n",{"type":48,"tag":817,"props":1348,"children":1349},{"class":819,"line":887},[1350],{"type":48,"tag":817,"props":1351,"children":1352},{},[1353],{"type":54,"value":947},{"type":48,"tag":167,"props":1355,"children":1357},{"id":1356},"eval-dataset",[1358],{"type":54,"value":1359},"Eval Dataset",{"type":48,"tag":57,"props":1361,"children":1362},{},[1363],{"type":54,"value":1364},"Optional. If provided, validation mAP is computed at each checkpoint interval.",{"type":48,"tag":167,"props":1366,"children":1368},{"id":1367},"checkpoint-handling",[1369],{"type":54,"value":1370},"Checkpoint Handling",{"type":48,"tag":57,"props":1372,"children":1373},{},[1374,1376,1382,1384,1390,1392,1397,1399,1405],{"type":54,"value":1375},"Training emits epoch-and-step checkpoints using the pattern\n",{"type":48,"tag":68,"props":1377,"children":1379},{"className":1378},[],[1380],{"type":54,"value":1381},"model_epoch_\u003Cepoch>_step_\u003Cstep>.pth",{"type":54,"value":1383},", plus a ",{"type":48,"tag":68,"props":1385,"children":1387},{"className":1386},[],[1388],{"type":54,"value":1389},"dd_model_latest.pth",{"type":54,"value":1391}," symlink. For\ndependent actions, use the model-specific or SDK-provided checkpoint resolver to\nselect the intended artifact. Evaluation, inference, export, and quantize should\nreceive the selected exact checkpoint path, not the ",{"type":48,"tag":68,"props":1393,"children":1395},{"className":1394},[],[1396],{"type":54,"value":1389},{"type":54,"value":1398}," symlink,\nunless the user explicitly asked for latest. Resume\u002Fretrain should set\n",{"type":48,"tag":68,"props":1400,"children":1402},{"className":1401},[],[1403],{"type":54,"value":1404},"train.resume_training_checkpoint_path",{"type":54,"value":1406}," to the exact checkpoint being resumed\nfrom.",{"type":48,"tag":167,"props":1408,"children":1410},{"id":1409},"important-parameters",[1411],{"type":54,"value":1412},"Important Parameters",{"type":48,"tag":417,"props":1414,"children":1415},{},[1416,1426,1436,1446,1456,1473,1482,1492],{"type":48,"tag":421,"props":1417,"children":1418},{},[1419,1424],{"type":48,"tag":425,"props":1420,"children":1421},{},[1422],{"type":54,"value":1423},"dataset.num_classes",{"type":54,"value":1425},": Number of object classes plus the background class. Default 91 (COCO). Must match annotations.",{"type":48,"tag":421,"props":1427,"children":1428},{},[1429,1434],{"type":48,"tag":425,"props":1430,"children":1431},{},[1432],{"type":54,"value":1433},"dataset.eval_class_ids",{"type":54,"value":1435},": Foreground category ids to include in COCO metrics. Set this to every object category id in custom datasets; the template default evaluates class id 1 only.",{"type":48,"tag":421,"props":1437,"children":1438},{},[1439,1444],{"type":48,"tag":425,"props":1440,"children":1441},{},[1442],{"type":54,"value":1443},"model.backbone",{"type":54,"value":1445},": Default resnet_50. Supported: resnet_50, gcvit_tiny, gcvit_small, gcvit_base, gcvit_large, gcvit_large_384 (more limited than DINO).",{"type":48,"tag":421,"props":1447,"children":1448},{},[1449,1454],{"type":48,"tag":425,"props":1450,"children":1451},{},[1452],{"type":54,"value":1453},"train.optim.lr",{"type":54,"value":1455},": Learning rate. Default 2e-4 (AdamW). lr_backbone is 2e-5.",{"type":48,"tag":421,"props":1457,"children":1458},{},[1459,1464,1466,1471],{"type":48,"tag":425,"props":1460,"children":1461},{},[1462],{"type":54,"value":1463},"train.optim.lr_steps",{"type":54,"value":1465},": MultiStep LR schedule. Default ",{"type":48,"tag":817,"props":1467,"children":1468},{},[1469],{"type":54,"value":1470},"40",{"type":54,"value":1472},". For short runs, set to match ~80% of total epochs.",{"type":48,"tag":421,"props":1474,"children":1475},{},[1476,1480],{"type":48,"tag":425,"props":1477,"children":1478},{},[1479],{"type":54,"value":1031},{"type":54,"value":1481},": Number of object queries. Default 300. Valid range 100-900.",{"type":48,"tag":421,"props":1483,"children":1484},{},[1485,1490],{"type":48,"tag":425,"props":1486,"children":1487},{},[1488],{"type":54,"value":1489},"model.dropout_ratio",{"type":54,"value":1491},": Dropout in transformer layers. Default 0.3 (higher than DINO's 0.0). Reduce for large datasets, increase for small datasets.",{"type":48,"tag":421,"props":1493,"children":1494},{},[1495,1499],{"type":48,"tag":425,"props":1496,"children":1497},{},[1498],{"type":54,"value":1054},{"type":54,"value":1500},": FFN