[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-bevfusion":3,"mdc--6jtp22-key":34,"related-repo-nvidia-tao-train-bevfusion":2214,"related-org-nvidia-tao-train-bevfusion":2320},{"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-bevfusion","train BEVFusion models for 3D object detection","BEVFusion for multi-sensor 3D object detection. Fuses LiDAR point clouds and camera images in bird's-eye-view (BEV) space, used in autonomous driving for robust 3D perception. Use when training, evaluating, or running inference for a TAO BEVFusion model. Trigger phrases include \"train BEVFusion\", \"LiDAR + camera fusion\", \"BEV 3D detection\", \"multi-sensor 3D perception\".",{"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","2026-07-14T05:27:19.253234","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-bevfusion","---\nname: tao-train-bevfusion\ndescription: BEVFusion for multi-sensor 3D object detection. Fuses LiDAR point clouds and camera images in bird's-eye-view\n  (BEV) space, used in autonomous driving for robust 3D perception. Use when training, evaluating, or running inference for\n  a TAO BEVFusion model. Trigger phrases include \"train BEVFusion\", \"LiDAR + camera fusion\", \"BEV 3D detection\", \"multi-sensor\n  3D perception\".\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- multi\n- sensor\n- 3d\n- detection\n---\n\n# BEVFusion\n\nBEVFusion for multi-sensor 3D object detection. Fuses LiDAR point clouds and camera images in bird's-eye-view (BEV) space. Used in autonomous driving for robust 3D perception.\n\nSet pretrained backbone paths for Swin image backbone.\n\nBEVFusion requires the BEVFusion-specific TAO container\n`nvcr.io\u002Fnvidia\u002Ftao\u002Ftao-toolkit:5.5.0-pyt`. The shared TAO PyTorch 7.0 RC image\ndoes not package `mmdet3d` and fails before any BEVFusion action can parse its\nspec. The model-skill action is named `dataset_convert`, but the 5.5 container\nCLI subtask is `bevfusion convert -e \u003Cspec>`.\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:** bevfusion\n- **Formats:** default\n- **Monitoring metric:** AP11\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| dataset_convert | root_dir | id |  | No |\n| evaluate | dataset.test_dataset | train_datasets | ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_val.pkl | No |\n| inference | dataset.root_dir | train_datasets |  | No |\n| inference | dataset.test_dataset | train_datasets | ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_val.pkl | No |\n| train | dataset.train_dataset | train_datasets | ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_train.pkl | No |\n| train | dataset.val_dataset | train_datasets | ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_val.pkl | No |\n| train | dataset.test_dataset | train_datasets | ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_val.pkl | No |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action** — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`.\n\n```python\nDATA_ROOT = \"\u002Fpath\u002Fto\u002Fkitti_root\"\nCONVERTED = DATA_ROOT  # BEVFusion 5.5 writes info pickles into root_dir.\nDATA_PREFIX = {\"pts\": \"training\u002Fvelodyne_reduced\", \"img\": \"training\u002Fimage_2\"}\n```\n\n**dataset_convert (mandatory data sources):**\n```python\n{\n    \"root_dir\": DATA_ROOT,\n    \"results_dir\": DATA_ROOT,\n    \"mode\": \"training\",\n}\n```\n\n**train (mandatory data sources):**\n```python\n{\n    \"train.num_epochs\": 30,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"dataset.root_dir\": DATA_ROOT,\n    \"dataset.train_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_train.pkl\", \"data_prefix\": DATA_PREFIX},\n    \"dataset.val_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n    \"dataset.test_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n}\n```\n\n**evaluate (mandatory data sources):**\n```python\n{\n    \"dataset.root_dir\": DATA_ROOT,\n    \"dataset.test_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n}\n```\n\n**inference (mandatory data sources):**\n```python\n{\n    \"dataset.root_dir\": DATA_ROOT,\n    \"dataset.test_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n}\n```\n## Eval Dataset\n\nOptional. Val dataset split is configured via ann_file in dataset config.\n\n## Important Parameters\n\n- **dataset.classes**: List of detection classes. Default [\"person\"]. Must match the annotation categories.\n- **dataset.type**: Dataset type. Options: KittiPersonDataset, TAO3DSyntheticDataset, TAO3DDataset.\n- **dataset.root_dir**: Root directory of the KITTI-style dataset.\n- **dataset.box_type_3d**: 3D box coordinate frame. Options: lidar, camera. Default lidar.\n- **train.optimizer.lr**: Learning rate. Default 2e-4 (AdamW). Use AmpOptimWrapper for mixed precision via optimizer.wrapper_type.\n- **input_modality**: Dict controlling sensor modalities. Keys: use_lidar (True), use_camera (True), use_radar (False), use_map (False).\n- **model.img_backbone**: Image backbone. Default mmdet.SwinTransformer (Swin-Tiny). embed_dims=96, depths=[2,2,6,2].\n- **model.view_transform.type**: View transform for BEV projection. Options: DepthLSSTransform, LSSTransform. Default DepthLSSTransform.\n- **model.point_cloud_range**: Spatial extent of LiDAR. Default [0,-40,-3,70.4,40,1].\n- **model.voxel_size**: Voxel dimensions. Default [0.05, 0.05, 0.1].\n- **dataset.train_dataset.batch_size**: Per-GPU batch size. Default 4.\n\n## Multi-GPU \u002F Multi-Node\n\n**Launch method:** `torchrun` (LIGHTNING_EXCLUDED_NETWORK). The entrypoint runs `torchrun --nnodes=N --nproc-per-node=M train.py`, NOT plain `python`.\n\n| Spec Key | Description | Default |\n|----------|-------------|---------|\n| `train.num_gpus` | Number of GPUs per node | 1 |\n| `train.gpu_ids` | GPU device indices | [0] |\n| `train.num_nodes` | Number of nodes | 1 |\n\n- `CUDA_VISIBLE_DEVICES` is explicitly set from `TAO_VISIBLE_DEVICES`\n- BEVFusion uses mmdet3d-based distributed training, not Lightning DDP\n- `NODE_RANK` is copied to `RANK` if `RANK` is unset\n\n**Multi-node env vars** (set by orchestrator):\n\n| Variable | Purpose |\n|----------|---------|\n| `WORLD_SIZE` | Number of nodes |\n| `NODE_RANK` | This node's rank |\n| `MASTER_ADDR` | Rank-0 node IP |\n| `MASTER_PORT` | Rank-0 port (default 29500) |\n| `NUM_GPU_PER_NODE` | GPUs per node |\n\n## Hardware\n\nMinimum 2 GPU(s), recommended 4 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. BEVFusion is memory-intensive due to multi-sensor fusion. A100 GPUs strongly recommended. Multi-GPU training expected.\n\n## Error Patterns\n\n**dataset_convert required**: Run the model-skill `dataset_convert` action\n(`bevfusion convert -e \u003Cspec>` in the BEVFusion 5.5 container) before training\nto produce `kitti_person_infos_train.pkl`, `kitti_person_infos_val.pkl`, and\n`training\u002Fvelodyne_reduced`. For direct local-docker 5.5 runs, set\n`results_dir` to the same mounted path as `root_dir`; the converter writes the\ninfo pickles there and later expects them under `root_dir` while reducing point\nclouds.\n\n**KITTI directory names**: The BEVFusion 5.5 converter writes reduced point\nclouds under `training\u002Fvelodyne_reduced` and expects camera images under\n`training\u002Fimage_2`. Do not use the stale `training\u002Flidar_reduced` or\n`training\u002Fimages\u002F` defaults when chaining dataset_convert into train\u002Fevaluate or\ninference.\n\n**BEVFusion 5.5 config surface**: Use the 5.5 dataclass keys in packaged\ntemplates. Remove newer top-level\u002Faction keys such as `model_name`,\n`wandb.group`, `wandb.run_id`, `train.checkpoint_interval_unit`,\n`evaluate.trt_engine`, `evaluate.batch_size`, `inference.trt_engine`, and\n`inference.batch_size`. For train, evaluate, and inference specs, keep the\nnon-running action stubs (`train`, `evaluate`, and `inference`) present with\nempty checkpoint strings where needed; the 5.5 runners materialize the full\nexperiment config before running the selected action. Use YAML null, not an\nempty string, for `train.pretrained_checkpoint` and\n`train.resume_training_checkpoint_path` when no checkpoint is intended.\n\n**`ModuleNotFoundError: No module named 'mmdet3d'`**: The shared TAO PyTorch\n7.0 RC image does not include the BEVFusion `mmdet3d` dependency. Use\n`nvcr.io\u002Fnvidia\u002Ftao\u002Ftao-toolkit:5.5.0-pyt`; it contains `mmdet3d` and exposes\nthe BEVFusion `convert`, `train`, `evaluate`, and `inference` subtasks.