hidden dim. Default 1024 (vs DINO's 2048). Increasing improves capacity but costs memory.",{"type":48,"tag":167,"props":1502,"children":1504},{"id":1503},"multi-gpu-multi-node",[1505],{"type":54,"value":1506},"Multi-GPU \u002F Multi-Node",{"type":48,"tag":57,"props":1508,"children":1509},{},[1510,1515,1517,1522],{"type":48,"tag":425,"props":1511,"children":1512},{},[1513],{"type":54,"value":1514},"Launch method:",{"type":54,"value":1516}," Lightning-managed (single ",{"type":48,"tag":68,"props":1518,"children":1520},{"className":1519},[],[1521],{"type":54,"value":810},{"type":54,"value":1523}," process, Lightning spawns workers).",{"type":48,"tag":468,"props":1525,"children":1526},{},[1527,1547],{"type":48,"tag":472,"props":1528,"children":1529},{},[1530],{"type":48,"tag":476,"props":1531,"children":1532},{},[1533,1537,1542],{"type":48,"tag":480,"props":1534,"children":1535},{},[1536],{"type":54,"value":489},{"type":48,"tag":480,"props":1538,"children":1539},{},[1540],{"type":54,"value":1541},"Description",{"type":48,"tag":480,"props":1543,"children":1544},{},[1545],{"type":54,"value":1546},"Default",{"type":48,"tag":506,"props":1548,"children":1549},{},[1550,1572,1597,1618],{"type":48,"tag":476,"props":1551,"children":1552},{},[1553,1562,1567],{"type":48,"tag":513,"props":1554,"children":1555},{},[1556],{"type":48,"tag":68,"props":1557,"children":1559},{"className":1558},[],[1560],{"type":54,"value":1561},"train.num_gpus",{"type":48,"tag":513,"props":1563,"children":1564},{},[1565],{"type":54,"value":1566},"Number of GPUs",{"type":48,"tag":513,"props":1568,"children":1569},{},[1570],{"type":54,"value":1571},"1",{"type":48,"tag":476,"props":1573,"children":1574},{},[1575,1584,1589],{"type":48,"tag":513,"props":1576,"children":1577},{},[1578],{"type":48,"tag":68,"props":1579,"children":1581},{"className":1580},[],[1582],{"type":54,"value":1583},"train.gpu_ids",{"type":48,"tag":513,"props":1585,"children":1586},{},[1587],{"type":54,"value":1588},"GPU device indices",{"type":48,"tag":513,"props":1590,"children":1591},{},[1592],{"type":48,"tag":817,"props":1593,"children":1594},{},[1595],{"type":54,"value":1596},"0",{"type":48,"tag":476,"props":1598,"children":1599},{},[1600,1609,1614],{"type":48,"tag":513,"props":1601,"children":1602},{},[1603],{"type":48,"tag":68,"props":1604,"children":1606},{"className":1605},[],[1607],{"type":54,"value":1608},"train.num_nodes",{"type":48,"tag":513,"props":1610,"children":1611},{},[1612],{"type":54,"value":1613},"Number of nodes",{"type":48,"tag":513,"props":1615,"children":1616},{},[1617],{"type":54,"value":1571},{"type":48,"tag":476,"props":1619,"children":1620},{},[1621,1630,1647],{"type":48,"tag":513,"props":1622,"children":1623},{},[1624],{"type":48,"tag":68,"props":1625,"children":1627},{"className":1626},[],[1628],{"type":54,"value":1629},"train.distributed_strategy",{"type":48,"tag":513,"props":1631,"children":1632},{},[1633,1639,1641],{"type":48,"tag":68,"props":1634,"children":1636},{"className":1635},[],[1637],{"type":54,"value":1638},"ddp",{"type":54,"value":1640}," or ",{"type":48,"tag":68,"props":1642,"children":1644},{"className":1643},[],[1645],{"type":54,"value":1646},"fsdp",{"type":48,"tag":513,"props":1648,"children":1649},{},[1650],{"type":48,"tag":68,"props":1651,"children":1653},{"className":1652},[],[1654],{"type":54,"value":1638},{"type":48,"tag":57,"props":1656,"children":1657},{},[1658,1660,1666,1667,1673,1674,1680,1681,1687],{"type":54,"value":1659},"Same DDP\u002FFSDP behavior as DINO. Multi-node requires ",{"type":48,"tag":68,"props":1661,"children":1663},{"className":1662},[],[1664],{"type":54,"value":1665},"WORLD_SIZE",{"type":54,"value":96},{"type":48,"tag":68,"props":1668,"children":1670},{"className":1669},[],[1671],{"type":54,"value":1672},"NODE_RANK",{"type":54,"value":96},{"type":48,"tag":68,"props":1675,"children":1677},{"className":1676},[],[1678],{"type":54,"value":1679},"MASTER_ADDR",{"type":54,"value":96},{"type":48,"tag":68,"props":1682,"children":1684},{"className":1683},[],[1685],{"type":54,"value":1686},"MASTER_PORT",{"type":54,"value":1688}," env vars set by orchestrator.",{"type":48,"tag":57,"props":1690,"children":1691},{},[1692,1694,1699,1701,1706,1708,1714,1715,1721],{"type":54,"value":1693},"When increasing ",{"type":48,"tag":68,"props":1695,"children":1697},{"className":1696},[],[1698],{"type":54,"value":1561},{"type":54,"value":1700},", also set ",{"type":48,"tag":68,"props":1702,"children":1704},{"className":1703},[],[1705],{"type":54,"value":1583},{"type":54,"value":1707}," to the same visible\ndevice range. For example, an 8-GPU single-node Slurm run must include both\n",{"type":48,"tag":68,"props":1709,"children":1711},{"className":1710},[],[1712],{"type":54,"value":1713},"\"train.num_gpus\": 8",{"type":54,"value":232},{"type":48,"tag":68,"props":1716,"children":1718},{"className":1717},[],[1719],{"type":54,"value":1720},"\"train.gpu_ids\": [0, 1, 2, 3, 4, 5, 6, 7]",{"type":54,"value":804},{"type":48,"tag":167,"props":1723,"children":1725},{"id":1724},"export-trt-defaults",[1726],{"type":54,"value":1727},"Export \u002F TRT Defaults",{"type":48,"tag":417,"props":1729,"children":1730},{},[1731,1736,1741,1746],{"type":48,"tag":421,"props":1732,"children":1733},{},[1734],{"type":54,"value":1735},"Export input: 640x640, opset 17",{"type":48,"tag":421,"props":1737,"children":1738},{},[1739],{"type":54,"value":1740},"TRT data types: FP32, FP16, INT8",{"type":48,"tag":421,"props":1742,"children":1743},{},[1744],{"type":54,"value":1745},"TRT workspace: 1024 MB",{"type":48,"tag":421,"props":1747,"children":1748},{},[1749],{"type":54,"value":1750},"TRT max_batch_size: 1",{"type":48,"tag":167,"props":1752,"children":1754},{"id":1753},"hardware",[1755],{"type":54,"value":1756},"Hardware",{"type":48,"tag":57,"props":1758,"children":1759},{},[1760],{"type":54,"value":1761},"Minimum 1 GPU(s), recommended 4 GPU(s). 16GB+ (V100 or A100) VRAM per GPU. Slightly lighter than DINO due to smaller FFN. batch_size=4 fits on most 16GB+ GPUs.",{"type":48,"tag":167,"props":1763,"children":1765},{"id":1764},"error-patterns",[1766],{"type":54,"value":1767},"Error Patterns",{"type":48,"tag":57,"props":1769,"children":1770},{},[1771,1776],{"type":48,"tag":425,"props":1772,"children":1773},{},[1774],{"type":54,"value":1775},"CUDA out of memory",{"type":54,"value":1777},": Reduce batch_size (4 -> 2 -> 1).",{"type":48,"tag":57,"props":1779,"children":1780},{},[1781,1786],{"type":48,"tag":425,"props":1782,"children":1783},{},[1784],{"type":54,"value":1785},"num_select must be \u003C num_queries * num_classes",{"type":54,"value":1787},": Same constraint as DINO.",{"type":48,"tag":57,"props":1789,"children":1790},{},[1791,1796,1798,1803],{"type":48,"tag":425,"props":1792,"children":1793},{},[1794],{"type":54,"value":1795},"return_interm_indices length must match num_feature_levels",{"type":54,"value":1797},": Default ",{"type":48,"tag":817,"props":1799,"children":1800},{},[1801],{"type":54,"value":1802},"1,2,3,4",{"type":54,"value":1804}," with num_feature_levels=4.",{"type":48,"tag":57,"props":1806,"children":1807},{},[1808,1813],{"type":48,"tag":425,"props":1809,"children":1810},{},[1811],{"type":54,"value":1812},"Dataset size smaller than total batch size",{"type":54,"value":1814},": Reduce batch_size or num_gpus.",{"type":48,"tag":57,"props":1816,"children":1817},{},[1818,1823,1825,1830,1832,1838,1840,1846,1848,1854,1856,1862],{"type":48,"tag":425,"props":1819,"children":1820},{},[1821],{"type":54,"value":1822},"AutoML metric extraction",{"type":54,"value":1824},": Deformable DETR emits detection metrics in structured training status and logs. For COCO\u002Fpaper-style benchmark comparisons, optimize ",{"type":48,"tag":68,"props":1826,"children":1828},{"className":1827},[],[1829],{"type":54,"value":457},{"type":54,"value":1831}," with ",{"type":48,"tag":68,"props":1833,"children":1835},{"className":1834},[],[1836],{"type":54,"value":1837},"direction: maximize",{"type":54,"value":1839},"; for explicit AP50 workflows, optimize ",{"type":48,"tag":68,"props":1841,"children":1843},{"className":1842},[],[1844],{"type":54,"value":1845},"val_mAP50",{"type":54,"value":1847},". Prefer ",{"type":48,"tag":68,"props":1849,"children":1851},{"className":1850},[],[1852],{"type":54,"value":1853},"results_dir\u002Ftrain\u002Fstatus.json",{"type":54,"value":1855}," or AutoML result state before parsing raw logs. Do not optimize ",{"type":48,"tag":68,"props":1857,"children":1859},{"className":1858},[],[1860],{"type":54,"value":1861},"val_loss",{"type":54,"value":1863}," for default detection model invocations.",{"type":48,"tag":167,"props":1865,"children":1867},{"id":1866},"spec-param-parent-model-inference",[1868],{"type":54,"value":1869},"Spec