\n\n**Post-evaluation SIGSEGV in BEVFusion 5.5**: Some local-docker runs can write\ncheckpoints or prediction files and still finish with TAO `Execution status:\nFAIL` after `Signal 11 (SIGSEGV)` in `cuMemRetainAllocationHandle`. Do not mark\nthe action successful from the Docker exit code alone; inspect the TAO log or\n`status.json`. If a checkpoint was produced before this failure, use only the\nexact intended checkpoint such as `epoch_1.pth` for downstream diagnostics and\ndo not treat `last_checkpoint` as a best checkpoint unless the action explicitly\nrequests the latest checkpoint.\n\n**Missing modality data**: Ensure both camera images and LiDAR point clouds are present if using multi-modal fusion.\n\n**Epoch numbering**: BEVFusion checkpoint epoch numbers may not follow standard zero-padded format.\n\n**Checkpoint handoff**: Use the SDK\u002Fmodel checkpoint resolver for parent-model\nselection. For direct local-docker chaining, inspect the train results and pass\nthe exact intended checkpoint path such as `epoch_1.pth`; use `latest.pth` only\nwhen the user explicitly asks for latest. Resume\u002Fretrain must set\n`train.resume: true` and `train.resume_training_checkpoint_path` to the exact\ncheckpoint being resumed.\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 `bevfusion.config.json`:\n\n| Action | Spec Field | Inference Function | Meaning |\n|---|---|---|---|\n| dataset_convert | `results_dir` | `output_dir` | current job results directory |\n| evaluate | `encryption_key` | `key` | encryption key |\n| evaluate | `evaluate.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| evaluate | `results_dir` | `output_dir` | current job results directory |\n| inference | `encryption_key` | `key` | encryption key |\n| inference | `inference.checkpoint` | `parent_model` | model file inferred from the parent job results folder |\n| inference | `results_dir` | `output_dir` | current job results directory |\n| train | `encryption_key` | `key` | encryption key |\n| train | `results_dir` | `output_dir` | current job results directory |\n| train | `train.pretrained_checkpoint` | `ptm_if_no_resume_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",{"data":35,"body":46},{"name":4,"description":6,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"tags":41},"Requires docker + nvidia-container-toolkit.",{"version":38,"author":39},"0.1.0","NVIDIA Corporation","Read Bash",[42,43,44,45],"multi","sensor","3d","detection",{"type":47,"children":48},"root",[49,58,64,69,107,114,207,213,322,356,362,398,405,618,624,643,683,691,740,748,838,846,881,889,923,929,934,940,1077,1083,1115,1209,1260,1270,1375,1381,1386,1392,1461,1502,1606,1667,1725,1735,1745,1784,1790,1819,1832,2173,2208],{"type":50,"tag":51,"props":52,"children":54},"element","h1",{"id":53},"bevfusion",[55],{"type":56,"value":57},"text","BEVFusion",{"type":50,"tag":59,"props":60,"children":61},"p",{},[62],{"type":56,"value":63},"BEVFusion for multi-sensor 3D object detection. Fuses LiDAR point clouds and camera images in bird's-eye-view (BEV) space. Used in autonomous driving for robust 3D perception.",{"type":50,"tag":59,"props":65,"children":66},{},[67],{"type":56,"value":68},"Set pretrained backbone paths for Swin image backbone.",{"type":50,"tag":59,"props":70,"children":71},{},[72,74,81,83,89,91,97,99,105],{"type":56,"value":73},"BEVFusion requires the BEVFusion-specific TAO container\n",{"type":50,"tag":75,"props":76,"children":78},"code",{"className":77},[],[79],{"type":56,"value":80},"nvcr.io\u002Fnvidia\u002Ftao\u002Ftao-toolkit:5.5.0-pyt",{"type":56,"value":82},". The shared TAO PyTorch 7.0 RC image\ndoes not package ",{"type":50,"tag":75,"props":84,"children":86},{"className":85},[],[87],{"type":56,"value":88},"mmdet3d",{"type":56,"value":90}," and fails before any BEVFusion action can parse its\nspec. The model-skill action is named ",{"type":50,"tag":75,"props":92,"children":94},{"className":93},[],[95],{"type":56,"value":96},"dataset_convert",{"type":56,"value":98},", but the 5.5 container\nCLI subtask is ",{"type":50,"tag":75,"props":100,"children":102},{"className":101},[],[103],{"type":56,"value":104},"bevfusion convert -e \u003Cspec>",{"type":56,"value":106},".",{"type":50,"tag":108,"props":109,"children":111},"h2",{"id":110},"dataclass-schemas",[112],{"type":56,"value":113},"Dataclass Schemas",{"type":50,"tag":59,"props":115,"children":116},{},[117,119,125,127,133,135,141,143,149,151,157,159,165,167,173,175,181,183,189,191,197,199,205],{"type":56,"value":118},"Generated TAO Core schemas are packaged in ",{"type":50,"tag":75,"props":120,"children":122},{"className":121},[],[123],{"type":56,"value":124},"schemas\u002F\u003Caction>.schema.json",{"type":56,"value":126},", with ",{"type":50,"tag":75,"props":128,"children":130},{"className":129},[],[131],{"type":56,"value":132},"schemas\u002Fmanifest.json",{"type":56,"value":134}," listing available actions. Each generated schema also emits ",{"type":50,"tag":75,"props":136,"children":138},{"className":137},[],[139],{"type":56,"value":140},"references\u002Fspec_template_\u003Caction>.yaml",{"type":56,"value":142}," from the schema top-level ",{"type":50,"tag":75,"props":144,"children":146},{"className":145},[],[147],{"type":56,"value":148},"default",{"type":56,"value":150}," field. AutoML enablement is declared at the model layer in ",{"type":50,"tag":75,"props":152,"children":154},{"className":153},[],[155],{"type":56,"value":156},"references\u002Fskill_info.yaml",{"type":56,"value":158}," via ",{"type":50,"tag":75,"props":160,"children":162},{"className":161},[],[163],{"type":56,"value":164},"automl_enabled",{"type":56,"value":166},". Runnable AutoML still requires ",{"type":50,"tag":75,"props":168,"children":170},{"className":169},[],[171],{"type":56,"value":172},"schemas\u002Ftrain.schema.json",{"type":56,"value":174}," and ",{"type":50,"tag":75,"props":176,"children":178},{"className":177},[],[179],{"type":56,"value":180},"references\u002Fspec_template_train.yaml",{"type":56,"value":182}," to exist and parse. Use the packaged train schema for ",{"type":50,"tag":75,"props":184,"children":186},{"className":185},[],[187],{"type":56,"value":188},"automl_default_parameters",{"type":56,"value":190},", ",{"type":50,"tag":75,"props":192,"children":194},{"className":193},[],[195],{"type":56,"value":196},"automl_disabled_parameters",{"type":56,"value":198},", defaults, min\u002Fmax bounds, enums, option weights, math conditions, dependencies, and popular parameters. Do not expect ",{"type":50,"tag":75,"props":200,"children":202},{"className":201},[],[203],{"type":56,"value":204},"~\u002Ftao-core",{"type":56,"value":206}," at runtime; maintainers regenerate schemas\u002Ftemplates before packaging the skill bank.",{"type":50,"tag":108,"props":208,"children":210},{"id":209},"train-action-policy",[211],{"type":56,"value":212},"Train Action Policy",{"type":50,"tag":59,"props":214,"children":215},{},[216,218,223,225,231,233,239,241,247,249,255,257,263,265,270,271,277,279,284,285,290,292,298,300,306,308,313,315,320],{"type":56,"value":217},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":50,"tag":75,"props":219,"children":221},{"className":220},[],[222],{"type":56,"value":156},{"type":56,"value":224}," and resolve the run override from either an explicit ",{"type":50,"tag":75,"props":226,"children":228},{"className":227},[],[229],{"type":56,"value":230},"automl_policy",{"type":56,"value":232}," value or the user's workflow request. Use ",{"type":50,"tag":75,"props":234,"children":236},{"className":235},[],[237],{"type":56,"value":238},"automl_policy: on",{"type":56,"value":240}," by default and only expose ",{"type":50,"tag":75,"props":242,"children":244},{"className":243},[],[245],{"type":56,"value":246},"on",{"type":56,"value":248}," \u002F ",{"type":50,"tag":75,"props":250,"children":252},{"className":251},[],[253],{"type":56,"value":254},"off",{"type":56,"value":256}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":50,"tag":75,"props":258,"children":260},{"className":259},[],[261],{"type":56,"value":262},"automl_policy: off",{"type":56,"value":264}," for this run only. When ",{"type":50,"tag":75,"props":266,"children":268},{"className":267},[],[269],{"type":56,"value":238},{"type":56,"value":190},{"type":50,"tag":75,"props":272,"children":274},{"className":273},[],[275],{"type":56,"value":276},"automl_enabled: true",{"type":56,"value":278},", and both ",{"type":50,"tag":75,"props":280,"children":282},{"className":281},[],[283],{"type":56,"value":172},{"type":56,"value":174},{"type":50,"tag":75,"props":286,"children":288},{"className":287},[],[289],{"type":56,"value":180},{"type":56,"value":291}," are packaged, route the train action through ",{"type":50,"tag":75,"props":293,"children":295},{"className":294},[],[296],{"type":56,"value":297},"tao-skill-bank:tao-run-automl",{"type":56,"value":299}," by default with this model's ",{"type":50,"tag":75,"props":301,"children":303},{"className":302},[],[304],{"type":56,"value":305},"skill_dir",{"type":56,"value":307},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":50,"tag":75,"props":309,"children":311},{"className":310},[],[312],{"type":56,"value":230},{"type":56,"value":314},". Use direct model training only when ",{"type":50,"tag":75,"props":316,"children":318},{"className":317},[],[319],{"type":56,"value":262},{"type":56,"value":321}," 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":50,"tag":59,"props":323,"children":324},{},[325,327,333,334,340,341,347,349,354],{"type":56,"value":326},"Non-train actions such as ",{"type":50,"tag":75,"props":328,"children":330},{"className":329},[],[331],{"type":56,"value":332},"evaluate",{"type":56,"value":190},{"type":50,"tag":75,"props":335,"children":337},{"className":336},[],[338],{"type":56,"value":339},"inference",{"type":56,"value":190},{"type":50,"tag":75,"props":342,"children":344},{"className":343},[],[345],{"type":56,"value":346},"export",{"type":56,"value":348},", and deploy flows stay in this model skill. The per-run ",{"type":50,"tag":75,"props":350,"children":352},{"className":351},[],[353],{"type":56,"value":230},{"type":56,"value":355}," override does not change model metadata.",{"type":50,"tag":108,"props":357,"children":359},{"id":358},"training-requirements",[360],{"type":56,"value":361},"Training Requirements",{"type":50,"tag":363,"props":364,"children":365},"ul",{},[366,378,388],{"type":50,"tag":367,"props":368,"children":369},"li",{},[370,376],{"type":50,"tag":371,"props":372,"children":373},"strong",{},[374],{"type":56,"value":375},"Dataset type:",{"type":56,"value":377}," bevfusion",{"type":50,"tag":367,"props":379,"children":380},{},[381,386],{"type":50,"tag":371,"props":382,"children":383},{},[384],{"type":56,"value":385},"Formats:",{"type":56,"value":387}," default",{"type":50,"tag":367,"props":389,"children":390},{},[391,396],{"type":50,"tag":371,"props":392,"children":393},{},[394],{"type":56,"value":395},"Monitoring metric:",{"type":56,"value":397}," AP11",{"type":50,"tag":399,"props":400,"children":402},"h3",{"id":401},"per-action-dataset-requirements",[403],{"type":56,"value":404},"Per-Action Dataset Requirements",{"type":50,"tag":406,"props":407,"children":408},"table",{},[409,443],{"type":50,"tag":410,"props":411,"children":412},"thead",{},[413],{"type":50,"tag":414,"props":415,"children":416},"tr",{},[417,423,428,433,438],{"type":50,"tag":418,"props":419,"children":420},"th",{},[421],{"type":56,"value":422},"Action",{"type":50,"tag":418,"props":424,"children":425},{},[426],{"type":56,"value":427},"Spec Key",{"type":50,"tag":418,"props":429,"children":430},{},[431],{"type":56,"value":432},"Source",{"type":50,"tag":418,"props":434,"children":435},{},[436],{"type":56,"value":437},"Files",{"type":50,"tag":418,"props":439,"children":440},{},[441],{"type":56,"value":442},"List?",{"type":50,"tag":444,"props":445,"children":446},"tbody",{},[447,473,499,522,545,571,595],{"type":50,"tag":414,"props":448,"children":449},{},[450,455,460,465,468],{"type":50,"tag":451,"props":452,"children":453},"td",{},[454],{"type":56,"value":96},{"type":50,"tag":451,"props":456,"children":457},{},[458],{"type":56,"value":459},"root_dir",{"type":50,"tag":451,"props":461,"children":462},{},[463],{"type":56,"value":464},"id",{"type":50,"tag":451,"props":466,"children":467},{},[],{"type":50,"tag":451,"props":469,"children":470},{},[471],{"type":56,"value":472},"No",{"type":50,"tag":414,"props":474,"children":475},{},[476,480,485,490,495],{"type":50,"tag":451,"props":477,"children":478},{},[479],{"type":56,"value":332},{"type":50,"tag":451,"props":481,"children":482},{},[483],{"type":56,"value":484},"dataset.test_dataset",{"type":50,"tag":451,"props":486,"children":487},{},[488],{"type":56,"value":489},"train_datasets",{"type":50,"tag":451,"props":491,"children":492},{},[493],{"type":56,"value":494},"ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_val.pkl",{"type":50,"tag":451,"props":496,"children":497},{},[498],{"type":56,"value":472},{"type":50,"tag":414,"props":500,"children":501},{},[502,506,511,515,518],{"type":50,"tag":451,"props":503,"children":504},{},[505],{"type":56,"value":339},{"type":50,"tag":451,"props":507,"children":508},{},[509],{"type":56,"value":510},"dataset.root_dir",{"type":50,"tag":451,"props":512,"children":513},{},[514],{"type":56,"value":489},{"type":50,"tag":451,"props":516,"children":517},{},[],{"type":50,"tag":451,"props":519,"children":520},{},[521],{"type":56,"value":472},{"type":50,"tag":414,"props":523,"children":524},{},[525,529,533,537,541],{"type":50,"tag":451,"props":526,"children":527},{},[528],{"type":56,"value":339},{"type":50,"tag":451,"props":530,"children":531},{},[532],{"type":56,"value":484},{"type":50,"tag":451,"props":534,"children":535},{},[536],{"type":56,"value":489},{"type":50,"tag":451,"props":538,"children":539},{},[540],{"type":56,"value":494},{"type":50,"tag":451,"props":542,"children":543},{},[544],{"type":56,"value":472},{"type":50,"tag":414,"props":546,"children":547},{},[548,553,558,562,567],{"type":50,"tag":451,"props":549,"children":550},{},[551],{"type":56,"value":552},"train",{"type":50,"tag":451,"props":554,"children":555},{},[556],{"type":56,"value":557},"dataset.train_dataset",{"type":50,"tag":451,"props":559,"children":560},{},[561],{"type":56,"value":489},{"type":50,"tag":451,"props":563,"children":564},{},[565],{"type":56,"value":566},"ann_file: results\u002F{dataset_convert_job_id}\u002Fkitti_person_infos_train.pkl",{"type":50,"tag":451,"props":568,"children":569},{},[570],{"type":56,"value":472},{"type":50,"tag":414,"props":572,"children":573},{},[574,578,583,587,591],{"type":50,"tag":451,"props":575,"children":576},{},[577],{"type":56,"value":552},{"type":50,"tag":451,"props":579,"children":580},{},[581],{"type":56,"value":582},"dataset.val_dataset",{"type":50,"tag":451,"props":584,"children":585},{},[586],{"type":56,"value":489},{"type":50,"tag":451,"props":588,"children":589},{},[590],{"type":56,"value":494},{"type":50,"tag":451,"props":592,"children":593},{},[594],{"type":56,"value":472},{"type":50,"tag":414,"props":596,"children":597},{},[598,602,606,610,614],{"type":50,"tag":451,"props":599,"children":600},{},[601],{"type":56,"value":552},{"type":50,"tag":451,"props":603,"children":604},{},[605],{"type":56,"value":484},{"type":50,"tag":451,"props":607,"children":608},{},[609],{"type":56,"value":489},{"type":50,"tag":451,"props":611,"children":612},{},[613],{"type":56,"value":494},{"type":50,"tag":451,"props":615,"children":616},{},[617],{"type":56,"value":472},{"type":50,"tag":399,"props":619,"children":621},{"id":620},"typical-spec-overrides",[622],{"type":56,"value":623},"Typical Spec Overrides",{"type":50,"tag":59,"props":625,"children":626},{},[627,629,634,636,642],{"type":56,"value":628},"Data source overrides are ",{"type":50,"tag":371,"props":630,"children":631},{},[632],{"type":56,"value":633},"mandatory for every action",{"type":56,"value":635}," — the agent MUST construct data source paths from the Per-Action Dataset Requirements table above and include them in ",{"type":50,"tag":75,"props":637,"children":639},{"className":638},[],[640],{"type":56,"value":641},"spec_overrides",{"type":56,"value":106},{"type":50,"tag":644,"props":645,"children":650},"pre",{"className":646,"code":647,"language":648,"meta":649,"style":649},"language-python shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","DATA_ROOT = \"\u002Fpath\u002Fto\u002Fkitti_root\"\nCONVERTED = DATA_ROOT  # BEVFusion 5.5 writes info pickles into root_dir.\nDATA_PREFIX = {\"pts\": \"training\u002Fvelodyne_reduced\", \"img\": \"training\u002Fimage_2\"}\n","python","",[651],{"type":50,"tag":75,"props":652,"children":653},{"__ignoreMap":649},[654,665,674],{"type":50,"tag":655,"props":656,"children":659},"span",{"class":657,"line":658},"line",1,[660],{"type":50,"tag":655,"props":661,"children":662},{},[663],{"type":56,"value":664},"DATA_ROOT = \"\u002Fpath\u002Fto\u002Fkitti_root\"\n",{"type":50,"tag":655,"props":666,"children":668},{"class":657,"line":667},2,[669],{"type":50,"tag":655,"props":670,"children":671},{},[672],{"type":56,"value":673},"CONVERTED = DATA_ROOT  # BEVFusion 5.5 writes info pickles into root_dir.