Param \u002F Parent Model Inference",{"type":48,"tag":57,"props":1871,"children":1872},{},[1873,1875,1881,1883,1889,1891,1897],{"type":54,"value":1874},"Model-specific inference mappings belong in this MD file, not in ",{"type":48,"tag":68,"props":1876,"children":1878},{"className":1877},[],[1879],{"type":54,"value":1880},"config.json",{"type":54,"value":1882},". Generated runners should read this section and apply the mappings with SDK helpers before ",{"type":48,"tag":68,"props":1884,"children":1886},{"className":1885},[],[1887],{"type":54,"value":1888},"create_job()",{"type":54,"value":1890},". This mirrors the old microservices ",{"type":48,"tag":68,"props":1892,"children":1894},{"className":1893},[],[1895],{"type":54,"value":1896},"infer_params.py",{"type":54,"value":1898}," flow.",{"type":48,"tag":57,"props":1900,"children":1901},{},[1902,1904,1910],{"type":54,"value":1903},"Inference mappings from TAO Core ",{"type":48,"tag":68,"props":1905,"children":1907},{"className":1906},[],[1908],{"type":54,"value":1909},"deformable_detr.config.json",{"type":54,"value":1911},":",{"type":48,"tag":468,"props":1913,"children":1914},{},[1915,1940],{"type":48,"tag":472,"props":1916,"children":1917},{},[1918],{"type":48,"tag":476,"props":1919,"children":1920},{},[1921,1925,1930,1935],{"type":48,"tag":480,"props":1922,"children":1923},{},[1924],{"type":54,"value":484},{"type":48,"tag":480,"props":1926,"children":1927},{},[1928],{"type":54,"value":1929},"Spec Field",{"type":48,"tag":480,"props":1931,"children":1932},{},[1933],{"type":54,"value":1934},"Inference Function",{"type":48,"tag":480,"props":1936,"children":1937},{},[1938],{"type":54,"value":1939},"Meaning",{"type":48,"tag":506,"props":1941,"children":1942},{},[1943,1973,2003,2031,2061,2088,2116,2146,2173,2200,2228,2256,2283,2310,2338,2365,2392,2421,2448,2476],{"type":48,"tag":476,"props":1944,"children":1945},{},[1946,1950,1959,1968],{"type":48,"tag":513,"props":1947,"children":1948},{},[1949],{"type":54,"value":102},{"type":48,"tag":513,"props":1951,"children":1952},{},[1953],{"type":48,"tag":68,"props":1954,"children":1956},{"className":1955},[],[1957],{"type":54,"value":1958},"encryption_key",{"type":48,"tag":513,"props":1960,"children":1961},{},[1962],{"type":48,"tag":68,"props":1963,"children":1965},{"className":1964},[],[1966],{"type":54,"value":1967},"key",{"type":48,"tag":513,"props":1969,"children":1970},{},[1971],{"type":54,"value":1972},"encryption key",{"type":48,"tag":476,"props":1974,"children":1975},{},[1976,1980,1989,1998],{"type":48,"tag":513,"props":1977,"children":1978},{},[1979],{"type":54,"value":102},{"type":48,"tag":513,"props":1981,"children":1982},{},[1983],{"type":48,"tag":68,"props":1984,"children":1986},{"className":1985},[],[1987],{"type":54,"value":1988},"evaluate.checkpoint",{"type":48,"tag":513,"props":1990,"children":1991},{},[1992],{"type":48,"tag":68,"props":1993,"children":1995},{"className":1994},[],[1996],{"type":54,"value":1997},"parent_model",{"type":48,"tag":513,"props":1999,"children":2000},{},[2001],{"type":54,"value":2002},"model file inferred from the parent job results folder",{"type":48,"tag":476,"props":2004,"children":2005},{},[2006,2010,2019,2027],{"type":48,"tag":513,"props":2007,"children":2008},{},[2009],{"type":54,"value":102},{"type":48,"tag":513,"props":2011,"children":2012},{},[2013],{"type":48,"tag":68,"props":2014,"children":2016},{"className":2015},[],[2017],{"type":54,"value":2018},"evaluate.trt_engine",{"type":48,"tag":513,"props":2020,"children":2021},{},[2022],{"type":48,"tag":68,"props":2023,"children":2025},{"className":2024},[],[2026],{"type":54,"value":1997},{"type":48,"tag":513,"props":2028,"children":2029},{},[2030],{"type":54,"value":2002},{"type":48,"tag":476,"props":2032,"children":2033},{},[2034,2038,2047,2056],{"type":48,"tag":513,"props":2035,"children":2036},{},[2037],{"type":54,"value":102},{"type":48,"tag":513,"props":2039,"children":2040},{},[2041],{"type":48,"tag":68,"props":2042,"children":2044},{"className":2043},[],[2045],{"type":54,"value":2046},"results_dir",{"type":48,"tag":513,"props":2048,"children":2049},{},[2050],{"type":48,"tag":68,"props":2051,"children":2053},{"className":2052},[],[2054],{"type":54,"value":2055},"output_dir",{"type":48,"tag":513,"props":2057,"children":2058},{},[2059],{"type":54,"value":2060},"current job results