\n",{"type":50,"tag":655,"props":675,"children":677},{"class":657,"line":676},3,[678],{"type":50,"tag":655,"props":679,"children":680},{},[681],{"type":56,"value":682},"DATA_PREFIX = {\"pts\": \"training\u002Fvelodyne_reduced\", \"img\": \"training\u002Fimage_2\"}\n",{"type":50,"tag":59,"props":684,"children":685},{},[686],{"type":50,"tag":371,"props":687,"children":688},{},[689],{"type":56,"value":690},"dataset_convert (mandatory data sources):",{"type":50,"tag":644,"props":692,"children":694},{"className":646,"code":693,"language":648,"meta":649,"style":649},"{\n    \"root_dir\": DATA_ROOT,\n    \"results_dir\": DATA_ROOT,\n    \"mode\": \"training\",\n}\n",[695],{"type":50,"tag":75,"props":696,"children":697},{"__ignoreMap":649},[698,706,714,722,731],{"type":50,"tag":655,"props":699,"children":700},{"class":657,"line":658},[701],{"type":50,"tag":655,"props":702,"children":703},{},[704],{"type":56,"value":705},"{\n",{"type":50,"tag":655,"props":707,"children":708},{"class":657,"line":667},[709],{"type":50,"tag":655,"props":710,"children":711},{},[712],{"type":56,"value":713},"    \"root_dir\": DATA_ROOT,\n",{"type":50,"tag":655,"props":715,"children":716},{"class":657,"line":676},[717],{"type":50,"tag":655,"props":718,"children":719},{},[720],{"type":56,"value":721},"    \"results_dir\": DATA_ROOT,\n",{"type":50,"tag":655,"props":723,"children":725},{"class":657,"line":724},4,[726],{"type":50,"tag":655,"props":727,"children":728},{},[729],{"type":56,"value":730},"    \"mode\": \"training\",\n",{"type":50,"tag":655,"props":732,"children":734},{"class":657,"line":733},5,[735],{"type":50,"tag":655,"props":736,"children":737},{},[738],{"type":56,"value":739},"}\n",{"type":50,"tag":59,"props":741,"children":742},{},[743],{"type":50,"tag":371,"props":744,"children":745},{},[746],{"type":56,"value":747},"train (mandatory data sources):",{"type":50,"tag":644,"props":749,"children":751},{"className":646,"code":750,"language":648,"meta":649,"style":649},"{\n    \"train.num_epochs\": 30,\n    \"train.checkpoint_interval\": 10,\n    \"train.validation_interval\": 10,\n    \"train.num_gpus\": 1,\n    \"dataset.root_dir\": DATA_ROOT,\n    \"dataset.train_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_train.pkl\", \"data_prefix\": DATA_PREFIX},\n    \"dataset.val_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n    \"dataset.test_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n}\n",[752],{"type":50,"tag":75,"props":753,"children":754},{"__ignoreMap":649},[755,762,770,778,786,794,803,812,821,830],{"type":50,"tag":655,"props":756,"children":757},{"class":657,"line":658},[758],{"type":50,"tag":655,"props":759,"children":760},{},[761],{"type":56,"value":705},{"type":50,"tag":655,"props":763,"children":764},{"class":657,"line":667},[765],{"type":50,"tag":655,"props":766,"children":767},{},[768],{"type":56,"value":769},"    \"train.num_epochs\": 30,\n",{"type":50,"tag":655,"props":771,"children":772},{"class":657,"line":676},[773],{"type":50,"tag":655,"props":774,"children":775},{},[776],{"type":56,"value":777},"    \"train.checkpoint_interval\": 10,\n",{"type":50,"tag":655,"props":779,"children":780},{"class":657,"line":724},[781],{"type":50,"tag":655,"props":782,"children":783},{},[784],{"type":56,"value":785},"    \"train.validation_interval\": 10,\n",{"type":50,"tag":655,"props":787,"children":788},{"class":657,"line":733},[789],{"type":50,"tag":655,"props":790,"children":791},{},[792],{"type":56,"value":793},"    \"train.num_gpus\": 1,\n",{"type":50,"tag":655,"props":795,"children":797},{"class":657,"line":796},6,[798],{"type":50,"tag":655,"props":799,"children":800},{},[801],{"type":56,"value":802},"    \"dataset.root_dir\": DATA_ROOT,\n",{"type":50,"tag":655,"props":804,"children":806},{"class":657,"line":805},7,[807],{"type":50,"tag":655,"props":808,"children":809},{},[810],{"type":56,"value":811},"    \"dataset.train_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_train.pkl\", \"data_prefix\": DATA_PREFIX},\n",{"type":50,"tag":655,"props":813,"children":815},{"class":657,"line":814},8,[816],{"type":50,"tag":655,"props":817,"children":818},{},[819],{"type":56,"value":820},"    \"dataset.val_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n",{"type":50,"tag":655,"props":822,"children":824},{"class":657,"line":823},9,[825],{"type":50,"tag":655,"props":826,"children":827},{},[828],{"type":56,"value":829},"    \"dataset.test_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n",{"type":50,"tag":655,"props":831,"children":833},{"class":657,"line":832},10,[834],{"type":50,"tag":655,"props":835,"children":836},{},[837],{"type":56,"value":739},{"type":50,"tag":59,"props":839,"children":840},{},[841],{"type":50,"tag":371,"props":842,"children":843},{},[844],{"type":56,"value":845},"evaluate (mandatory data sources):",{"type":50,"tag":644,"props":847,"children":849},{"className":646,"code":848,"language":648,"meta":649,"style":649},"{\n    \"dataset.root_dir\": DATA_ROOT,\n    \"dataset.test_dataset\": {\"ann_file\": f\"{CONVERTED}\u002Fkitti_person_infos_val.pkl\", \"data_prefix\": DATA_PREFIX},\n}\n",[850],{"type":50,"tag":75,"props":851,"children":852},{"__ignoreMap":649},[853,860,867,874],{"type":50,"tag":655,"props":854,"children":855},{"class":657,"line":658},[856],{"type":50,"tag":655,"props":857,"children":858},{},[859],{"type":56,"value":705},{"type":50,"tag":655,"props":861,"children":862},{"class":657,"line":667},[863],{"type":50,"tag":655,"props":864,"children":865},{},[866],{"type":56,"value":802},{"type":50,"tag":655,"props":868,"children":869},{"class":657,"line":676},[870],{"type":50,"tag":655,"props":871,"children":872},{},[873],{"type":56,"value":829},{"type":50,"tag":655,"props":875,"children":876},{"class":657,"line":724},[877],{"type":50,"tag":655,"props":878,"children":879},{},[880],{"type":56,"value":739},{"type":50,"tag":59,"props":882,"children":883},{},[884],{"type":50,"tag":371,"props":885,"children":886},{},[887],{"type":56,"value":888},"inference (mandatory data sources):",{"type":50,"tag":644,"props":890,"children":891},{"className":646,"code":848,"language":648,"meta":649,"style":649},[892],{"type":50,"tag":75,"props":893,"children":894},{"__ignoreMap":649},[895,902,909,916],{"type":50,"tag":655,"props":896,"children":897},{"class":657,"line":658},[898],{"type":50,"tag":655,"props":899,"children":900},{},[901],{"type":56,"value":705},{"type":50,"tag":655,"props":903,"children":904},{"class":657,"line":667},[905],{"type":50,"tag":655,"props":906,"children":907},{},[908],{"type":56,"value":802},{"type":50,"tag":655,"props":910,"children":911},{"class":657,"line":676},[912],{"type":50,"tag":655,"props":913,"children":914},{},[915],{"type":56,"value":829},{"type":50,"tag":655,"props":917,"children":918},{"class":657,"line":724},[919],{"type":50,"tag":655,"props":920,"children":921},{},[922],{"type":56,"value":739},{"type":50,"tag":108,"props":924,"children":926},{"id":925},"eval-dataset",[927],{"type":56,"value":928},"Eval Dataset",{"type":50,"tag":59,"props":930,"children":931},{},[932],{"type":56,"value":933},"Optional. Val dataset split is configured via ann_file in dataset config.",{"type":50,"tag":108,"props":935,"children":937},{"id":936},"important-parameters",[938],{"type":56,"value":939},"Important Parameters",{"type":50,"tag":363,"props":941,"children":942},{},[943,960,970,979,989,999,1009,1025,1035,1051,1067],{"type":50,"tag":367,"props":944,"children":945},{},[946,951,953,958],{"type":50,"tag":371,"props":947,"children":948},{},[949],{"type":56,"value":950},"dataset.classes",{"type":56,"value":952},": List of detection classes. Default ",{"type":50,"tag":655,"props":954,"children":955},{},[956],{"type":56,"value":957},"\"person\"",{"type":56,"value":959},". Must match the annotation categories.",{"type":50,"tag":367,"props":961,"children":962},{},[963,968],{"type":50,"tag":371,"props":964,"children":965},{},[966],{"type":56,"value":967},"dataset.type",{"type":56,"value":969},": Dataset type. Options: KittiPersonDataset, TAO3DSyntheticDataset, TAO3DDataset.",{"type":50,"tag":367,"props":971,"children":972},{},[973,977],{"type":50,"tag":371,"props":974,"children":975},{},[976],{"type":56,"value":510},{"type":56,"value":978},": Root directory of the KITTI-style dataset.",{"type":50,"tag":367,"props":980,"children":981},{},[982,987],{"type":50,"tag":371,"props":983,"children":984},{},[985],{"type":56,"value":986},"dataset.box_type_3d",{"type":56,"value":988},": 3D box coordinate frame. Options: lidar, camera. Default lidar.",{"type":50,"tag":367,"props":990,"children":991},{},[992,997],{"type":50,"tag":371,"props":993,"children":994},{},[995],{"type":56,"value":996},"train.optimizer.lr",{"type":56,"value":998},": Learning rate. Default 2e-4 (AdamW). Use AmpOptimWrapper for mixed precision via optimizer.wrapper_type.",{"type":50,"tag":367,"props":1000,"children":1001},{},[1002,1007],{"type":50,"tag":371,"props":1003,"children":1004},{},[1005],{"type":56,"value":1006},"input_modality",{"type":56,"value":1008},": Dict controlling sensor modalities. Keys: use_lidar (True), use_camera (True), use_radar (False), use_map (False).",{"type":50,"tag":367,"props":1010,"children":1011},{},[1012,1017,1019,1024],{"type":50,"tag":371,"props":1013,"children":1014},{},[1015],{"type":56,"value":1016},"model.img_backbone",{"type":56,"value":1018},": Image backbone. Default mmdet.SwinTransformer (Swin-Tiny). embed_dims=96, depths=",{"type":50,"tag":655,"props":1020,"children":1021},{},[1022],{"type":56,"value":1023},"2,2,6,2",{"type":56,"value":106},{"type":50,"tag":367,"props":1026,"children":1027},{},[1028,1033],{"type":50,"tag":371,"props":1029,"children":1030},{},[1031],{"type":56,"value":1032},"model.view_transform.type",{"type":56,"value":1034},": View transform for BEV projection. Options: DepthLSSTransform, LSSTransform. Default DepthLSSTransform.",{"type":50,"tag":367,"props":1036,"children":1037},{},[1038,1043,1045,1050],{"type":50,"tag":371,"props":1039,"children":1040},{},[1041],{"type":56,"value":1042},"model.point_cloud_range",{"type":56,"value":1044},": Spatial extent of LiDAR. Default ",{"type":50,"tag":655,"props":1046,"children":1047},{},[1048],{"type":56,"value":1049},"0,-40,-3,70.4,40,1",{"type":56,"value":106},{"type":50,"tag":367,"props":1052,"children":1053},{},[1054,1059,1061,1066],{"type":50,"tag":371,"props":1055,"children":1056},{},[1057],{"type":56,"value":1058},"model.voxel_size",{"type":56,"value":1060},": Voxel dimensions. Default ",{"type":50,"tag":655,"props":1062,"children":1063},{},[1064],{"type":56,"value":1065},"0.05, 0.05, 0.1",{"type":56,"value":106},{"type":50,"tag":367,"props":1068,"children":1069},{},[1070,1075],{"type":50,"tag":371,"props":1071,"children":1072},{},[1073],{"type":56,"value":1074},"dataset.train_dataset.batch_size",{"type":56,"value":1076},": Per-GPU batch size. Default 4.",{"type":50,"tag":108,"props":1078,"children":1080},{"id":1079},"multi-gpu-multi-node",[1081],{"type":56,"value":1082},"Multi-GPU \u002F Multi-Node",{"type":50,"tag":59,"props":1084,"children":1085},{},[1086,1091,1093,1099,1101,1107,1109,1114],{"type":50,"tag":371,"props":1087,"children":1088},{},[1089],{"type":56,"value":1090},"Launch method:",{"type":56,"value":1092}," ",{"type":50,"tag":75,"props":1094,"children":1096},{"className":1095},[],[1097],{"type":56,"value":1098},"torchrun",{"type":56,"value":1100}," (LIGHTNING_EXCLUDED_NETWORK). The entrypoint runs ",{"type":50,"tag":75,"props":1102,"children":1104},{"className":1103},[],[1105],{"type":56,"value":1106},"torchrun --nnodes=N --nproc-per-node=M train.py",{"type":56,"value":1108},", NOT plain ",{"type":50,"tag":75,"props":1110,"children":1112},{"className":1111},[],[1113],{"type":56,"value":648},{"type":56,"value":106},{"type":50,"tag":406,"props":1116,"children":1117},{},[1118,1138],{"type":50,"tag":410,"props":1119,"children":1120},{},[1121],{"type":50,"tag":414,"props":1122,"children":1123},{},[1124,1128,1133],{"type":50,"tag":418,"props":1125,"children":1126},{},[1127],{"type":56,"value":427},{"type":50,"tag":418,"props":1129,"children":1130},{},[1131],{"type":56,"value":1132},"Description",{"type":50,"tag":418,"props":1134,"children":1135},{},[1136],{"type":56,"value":1137},"Default",{"type":50,"tag":444,"props":1139,"children":1140},{},[1141,1163,1188],{"type":50,"tag":414,"props":1142,"children":1143},{},[1144,1153,1158],{"type":50,"tag":451,"props":1145,"children":1146},{},[1147],{"type":50,"tag":75,"props":1148,"children":1150},{"className":1149},[],[1151],{"type":56,"value":1152},"train.num_gpus",{"type":50,"tag":451,"props":1154,"children":1155},{},[1156],{"type":56,"value":1157},"Number of GPUs per node",{"type":50,"tag":451,"props":1159,"children":1160},{},[1161],{"type":56,"value":1162},"1",{"type":50,"tag":414,"props":1164,"children":1165},{},[1166,1175,1180],{"type":50,"tag":451,"props":1167,"children":1168},{},[1169],{"type":50,"tag":75,"props":1170,"children":1172},{"className":1171},[],[1173],{"type":56,"value":1174},"train.gpu_ids",{"type":50,"tag":451,"props":1176,"children":1177},{},[1178],{"type":56,"value":1179},"GPU device indices",{"type":50,"tag":451,"props":1181,"children":1182},{},[1183],{"type":50,"tag":655,"props":1184,"children":1185},{},[1186],{"type":56,"value":1187},"0",{"type":50,"tag":414,"props":1189,"children":1190},{},[1191,1200,1205],{"type":50,"tag":451,"props":1192,"children":1193},{},[1194],{"type":50,"tag":75,"props":1195,"children":1197},{"className":1196},[],[1198],{"type":56,"value":1199},"train.num_nodes",{"type":50,"tag":451,"props":1201,"children":1202},{},[1203],{"type":56,"value":1204},"Number of nodes",{"type":50,"tag":451,"props":1206,"children":1207},{},[1208],{"type":56,"value":1162},{"type":50,"tag":363,"props":1210,"children":1211},{},[1212,1229,1234],{"type":50,"tag":367,"props":1213,"children":1214},{},[1215,1221,1223],{"type":50,"tag":75,"props":1216,"children":1218},{"className":1217},[],[1219],{"type":56,"value":1220},"CUDA_VISIBLE_DEVICES",{"type":56,"value":1222}," is explicitly set from ",{"type":50,"tag":75,"props":1224,"children":1226},{"className":1225},[],[1227],{"type":56,"value":1228},"TAO_VISIBLE_DEVICES",{"type":50,"tag":367,"props":1230,"children":1231},{},[1232],{"type":56,"value":1233},"BEVFusion uses mmdet3d-based distributed training, not Lightning DDP",{"type":50,"tag":367,"props":1235,"children":1236},{},[1237,1243,1245,1251,1253,1258],{"type":50,"tag":75,"props":1238,"children":1240},{"className":1239},[],[1241],{"type":56,"value":1242},"NODE_RANK",{"type":56,"value":1244}," is copied to ",{"type":50,"tag":75,"props":1246,"children":1248},{"className":1247},[],[1249],{"type":56,"value":1250},"RANK",{"type":56,"value":1252}," if ",{"type":50,"tag":75,"props":1254,"children":1256},{"className":1255},[],[1257],{"type":56,"value":1250},{"type":56,"value":1259}," is unset",{"type":50,"tag":59,"props":1261,"children":1262},{},[1263,1268],{"type":50,"tag":371,"props":1264,"children":1265},{},[1266],{"type":56,"value":1267},"Multi-node env vars",{"type":56,"value":1269}," (set by orchestrator):",{"type":50,"tag":406,"props":1271,"children":1272},{},[1273,1289],{"type":50,"tag":410,"props":1274,"children":1275},{},[1276],{"type":50,"tag":414,"props":1277,"children":1278},{},[1279,1284],{"type":50,"tag":418,"props":1280,"children":1281},{},[1282],{"type":56,"value":1283},"Variable",{"type":50,"tag":418,"props":1285,"children":1286},{},[1287],{"type":56,"value":1288},"Purpose",{"type":50,"tag":444,"props":1290,"children":1291},{},[1292,1308,1324,1341,1358],{"type":50,"tag":414,"props":1293,"children":1294},{},[1295,1304],{"type":50,"tag":451,"props":1296,"children":1297},{},[1298],{"type":50,"tag":75,"props":1299,"children":1301},{"className":1300},[],[1302],{"type":56,"value":1303},"WORLD_SIZE",{"type":50,"tag":451,"props":1305,"children":1306},{},[1307],{"type":56,"value":1204},{"type":50,"tag":414,"props":1309,"children":1310},{},[1311,1319],{"type":50,"tag":451,"props":1312,"children":1313},{},[1314],{"type":50,"tag":75,"props":1315,"children":1317},{"className":1316},[],[1318],{"type":56,"value":1242},{"type":50,"tag":451,"props":1320,"children":1321},{},[1322],{"type":56,"value":1323},"This node's rank",{"type":50,"tag":414,"props":1325,"children":1326},{},[1327,1336],{"type":50,"tag":451,"props":1328,"children":1329},{},[1330],{"type":50,"tag":75,"props":1331,"children":1333},{"className":1332},[],[1334],{"type":56,"value":1335},"MASTER_ADDR",{"type":50,"tag":451,"props":1337,"children":1338},{},[1339],{"type":56,"value":1340},"Rank-0 node IP",{"type":50,"tag":414,"props":1342,"children":1343},{},[1344,1353],{"type":50,"tag":451,"props":1345,"children":1346},{},[1347],{"type":50,"tag":75,"props":1348,"children":1350},{"className":1349},[],[1351],{"type":56,"value":1352},"MASTER_PORT",{"type":50,"tag":451,"props":1354,"children":1355},{},[1356],{"type":56,"value":1357},"Rank-0 port (default 29500)",{"type":50,"tag":414,"props":1359,"children":1360},{},[1361,1370],{"type":50,"tag":451,"props":1362,"children":1363},{},[1364],{"type":50,"tag":75,"props":1365,"children":1367},{"className":1366},[],[1368],{"type":56,"value":1369},"NUM_GPU_PER_NODE",{"type":50,"tag":451,"props":1371,"children":1372},{},[1373],{"type":56,"value":1374},"GPUs per node",{"type":50,"tag":108,"props":1376,"children":1378},{"id":1377},"hardware",[1379],{"type":56,"value":1380},"Hardware",{"type":50,"tag":59,"props":1382,"children":1383},{},[1384],{"type":56,"value":1385},"Minimum 2 GPU(s), recommended 4 GPU(s). 