directory",{"type":48,"tag":476,"props":2062,"children":2063},{},[2064,2068,2076,2084],{"type":48,"tag":513,"props":2065,"children":2066},{},[2067],{"type":54,"value":116},{"type":48,"tag":513,"props":2069,"children":2070},{},[2071],{"type":48,"tag":68,"props":2072,"children":2074},{"className":2073},[],[2075],{"type":54,"value":1958},{"type":48,"tag":513,"props":2077,"children":2078},{},[2079],{"type":48,"tag":68,"props":2080,"children":2082},{"className":2081},[],[2083],{"type":54,"value":1967},{"type":48,"tag":513,"props":2085,"children":2086},{},[2087],{"type":54,"value":1972},{"type":48,"tag":476,"props":2089,"children":2090},{},[2091,2095,2104,2112],{"type":48,"tag":513,"props":2092,"children":2093},{},[2094],{"type":54,"value":116},{"type":48,"tag":513,"props":2096,"children":2097},{},[2098],{"type":48,"tag":68,"props":2099,"children":2101},{"className":2100},[],[2102],{"type":54,"value":2103},"export.checkpoint",{"type":48,"tag":513,"props":2105,"children":2106},{},[2107],{"type":48,"tag":68,"props":2108,"children":2110},{"className":2109},[],[2111],{"type":54,"value":1997},{"type":48,"tag":513,"props":2113,"children":2114},{},[2115],{"type":54,"value":2002},{"type":48,"tag":476,"props":2117,"children":2118},{},[2119,2123,2132,2141],{"type":48,"tag":513,"props":2120,"children":2121},{},[2122],{"type":54,"value":116},{"type":48,"tag":513,"props":2124,"children":2125},{},[2126],{"type":48,"tag":68,"props":2127,"children":2129},{"className":2128},[],[2130],{"type":54,"value":2131},"export.onnx_file",{"type":48,"tag":513,"props":2133,"children":2134},{},[2135],{"type":48,"tag":68,"props":2136,"children":2138},{"className":2137},[],[2139],{"type":54,"value":2140},"create_onnx_file",{"type":48,"tag":513,"props":2142,"children":2143},{},[2144],{"type":54,"value":2145},"output ONNX path",{"type":48,"tag":476,"props":2147,"children":2148},{},[2149,2153,2161,2169],{"type":48,"tag":513,"props":2150,"children":2151},{},[2152],{"type":54,"value":116},{"type":48,"tag":513,"props":2154,"children":2155},{},[2156],{"type":48,"tag":68,"props":2157,"children":2159},{"className":2158},[],[2160],{"type":54,"value":2046},{"type":48,"tag":513,"props":2162,"children":2163},{},[2164],{"type":48,"tag":68,"props":2165,"children":2167},{"className":2166},[],[2168],{"type":54,"value":2055},{"type":48,"tag":513,"props":2170,"children":2171},{},[2172],{"type":54,"value":2060},{"type":48,"tag":476,"props":2174,"children":2175},{},[2176,2180,2188,2196],{"type":48,"tag":513,"props":2177,"children":2178},{},[2179],{"type":54,"value":109},{"type":48,"tag":513,"props":2181,"children":2182},{},[2183],{"type":48,"tag":68,"props":2184,"children":2186},{"className":2185},[],[2187],{"type":54,"value":1958},{"type":48,"tag":513,"props":2189,"children":2190},{},[2191],{"type":48,"tag":68,"props":2192,"children":2194},{"className":2193},[],[2195],{"type":54,"value":1967},{"type":48,"tag":513,"props":2197,"children":2198},{},[2199],{"type":54,"value":1972},{"type":48,"tag":476,"props":2201,"children":2202},{},[2203,2207,2216,2224],{"type":48,"tag":513,"props":2204,"children":2205},{},[2206],{"type":54,"value":109},{"type":48,"tag":513,"props":2208,"children":2209},{},[2210],{"type":48,"tag":68,"props":2211,"children":2213},{"className":2212},[],[2214],{"type":54,"value":2215},"inference.checkpoint",{"type":48,"tag":513,"props":2217,"children":2218},{},[2219],{"type":48,"tag":68,"props":2220,"children":2222},{"className":2221},[],[2223],{"type":54,"value":1997},{"type":48,"tag":513,"props":2225,"children":2226},{},[2227],{"type":54,"value":2002},{"type":48,"tag":476,"props":2229,"children":2230},{},[2231,2235,2244,2252],{"type":48,"tag":513,"props":2232,"children":2233},{},[2234],{"type":54,"value":109},{"type":48,"tag":513,"props":2236,"children":2237},{},[2238],{"type":48,"tag":68,"props":2239,"children":2241},{"className":2240},[],[2242],{"type":54,"value":2243},"inference.trt_engine",{"type":48,"tag":513,"props":2245,"children":2246},{},[2247],{"type":48,"tag":68,"props":2248,"children":2250},{"className":2249},[],[2251],{"type":54,"value":1997},{"type":48,"tag":513,"props":2253,"children":2254},{},[2255],{"type":54,"value":2002},{"type":48,"tag":476,"props":2257,"children":2258},{},[2259,2263,2271,2279],{"type":48,"tag":513,"props":2260,"children":2261},{},[2262],{"type":54,"value":109},{"type":48,"tag":513,"props":2264,"children":2265},{},[2266],{"type":48,"tag":68,"props":2267,