24GB+ (A100 recommended) VRAM per GPU. BEVFusion is memory-intensive due to multi-sensor fusion. A100 GPUs strongly recommended. Multi-GPU training expected.",{"type":50,"tag":108,"props":1387,"children":1389},{"id":1388},"error-patterns",[1390],{"type":56,"value":1391},"Error Patterns",{"type":50,"tag":59,"props":1393,"children":1394},{},[1395,1400,1402,1407,1409,1414,1416,1422,1423,1429,1431,1437,1439,1445,1447,1452,1454,1459],{"type":50,"tag":371,"props":1396,"children":1397},{},[1398],{"type":56,"value":1399},"dataset_convert required",{"type":56,"value":1401},": Run the model-skill ",{"type":50,"tag":75,"props":1403,"children":1405},{"className":1404},[],[1406],{"type":56,"value":96},{"type":56,"value":1408}," action\n(",{"type":50,"tag":75,"props":1410,"children":1412},{"className":1411},[],[1413],{"type":56,"value":104},{"type":56,"value":1415}," in the BEVFusion 5.5 container) before training\nto produce ",{"type":50,"tag":75,"props":1417,"children":1419},{"className":1418},[],[1420],{"type":56,"value":1421},"kitti_person_infos_train.pkl",{"type":56,"value":190},{"type":50,"tag":75,"props":1424,"children":1426},{"className":1425},[],[1427],{"type":56,"value":1428},"kitti_person_infos_val.pkl",{"type":56,"value":1430},", and\n",{"type":50,"tag":75,"props":1432,"children":1434},{"className":1433},[],[1435],{"type":56,"value":1436},"training\u002Fvelodyne_reduced",{"type":56,"value":1438},". For direct local-docker 5.5 runs, set\n",{"type":50,"tag":75,"props":1440,"children":1442},{"className":1441},[],[1443],{"type":56,"value":1444},"results_dir",{"type":56,"value":1446}," to the same mounted path as ",{"type":50,"tag":75,"props":1448,"children":1450},{"className":1449},[],[1451],{"type":56,"value":459},{"type":56,"value":1453},"; the converter writes the\ninfo pickles there and later expects them under ",{"type":50,"tag":75,"props":1455,"children":1457},{"className":1456},[],[1458],{"type":56,"value":459},{"type":56,"value":1460}," while reducing point\nclouds.",{"type":50,"tag":59,"props":1462,"children":1463},{},[1464,1469,1471,1476,1478,1484,1486,1492,1494,1500],{"type":50,"tag":371,"props":1465,"children":1466},{},[1467],{"type":56,"value":1468},"KITTI directory names",{"type":56,"value":1470},": The BEVFusion 5.5 converter writes reduced point\nclouds under ",{"type":50,"tag":75,"props":1472,"children":1474},{"className":1473},[],[1475],{"type":56,"value":1436},{"type":56,"value":1477}," and expects camera images under\n",{"type":50,"tag":75,"props":1479,"children":1481},{"className":1480},[],[1482],{"type":56,"value":1483},"training\u002Fimage_2",{"type":56,"value":1485},". Do not use the stale ",{"type":50,"tag":75,"props":1487,"children":1489},{"className":1488},[],[1490],{"type":56,"value":1491},"training\u002Flidar_reduced",{"type":56,"value":1493}," or\n",{"type":50,"tag":75,"props":1495,"children":1497},{"className":1496},[],[1498],{"type":56,"value":1499},"training\u002Fimages\u002F",{"type":56,"value":1501}," defaults when chaining dataset_convert into train\u002Fevaluate or\ninference.",{"type":50,"tag":59,"props":1503,"children":1504},{},[1505,1510,1512,1518,1520,1526,1527,1533,1534,1540,1541,1547,1548,1554,1555,1561,1562,1568,1570,1575,1576,1581,1583,1588,1590,1596,1598,1604],{"type":50,"tag":371,"props":1506,"children":1507},{},[1508],{"type":56,"value":1509},"BEVFusion 5.5 config surface",{"type":56,"value":1511},": Use the 5.5 dataclass keys in packaged\ntemplates. Remove newer top-level\u002Faction keys such as ",{"type":50,"tag":75,"props":1513,"children":1515},{"className":1514},[],[1516],{"type":56,"value":1517},"model_name",{"type":56,"value":1519},",\n",{"type":50,"tag":75,"props":1521,"children":1523},{"className":1522},[],[1524],{"type":56,"value":1525},"wandb.group",{"type":56,"value":190},{"type":50,"tag":75,"props":1528,"children":1530},{"className":1529},[],[1531],{"type":56,"value":1532},"wandb.run_id",{"type":56,"value":190},{"type":50,"tag":75,"props":1535,"children":1537},{"className":1536},[],[1538],{"type":56,"value":1539},"train.checkpoint_interval_unit",{"type":56,"value":1519},{"type":50,"tag":75,"props":1542,"children":1544},{"className":1543},[],[1545],{"type":56,"value":1546},"evaluate.trt_engine",{"type":56,"value":190},{"type":50,"tag":75,"props":1549,"children":1551},{"className":1550},[],[1552],{"type":56,"value":1553},"evaluate.batch_size",{"type":56,"value":190},{"type":50,"tag":75,"props":1556,"children":1558},{"className":1557},[],[1559],{"type":56,"value":1560},"inference.trt_engine",{"type":56,"value":1430},{"type":50,"tag":75,"props":1563,"children":1565},{"className":1564},[],[1566],{"type":56,"value":1567},"inference.batch_size",{"type":56,"value":1569},". For train, evaluate, and inference specs, keep the\nnon-running action stubs (",{"type":50,"tag":75,"props":1571,"children":1573},{"className":1572},[],[1574],{"type":56,"value":552},{"type":56,"value":190},{"type":50,"tag":75,"props":1577,"children":1579},{"className":1578},[],[1580],{"type":56,"value":332},{"type":56,"value":1582},", and ",{"type":50,"tag":75,"props":1584,"children":1586},{"className":1585},[],[1587],{"type":56,"value":339},{"type":56,"value":1589},") present with\nempty checkpoint strings where needed; the 5.5 runners materialize the full\nexperiment config before running the selected action. Use YAML null, not an\nempty string, for ",{"type":50,"tag":75,"props":1591,"children":1593},{"className":1592},[],[1594],{"type":56,"value":1595},"train.pretrained_checkpoint",{"type":56,"value":1597}," and\n",{"type":50,"tag":75,"props":1599,"children":1601},{"className":1600},[],[1602],{"type":56,"value":1603},"train.resume_training_checkpoint_path",{"type":56,"value":1605}," when no checkpoint is intended.",{"type":50,"tag":59,"props":1607,"children":1608},{},[1609,1618,1620,1625,1627,1632,1634,1639,1641,1647,1648,1653,1654,1659,1660,1665],{"type":50,"tag":371,"props":1610,"children":1611},{},[1612],{"type":50,"tag":75,"props":1613,"children":1615},{"className":1614},[],[1616],{"type":56,"value":1617},"ModuleNotFoundError: No module named 'mmdet3d'",{"type":56,"value":1619},": The shared TAO PyTorch\n7.0 RC image does not include the BEVFusion ",{"type":50,"tag":75,"props":1621,"children":1623},{"className":1622},[],[1624],{"type":56,"value":88},{"type":56,"value":1626}," dependency. Use\n",{"type":50,"tag":75,"props":1628,"children":1630},{"className":1629},[],[1631],{"type":56,"value":80},{"type":56,"value":1633},"; it contains ",{"type":50,"tag":75,"props":1635,"children":1637},{"className":1636},[],[1638],{"type":56,"value":88},{"type":56,"value":1640}," and exposes\nthe BEVFusion ",{"type":50,"tag":75,"props":1642,"children":1644},{"className":1643},[],[1645],{"type":56,"value":1646},"convert",{"type":56,"value":190},{"type":50,"tag":75,"props":1649,"children":1651},{"className":1650},[],[1652],{"type":56,"value":552},{"type":56,"value":190},{"type":50,"tag":75,"props":1655,"children":1657},{"className":1656},[],[1658],{"type":56,"value":332},{"type":56,"value":1582},{"type":50,"tag":75,"props":1661,"children":1663},{"className":1662},[],[1664],{"type":56,"value":339},{"type":56,"value":1666}," subtasks.",{"type":50,"tag":59,"props":1668,"children":1669},{},[1670,1675,1677,1683,1685,1691,1693,1699,1701,1707,1709,1715,1717,1723],{"type":50,"tag":371,"props":1671,"children":1672},{},[1673],{"type":56,"value":1674},"Post-evaluation SIGSEGV in BEVFusion 5.5",{"type":56,"value":1676},": Some local-docker runs can write\ncheckpoints or prediction files and still finish with TAO ",{"type":50,"tag":75,"props":1678,"children":1680},{"className":1679},[],[1681],{"type":56,"value":1682},"Execution status: FAIL",{"type":56,"value":1684}," after ",{"type":50,"tag":75,"props":1686,"children":1688},{"className":1687},[],[1689],{"type":56,"value":1690},"Signal 11 (SIGSEGV)",{"type":56,"value":1692}," in ",{"type":50,"tag":75,"props":1694,"children":1696},{"className":1695},[],[1697],{"type":56,"value":1698},"cuMemRetainAllocationHandle",{"type":56,"value":1700},". Do not mark\nthe action successful from the Docker exit code alone; inspect the TAO log or\n",{"type":50,"tag":75,"props":1702,"children":1704},{"className":1703},[],[1705],{"type":56,"value":1706},"status.json",{"type":56,"value":1708},". If a checkpoint was produced before this failure, use only the\nexact intended checkpoint such as ",{"type":50,"tag":75,"props":1710,"children":1712},{"className":1711},[],[1713],{"type":56,"value":1714},"epoch_1.pth",{"type":56,"value":1716}," for downstream diagnostics and\ndo not treat ",{"type":50,"tag":75,"props":1718,"children":1720},{"className":1719},[],[1721],{"type":56,"value":1722},"last_checkpoint",{"type":56,"value":1724}," as a best checkpoint unless the action explicitly\nrequests the latest checkpoint.",{"type":50,"tag":59,"props":1726,"children":1727},{},[1728,1733],{"type":50,"tag":371,"props":1729,"children":1730},{},[1731],{"type":56,"value":1732},"Missing modality data",{"type":56,"value":1734},": Ensure both camera images and LiDAR point clouds are present if using multi-modal fusion.",{"type":50,"tag":59,"props":1736,"children":1737},{},[1738,1743],{"type":50,"tag":371,"props":1739,"children":1740},{},[1741],{"type":56,"value":1742},"Epoch numbering",{"type":56,"value":1744},": BEVFusion checkpoint epoch numbers may not follow standard zero-padded format.",{"type":50,"tag":59,"props":1746,"children":1747},{},[1748,1753,1755,1760,1762,1768,1770,1776,1777,1782],{"type":50,"tag":371,"props":1749,"children":1750},{},[1751],{"type":56,"value":1752},"Checkpoint handoff",{"type":56,"value":1754},": Use the SDK\u002Fmodel checkpoint resolver for parent-model\nselection. For direct local-docker chaining, inspect the train results and pass\nthe exact intended checkpoint path such as ",{"type":50,"tag":75,"props":1756,"children":1758},{"className":1757},[],[1759],{"type":56,"value":1714},{"type":56,"value":1761},"; use ",{"type":50,"tag":75,"props":1763,"children":1765},{"className":1764},[],[1766],{"type":56,"value":1767},"latest.pth",{"type":56,"value":1769}," only\nwhen the user explicitly asks for latest. Resume\u002Fretrain must set\n",{"type":50,"tag":75,"props":1771,"children":1773},{"className":1772},[],[1774],{"type":56,"value":1775},"train.resume: true",{"type":56,"value":174},{"type":50,"tag":75,"props":1778,"children":1780},{"className":1779},[],[1781],{"type":56,"value":1603},{"type":56,"value":1783}," to the exact\ncheckpoint being resumed.",{"type":50,"tag":108,"props":1785,"children":1787},{"id":1786},"spec-param-parent-model-inference",[1788],{"type":56,"value":1789},"Spec Param \u002F Parent Model Inference",{"type":50,"tag":59,"props":1791,"children":1792},{},[1793,1795,1801,1803,1809,1811,1817],{"type":56,"value":1794},"Model-specific inference mappings belong in this MD file, not in ",{"type":50,"tag":75,"props":1796,"children":1798},{"className":1797},[],[1799],{"type":56,"value":1800},"config.json",{"type":56,"value":1802},". Generated runners should read this section and apply the mappings with SDK helpers before ",{"type":50,"tag":75,"props":1804,"children":1806},{"className":1805},[],[1807],{"type":56,"value":1808},"create_job()",{"type":56,"value":1810},". This mirrors the old microservices ",{"type":50,"tag":75,"props":1812,"children":1814},{"className":1813},[],[1815],{"type":56,"value":1816},"infer_params.py",{"type":56,"value":1818}," flow.",{"type":50,"tag":59,"props":1820,"children":1821},{},[1822,1824,1830],{"type":56,"value":1823},"Inference mappings from TAO Core ",{"type":50,"tag":75,"props":1825,"children":1827},{"className":1826},[],[1828],{"type":56,"value":1829},"bevfusion.config.json",{"type":56,"value":1831},":",{"type":50,"tag":406,"props":1833,"children":1834},{},[1835,1860],{"type":50,"tag":410,"props":1836,"children":1837},{},[1838],{"type":50,"tag":414,"props":1839,"children":1840},{},[1841,1845,1850,1855],{"type":50,"tag":418,"props":1842,"children":1843},{},[1844],{"type":56,"value":422},{"type":50,"tag":418,"props":1846,"children":1847},{},[1848],{"type":56,"value":1849},"Spec Field",{"type":50,"tag":418,"props":1851,"children":1852},{},[1853],{"type":56,"value":1854},"Inference Function",{"type":50,"tag":418,"props":1856,"children":1857},{},[1858],{"type":56,"value":1859},"Meaning",{"type":50,"tag":444,"props":1861,"children":1862},{},[1863,1892,1922,1952,1979,2006,2034,2061,2088,2115,2144],{"type":50,"tag":414,"props":1864,"children":1865},{},[1866,1870,1878,1887],{"type":50,"tag":451,"props":1867,"children":1868},{},[1869],{"type":56,"value":96},{"type":50,"tag":451,"props":1871,"children":1872},{},[1873],{"type":50,"tag":75,"props":1874,"children":1876},{"className":1875},[],[1877],{"type":56,"value":1444},{"type":50,"tag":451,"props":1879,"children":1880},{},[1881],{"type":50,"tag":75,"props":1882,"children":1884},{"className":1883},[],[1885],{"type":56,"value":1886},"output_dir",{"type":50,"tag":451,"props":1888,"children":1889},{},[1890],{"type":56,"value":1891},"current job results directory",{"type":50,"tag":414,"props":1893,"children":1894},{},[1895,1899,1908,1917],{"type":50,"tag":451,"props":1896,"children":1897},{},[1898],{"type":56,"value":332},{"type":50,"tag":451,"props":1900,"children":1901},{},[1902],{"type":50,"tag":75,"props":1903,"children":1905},{"className":1904},[],[1906],{"type":56,"value":1907},"encryption_key",{"type":50,"tag":451,"props":1909,"children":1910},{},[1911],{"type":50,"tag":75,"props":1912,"children":1914},{"className":1913},[],[1915],{"type":56,"value":1916},"key",{"type":50,"tag":451,"props":1918,"children":1919},{},[1920],{"type":56,"value":1921},"encryption key",{"type":50,"tag":414,"props":1923,"children":1924},{},[1925,1929,1938,1947],{"type":50,"tag":451,"props":1926,"children":1927},{},[1928],{"type":56,"value":332},{"type":50,"tag":451,"props":1930,"children":1931},{},[1932],{"type":50,"tag":75,"props":1933,"children":1935},{"className":1934},[],[1936],{"type":56,"value":1937},"evaluate.checkpoint",{"type":50,"tag":451,"props":1939,"children":1940},{},[1941],{"type":50,"tag":75,"props":1942,"children":1944},{"className":1943},[],[1945],{"type":56,"value":1946},"parent_model",{"type":50,"tag":451,"props":1948,"children":1949},{},[1950],{"type":56,"value":1951},"model file inferred from the parent job results folder",{"type":50,"tag":414,"props":1953,"children":1954},{},[1955,1959,1967,1975],{"type":50,"tag":451,"props":1956,"children":1957},{},[1958],{"type":56,"value":332},{"type":50,"tag":451,"props":1960,"children":1961},{},[1962],{"type":50,"tag":75,"props":1963,"children":1965},{"className":1964},[],[1966],{"type":56,"value":1444},{"type":50,"tag":451,"props":1968,"children":1969},{},[1970],{"type":50,"tag":75,"props":1971,"children":1973},{"className":1972},[],[1974],{"type":56,"value":1886},{"type":50,"tag":451,"props":1976,"children":1977},{},[1978],{"type":56,"value":1891},{"type":50,"tag":414,"props":1980,"children":1981},{},[1982,1986,1994,2002],{"type":50,"tag":451,"props":1983,"children":1984},{},[1985],{"type":56,"value":339},{"type":50,"tag":451,"props":1987,"children":1988},{},[1989],{"type":50,"tag":75,"props":1990,"children":1992},{"className":1991},[],[1993],{"type":56,"value":1907},{"type":50,"tag":451,"props":1995,"children":1996},{},[1997],{"type":50,"tag":75,"props":1998,"children":2000},{"className":1999},[],[2001],{"type":56,"value":1916},{"type":50,"tag":451,"props":2003,"children":2004},{},[2005],{"type":56,"value":1921},{"type":50,"tag":414,"props":2007,"children":2008},{},[2009,2013,2022,2030],{"type":50,"tag":451,"props":2010,"children":2011},{},[2012],{"type":56,"value":339},{"type":50,"tag":451,"props":2014,"children":2015},{},[2016],{"type":50,"tag":75,"props":2017,"children":2019},{"className":2018},[],[2020],{"type":56,"value":2021},"inference.checkpoint",{"type":50,"tag":451,"props":2023,"children":2024},{},[2025],{"type":50,"tag":75,"props":2026,"children":2028},{"className":2027},[],[2029],{"type":56,"value":1946},{"type":50,"tag":451,"props":2031,"children":2032},{},[2033],{"type":56,"value":1951},{"type":50,"tag":414,"props":2035,"children":2036},{},[2037,2041,2049,2057],{"type":50,"tag":451,"props":2038,"children":2039},{},[2040],{"type":56,"value":339},{"type":50,"tag":451,"props":2042,"children":2043},{},[2044],{"type":50,"tag":75,"props":2045,"children":2047},{"className":2046},[],[2048],{"type":56,"value":1444},{"type":50,"tag":451,"props":2050,"children":2051},{},[2052],{"type":50,"tag":75,"props":2053,"children":2055},{"className":2054},[],[2056],{"type":56,"value":1886},{"type":50,"tag":451,"props":2058,"children":2059},{},[2060],{"type":56,"value":1891},{"type":50,"tag":414,"props":2062,"children":2063},{},[2064,2068,2076,2084],{"type":50,"tag":451,"props":2065,"children":2066},{},[2067],{"type":56,"value":552},{"type":50,"tag":451,"props":2069,"children":2070},{},[2071],{"type":50,"tag":75,"props":2072,"children":2074},{"className":2073},[],[2075],{"type":56,"value":1907},{"type":50,"tag":451,"props":2077,"children":2078},{},[2079],{"type":50,"tag":75,"props":2080,"children":2082},{"className":2081},[],[2083],{"type":56,"value":1916},{"type":50,"tag":451,"props":2085,"children":2086},{},[2087],{"type":56,"value":1921},{"type":50,"tag":414,"props":2089,"children":2090},{},[2091,2095,2103,2111],{"type":50,"tag":451,"props":2092,"children":2093},{},[2094],{"type":56,"value":552},{"type":50,"tag":451,"props":2096,"children":2097},{},[2098],{"type":50,"tag":75,"props":2099,"children":2101},{"className":2100},[],[2102],{"type":56,"value":1444},{"type":50,"tag":451,"props":2104,"children":2105},{},[2106],{"type":50,"tag":75,"props":2107,"children":2109},{"className":2108},[],[2110],{"type":56,"value":1886},{"type":50,"tag":451,"props":2112,"children":2113},{},[2114],{"type":56,"value":1891},{"type":50,"tag":414,"props":2116,"children":2117},{},[2118,2122,2130,2139],{"type":50,"tag":451,"props":2119,"children":2120},{},[2121],{"type":56,"value":552},{"type":50,"tag":451,"props":2123,"children":2124},{},[2125],{"type":50,"tag":75,"props":2126,"children":2128},{"className":2127},[],[2129],{"type":56,"value":1595},{"type":50,"tag":451,"props":2131,"children":2132},{},[2133],{"type":50,"tag":75,"props":2134,"children":2136},{"className":2135},[],[2137],{"type":56,"value":2138},"ptm_if_no_resume_model",{"type":50,"tag":451,"props":2140,"children":2141},{},[2142],{"type":56,"value":2143},"PTM