"children":2269},{"className":2268},[],[2270],{"type":54,"value":2046},{"type":48,"tag":513,"props":2272,"children":2273},{},[2274],{"type":48,"tag":68,"props":2275,"children":2277},{"className":2276},[],[2278],{"type":54,"value":2055},{"type":48,"tag":513,"props":2280,"children":2281},{},[2282],{"type":54,"value":2060},{"type":48,"tag":476,"props":2284,"children":2285},{},[2286,2290,2298,2306],{"type":48,"tag":513,"props":2287,"children":2288},{},[2289],{"type":54,"value":124},{"type":48,"tag":513,"props":2291,"children":2292},{},[2293],{"type":48,"tag":68,"props":2294,"children":2296},{"className":2295},[],[2297],{"type":54,"value":1958},{"type":48,"tag":513,"props":2299,"children":2300},{},[2301],{"type":48,"tag":68,"props":2302,"children":2304},{"className":2303},[],[2305],{"type":54,"value":1967},{"type":48,"tag":513,"props":2307,"children":2308},{},[2309],{"type":54,"value":1972},{"type":48,"tag":476,"props":2311,"children":2312},{},[2313,2317,2326,2334],{"type":48,"tag":513,"props":2314,"children":2315},{},[2316],{"type":54,"value":124},{"type":48,"tag":513,"props":2318,"children":2319},{},[2320],{"type":48,"tag":68,"props":2321,"children":2323},{"className":2322},[],[2324],{"type":54,"value":2325},"quantize.model_path",{"type":48,"tag":513,"props":2327,"children":2328},{},[2329],{"type":48,"tag":68,"props":2330,"children":2332},{"className":2331},[],[2333],{"type":54,"value":1997},{"type":48,"tag":513,"props":2335,"children":2336},{},[2337],{"type":54,"value":2002},{"type":48,"tag":476,"props":2339,"children":2340},{},[2341,2345,2353,2361],{"type":48,"tag":513,"props":2342,"children":2343},{},[2344],{"type":54,"value":124},{"type":48,"tag":513,"props":2346,"children":2347},{},[2348],{"type":48,"tag":68,"props":2349,"children":2351},{"className":2350},[],[2352],{"type":54,"value":2046},{"type":48,"tag":513,"props":2354,"children":2355},{},[2356],{"type":48,"tag":68,"props":2357,"children":2359},{"className":2358},[],[2360],{"type":54,"value":2055},{"type":48,"tag":513,"props":2362,"children":2363},{},[2364],{"type":54,"value":2060},{"type":48,"tag":476,"props":2366,"children":2367},{},[2368,2372,2380,2388],{"type":48,"tag":513,"props":2369,"children":2370},{},[2371],{"type":54,"value":94},{"type":48,"tag":513,"props":2373,"children":2374},{},[2375],{"type":48,"tag":68,"props":2376,"children":2378},{"className":2377},[],[2379],{"type":54,"value":1958},{"type":48,"tag":513,"props":2381,"children":2382},{},[2383],{"type":48,"tag":68,"props":2384,"children":2386},{"className":2385},[],[2387],{"type":54,"value":1967},{"type":48,"tag":513,"props":2389,"children":2390},{},[2391],{"type":54,"value":1972},{"type":48,"tag":476,"props":2393,"children":2394},{},[2395,2399,2407,2416],{"type":48,"tag":513,"props":2396,"children":2397},{},[2398],{"type":54,"value":94},{"type":48,"tag":513,"props":2400,"children":2401},{},[2402],{"type":48,"tag":68,"props":2403,"children":2405},{"className":2404},[],[2406],{"type":54,"value":73},{"type":48,"tag":513,"props":2408,"children":2409},{},[2410],{"type":48,"tag":68,"props":2411,"children":2413},{"className":2412},[],[2414],{"type":54,"value":2415},"ptm_if_no_resume_model",{"type":48,"tag":513,"props":2417,"children":2418},{},[2419],{"type":54,"value":2420},"PTM when no resume checkpoint exists",{"type":48,"tag":476,"props":2422,"children":2423},{},[2424,2428,2436,2444],{"type":48,"tag":513,"props":2425,"children":2426},{},[2427],{"type":54,"value":94},{"type":48,"tag":513,"props":2429,"children":2430},{},[2431],{"type":48,"tag":68,"props":2432,"children":2434},{"className":2433},[],[2435],{"type":54,"value":2046},{"type":48,"tag":513,"props":2437,"children":2438},{},[2439],{"type":48,"tag":68,"props":2440,"children":2442},{"className":2441},[],[2443],{"type":54,"value":2055},{"type":48,"tag":513,"props":2445,"children":2446},{},[2447],{"type":54,"value":2060},{"type":48,"tag":476,"props":2449,"children":2450},{},[2451,2455,2463,2471],{"type":48,"tag":513,"props":2452,"children":2453},{},[2454],{"type":54,"value":94},{"type":48,"tag":513,"props":2456,"children":2457},{},[2458],{"type":48,"tag":68,"props":2459,"children":2461},{"className":2460},[],[2462],{"type":54,"value":81},{"type":48,"tag":513,"props":2464,"children":2465},{},[2466],{"type":48,"tag":68,"props":2467,"children":2469},{"className":2468},[],[2470],{"type":54,"value":2415},{"type":48,"tag":513,"props":2472,"children":2473},{},[2474],{"type":54,"value":2475},"full