when no resume checkpoint exists",{"type":50,"tag":414,"props":2145,"children":2146},{},[2147,2151,2159,2168],{"type":50,"tag":451,"props":2148,"children":2149},{},[2150],{"type":56,"value":552},{"type":50,"tag":451,"props":2152,"children":2153},{},[2154],{"type":50,"tag":75,"props":2155,"children":2157},{"className":2156},[],[2158],{"type":56,"value":1603},{"type":50,"tag":451,"props":2160,"children":2161},{},[2162],{"type":50,"tag":75,"props":2163,"children":2165},{"className":2164},[],[2166],{"type":56,"value":2167},"resume_model",{"type":50,"tag":451,"props":2169,"children":2170},{},[2171],{"type":56,"value":2172},"model file inferred from the current job results folder",{"type":50,"tag":59,"props":2174,"children":2175},{},[2176,2178,2183,2185,2191,2193,2199,2201,2206],{"type":56,"value":2177},"For ",{"type":50,"tag":75,"props":2179,"children":2181},{"className":2180},[],[2182],{"type":56,"value":1946},{"type":56,"value":2184}," or ",{"type":50,"tag":75,"props":2186,"children":2188},{"className":2187},[],[2189],{"type":56,"value":2190},"parent_model_folder",{"type":56,"value":2192},", pass the upstream train\u002Fexport\u002FAutoML child job id as ",{"type":50,"tag":75,"props":2194,"children":2196},{"className":2195},[],[2197],{"type":56,"value":2198},"parent_job_id",{"type":56,"value":2200},". 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":50,"tag":75,"props":2202,"children":2204},{"className":2203},[],[2205],{"type":56,"value":1800},{"type":56,"value":2207}," and do not patch generated runner scripts to guess checkpoint paths.",{"type":50,"tag":2209,"props":2210,"children":2211},"style",{},[2212],{"type":56,"value":2213},"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":2215,"total":2319},[2216,2233,2247,2261,2273,2290,2305],{"slug":2217,"name":2217,"fn":2218,"description":2219,"org":2220,"tags":2221,"stars":23,"repoUrl":24,"updatedAt":2232},"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},[2222,2225,2228,2229],{"name":2223,"slug":2224,"type":15},"Data Analysis","data-analysis",{"name":2226,"slug":2227,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2230,"slug":2231,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2234,"name":2234,"fn":2235,"description":2236,"org":2237,"tags":2238,"stars":23,"repoUrl":24,"updatedAt":2246},"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},[2239,2242,2245],{"name":2240,"slug":2241,"type":15},"Deployment","deployment",{"name":2243,"slug":2244,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2248,"name":2248,"fn":2249,"description":2250,"org":2251,"tags":2252,"stars":23,"repoUrl":24,"updatedAt":2260},"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},[2253,2256,2257],{"name":2254,"slug":2255,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2258,"slug":2259,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2262,"name":2262,"fn":2263,"description":2264,"org":2265,"tags":2266,"stars":23,"repoUrl":24,"updatedAt":2272},"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},[2267,2268,2269],{"name":2223,"slug":2224,"type":15},{"name":9,"slug":8,"type":15},{"name":2270,"slug":2271,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":2274,"name":2274,"fn":2275,"description":2276,"org":2277,"tags":2278,"stars":23,"repoUrl":24,"updatedAt":2289},"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},[2279,2282,2285,2286],{"name":2280,"slug":2281,"type":15},"Automation","automation",{"name":2283,"slug":2284,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":2287,"slug":2288,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2291,"name":2291,"fn":2292,"description":2293,"org":2294,"tags":2295,"stars":23,"repoUrl":24,"updatedAt":2304},"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},[2296,2297,2300,2301],{"name":2240,"slug":2241,"type":15},{"name":2298,"slug":2299,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2302,"slug":2303,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2306,"name":2306,"fn":2307,"description":2308,"org":2309,"tags":2310,"stars":23,"repoUrl":24,"updatedAt":2318},"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},[2311,2312,2315],{"name":9,"slug":8,"type":15},{"name":2313,"slug":2314,"type":15},"Quantum Computing","quantum-computing",{"name":2316,"slug":2317,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":2321,"total":2469},[2322,2340,2355,2366,2378,2392,2405,2417,2428,2437,2451,2460],{"slug":2323,"name":2323,"fn":2324,"description":2325,"org":2326,"tags":2327,"stars":2337,"repoUrl":2338,"updatedAt":2339},"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},[2328,2331,2334],{"name":2329,"slug":2330,"type":15},"Documentation","documentation",{"name":2332,"slug":2333,"type":15},"MCP","mcp",{"name":2335,"slug":2336,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":2341,"name":2341,"fn":2342,"description":2343,"org":2344,"tags":2345,"stars":2352,"repoUrl":2353,"updatedAt":2354},"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},[2346,2349,2350],{"name":2347,"slug":2348,"type":15},"Containers","containers",{"name":2240,"slug":2241,"type":15},{"name":2351,"slug":648,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":2356,"name":2356,"fn":2357,"description":2358,"org":2359,"tags":2360,"stars":2352,"repoUrl":2353,"updatedAt":2365},"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},[2361,2364],{"name":2362,"slug":2363,"type":15},"CI\u002FCD","ci-cd",{"name":2240,"slug":2241,"type":15},"2026-07-14T05:25:59.97109",{"slug":2367,"name":2367,"fn":2368,"description":2369,"org":2370,"tags":2371,"stars":2352,"repoUrl":2353,"updatedAt":2377},"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},[2372,2373,2374],{"name":2362,"slug":2363,"type":15},{"name":2240,"slug":2241,"type":15},{"name":2375,"slug":2376,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":2379,"name":2379,"fn":2380,"description":2381,"org":2382,"tags":2383,"stars":2352,"repoUrl":2353,"updatedAt":2391},"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},[2384,2387,2388],{"name":2385,"slug":2386,"type":15},"Debugging","debugging",{"name":2375,"slug":2376,"type":15},{"name":2389,"slug":2390,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":2393,"name":2393,"fn":2394,"description":2395,"org":2396,"tags":2397,"stars":2352,"repoUrl":2353,"updatedAt":2404},"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},[2398,2401],{"name":2399,"slug":2400,"type":15},"Best Practices","best-practices",{"name":2402,"slug":2403,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":2406,"name":2406,"fn":2407,"description":2408,"org":2409,"tags":2410,"stars":2352,"repoUrl":2353,"updatedAt":2416},"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},[2411,2412,2415],{"name":17,"slug":18,"type":15},{"name":2413,"slug":2414,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":2418,"name":2418,"fn":2419,"description":2420,"org":2421,"tags":2422,"stars":2352,"repoUrl":2353,"updatedAt":2427},"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},[2423,2426],{"name":2424,"slug":2425,"type":15},"QA","qa",{"name":2270,"slug":2271,"type":15},"2026-07-14T05:25:53.673039",{"slug":2429,"name":2429,"fn":2430,"description":2431,"org":2432,"tags":2433,"stars":2352,"repoUrl":2353,"updatedAt":2436},"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},[2434,2435],{"name":2240,"slug":2241,"type":15},{"name":2243,"slug":2244,"type":15},"2026-07-14T05:25:49.362534",{"slug":2438,"name":2438,"fn":2439,"description":2440,"org":2441,"tags":2442,"stars":2352,"repoUrl":2353,"updatedAt":2450},"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},[2443,2446,2447],{"name":2444,"slug":2445,"type":15},"Code Review","code-review",{"name":2375,"slug":2376,"type":15},{"name":2448,"slug":2449,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":2452,"name":2452,"fn":2453,"description":2454,"org":2455,"tags":2456,"stars":2352,"repoUrl":2353,"updatedAt":2459},"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},[2457,2458],{"name":2424,"slug":2425,"type":15},{"name":2270,"slug":2271,"type":15},"2026-07-14T05:25:54.928983",{"slug":2461,"name":2461,"fn":2462,"description":2463,"org":2464,"tags":2465,"stars":2352,"repoUrl":2353,"updatedAt":2468},"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},[2466,2467],{"name":2280,"slug":2281,"type":15},{"name":2362,"slug":2363,"type":15},"2026-07-30T05:29:03.275638",496]