model PTM when no resume checkpoint exists",{"type":48,"tag":476,"props":2477,"children":2478},{},[2479,2483,2491,2500],{"type":48,"tag":513,"props":2480,"children":2481},{},[2482],{"type":54,"value":94},{"type":48,"tag":513,"props":2484,"children":2485},{},[2486],{"type":48,"tag":68,"props":2487,"children":2489},{"className":2488},[],[2490],{"type":54,"value":1404},{"type":48,"tag":513,"props":2492,"children":2493},{},[2494],{"type":48,"tag":68,"props":2495,"children":2497},{"className":2496},[],[2498],{"type":54,"value":2499},"resume_model",{"type":48,"tag":513,"props":2501,"children":2502},{},[2503],{"type":54,"value":2504},"model file inferred from the current job results folder",{"type":48,"tag":57,"props":2506,"children":2507},{},[2508,2510,2515,2516,2522,2524,2530,2532,2537],{"type":54,"value":2509},"For ",{"type":48,"tag":68,"props":2511,"children":2513},{"className":2512},[],[2514],{"type":54,"value":1997},{"type":54,"value":1640},{"type":48,"tag":68,"props":2517,"children":2519},{"className":2518},[],[2520],{"type":54,"value":2521},"parent_model_folder",{"type":54,"value":2523},", pass the upstream train\u002Fexport\u002FAutoML child job id as ",{"type":48,"tag":68,"props":2525,"children":2527},{"className":2526},[],[2528],{"type":54,"value":2529},"parent_job_id",{"type":54,"value":2531},". 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":48,"tag":68,"props":2533,"children":2535},{"className":2534},[],[2536],{"type":54,"value":1880},{"type":54,"value":2538}," and do not patch generated runner scripts to guess checkpoint paths.",{"type":48,"tag":167,"props":2540,"children":2542},{"id":2541},"deployment",[2543],{"type":54,"value":2544},"Deployment",{"type":48,"tag":417,"props":2546,"children":2547},{},[2548],{"type":48,"tag":421,"props":2549,"children":2550},{},[2551],{"type":48,"tag":2552,"props":2553,"children":2555},"a",{"href":2554},"references\u002Ftao-deploy-deformable-detr.md",[2556],{"type":54,"value":2557},"tao-deploy-deformable-detr",{"type":48,"tag":2559,"props":2560,"children":2561},"style",{},[2562],{"type":54,"value":2563},"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":2565,"total":2667},[2566,2583,2595,2609,2621,2638,2653],{"slug":2567,"name":2567,"fn":2568,"description":2569,"org":2570,"tags":2571,"stars":23,"repoUrl":24,"updatedAt":2582},"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},[2572,2575,2578,2579],{"name":2573,"slug":2574,"type":15},"Data Analysis","data-analysis",{"name":2576,"slug":2577,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2580,"slug":2581,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2584,"name":2584,"fn":2585,"description":2586,"org":2587,"tags":2588,"stars":23,"repoUrl":24,"updatedAt":2594},"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},[2589,2590,2593],{"name":2544,"slug":2541,"type":15},{"name":2591,"slug":2592,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2596,"name":2596,"fn":2597,"description":2598,"org":2599,"tags":2600,"stars":23,"repoUrl":24,"updatedAt":2608},"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},[2601,2604,2605],{"name":2602,"slug":2603,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2606,"slug":2607,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2610,"name":2610,"fn":2611,"description":2612,"org":2613,"tags":2614,"stars":23,"repoUrl":24,"updatedAt":2620},"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},[2615,2616,2617],{"name":2573,"slug":2574,"type":15},{"name":9,"slug":8,"type":15},{"name":2618,"slug":2619,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":2622,"name":2622,"fn":2623,"description":2624,"org":2625,"tags":2626,"stars":23,"repoUrl":24,"updatedAt":2637},"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},[2627,2630,2633,2634],{"name":2628,"slug":2629,"type":15},"Automation","automation",{"name":2631,"slug":2632,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":2635,"slug":2636,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2639,"name":2639,"fn":2640,"description":2641,"org":2642,"tags":2643,"stars":23,"repoUrl":24,"updatedAt":2652},"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},[2644,2645,2648,2649],{"name":2544,"slug":2541,"type":15},{"name":2646,"slug":2647,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2650,"slug":2651,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2654,"name":2654,"fn":2655,"description":2656,"org":2657,"tags":2658,"stars":23,"repoUrl":24,"updatedAt":2666},"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},[2659,2660,2663],{"name":9,"slug":8,"type":15},{"name":2661,"slug":2662,"type":15},"Quantum Computing","quantum-computing",{"name":2664,"slug":2665,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":2669,"total":2817},[2670,2688,2703,2714,2726,2740,2753,2765,2776,2785,2799,2808],{"slug":2671,"name":2671,"fn":2672,"description":2673,"org":2674,"tags":2675,"stars":2685,"repoUrl":2686,"updatedAt":2687},"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},[2676,2679,2682],{"name":2677,"slug":2678,"type":15},"Documentation","documentation",{"name":2680,"slug":2681,"type":15},"MCP","mcp",{"name":2683,"slug":2684,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":2689,"name":2689,"fn":2690,"description":2691,"org":2692,"tags":2693,"stars":2700,"repoUrl":2701,"updatedAt":2702},"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},[2694,2697,2698],{"name":2695,"slug":2696,"type":15},"Containers","containers",{"name":2544,"slug":2541,"type":15},{"name":2699,"slug":810,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":2704,"name":2704,"fn":2705,"description":2706,"org":2707,"tags":2708,"stars":2700,"repoUrl":2701,"updatedAt":2713},"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},[2709,2712],{"name":2710,"slug":2711,"type":15},"CI\u002FCD","ci-cd",{"name":2544,"slug":2541,"type":15},"2026-07-14T05:25:59.97109",{"slug":2715,"name":2715,"fn":2716,"description":2717,"org":2718,"tags":2719,"stars":2700,"repoUrl":2701,"updatedAt":2725},"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},[2720,2721,2722],{"name":2710,"slug":2711,"type":15},{"name":2544,"slug":2541,"type":15},{"name":2723,"slug":2724,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":2727,"name":2727,"fn":2728,"description":2729,"org":2730,"tags":2731,"stars":2700,"repoUrl":2701,"updatedAt":2739},"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},[2732,2735,2736],{"name":2733,"slug":2734,"type":15},"Debugging","debugging",{"name":2723,"slug":2724,"type":15},{"name":2737,"slug":2738,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":2741,"name":2741,"fn":2742,"description":2743,"org":2744,"tags":2745,"stars":2700,"repoUrl":2701,"updatedAt":2752},"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},[2746,2749],{"name":2747,"slug":2748,"type":15},"Best Practices","best-practices",{"name":2750,"slug":2751,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":2754,"name":2754,"fn":2755,"description":2756,"org":2757,"tags":2758,"stars":2700,"repoUrl":2701,"updatedAt":2764},"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},[2759,2760,2763],{"name":17,"slug":18,"type":15},{"name":2761,"slug":2762,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":2766,"name":2766,"fn":2767,"description":2768,"org":2769,"tags":2770,"stars":2700,"repoUrl":2701,"updatedAt":2775},"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},[2771,2774],{"name":2772,"slug":2773,"type":15},"QA","qa",{"name":2618,"slug":2619,"type":15},"2026-07-14T05:25:53.673039",{"slug":2777,"name":2777,"fn":2778,"description":2779,"org":2780,"tags":2781,"stars":2700,"repoUrl":2701,"updatedAt":2784},"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},[2782,2783],{"name":2544,"slug":2541,"type":15},{"name":2591,"slug":2592,"type":15},"2026-07-14T05:25:49.362534",{"slug":2786,"name":2786,"fn":2787,"description":2788,"org":2789,"tags":2790,"stars":2700,"repoUrl":2701,"updatedAt":2798},"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},[2791,2794,2795],{"name":2792,"slug":2793,"type":15},"Code Review","code-review",{"name":2723,"slug":2724,"type":15},{"name":2796,"slug":2797,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":2800,"name":2800,"fn":2801,"description":2802,"org":2803,"tags":2804,"stars":2700,"repoUrl":2701,"updatedAt":2807},"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},[2805,2806],{"name":2772,"slug":2773,"type":15},{"name":2618,"slug":2619,"type":15},"2026-07-14T05:25:54.928983",{"slug":2809,"name":2809,"fn":2810,"description":2811,"org":2812,"tags":2813,"stars":2700,"repoUrl":2701,"updatedAt":2816},"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},[2814,2815],{"name":2628,"slug":2629,"type":15},{"name":2710,"slug":2711,"type":15},"2026-07-30T05:29:03.275638",496]