[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-fast-foundation-stereo":3,"mdc--frriz-key":31,"related-repo-nvidia-tao-train-fast-foundation-stereo":1916,"related-org-nvidia-tao-train-fast-foundation-stereo":2018},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":20,"repoUrl":21,"updatedAt":22,"license":23,"forks":24,"topics":25,"repo":26,"sourceUrl":29,"mdContent":30},"tao-train-fast-foundation-stereo","estimate stereo depth with FastFoundationStereo","Real-time stereo depth estimation using FastFoundationStereo (FFS), the distilled bp2 commercial variant of FoundationStereo. Predicts disparity maps from stereo image pairs with ~10× lower latency than full FoundationStereo. Use when training, evaluating, exporting, or running inference for a TAO FastFoundationStereo (FFS) model. Trigger phrases include \"train fast stereo\", \"real-time stereo disparity\", \"FastFoundationStereo\", \"distilled stereo depth\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,17],{"name":13,"slug":14,"type":15},"Performance","performance","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:30:15.129678","Apache-2.0",281,[],{"repoUrl":21,"stars":20,"forks":24,"topics":27,"description":28},[],"AI agent skills published by NVIDIA","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills\u002Ftree\u002FHEAD\u002Fskills\u002Ftao-train-fast-foundation-stereo","---\nname: tao-train-fast-foundation-stereo\ndescription: Real-time stereo depth estimation using FastFoundationStereo (FFS), the distilled bp2 commercial variant of\n  FoundationStereo. Predicts disparity maps from stereo image pairs with ~10× lower latency than full FoundationStereo. Use\n  when training, evaluating, exporting, or running inference for a TAO FastFoundationStereo (FFS) model. Trigger phrases\n  include \"train fast stereo\", \"real-time stereo disparity\", \"FastFoundationStereo\", \"distilled stereo depth\".\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- stereo\n- depth\n- estimation\n- realtime\n- distilled\n---\n\n# Depth Net Fast Stereo\n\nReal-time stereo depth estimation using **FastFoundationStereo (FFS)** — the bp2 commercial distilled variant of FoundationStereo. Predicts disparity maps from rectified stereo image pairs with per-layer pruned widths for real-time inference.\n\nThe mono \u002F stereo \u002F fast-stereo skills share the unified TAO `depth_net` CLI; FFS is selected via `model.model_type: FastFoundationStereo`. FFS differs from `FoundationStereo` only in pruned per-layer widths and a serialized forward path; everything else (entrypoint, action verbs, dataset classes, deploy chain) is identical to `depth-net-stereo`.\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, TensorRT `inference`), read `references\u002Ftao-deploy-fast-foundation-stereo.md` first. The deploy spec template lives at `references\u002Fspec_template_deploy.yaml`.\n\n## When to Use\n\nUse this skill to train, evaluate, export, or run inference for a TAO FastFoundationStereo model. Two supported use cases:\n\nFFS raw-deploy and bp2-finetune flows require a pre-trained bp2 commercial checkpoint (`model_best_bp2_serialize.pth`). The default PyT image does not guarantee that this file is present on disk, so treat the checkpoint path as a required user\u002Fregistry artifact. If no bp2 checkpoint is available, scratch training is still usable for workflow validation, but the resulting metrics are not representative of the bp2 model.\n\n1. **Raw deploy** — use the bp2 ckpt as-is. Skip `train`; run `inference` \u002F `evaluate` \u002F `export` \u002F `gen_trt_engine` directly with the bp2 file as the action's checkpoint.\n2. **Finetune on user data** — set `train.pretrained_model_path` to the bp2 file, train on user data, then verify + deploy on the resulting ckpt. The full 7-action sequence (train → evaluate pyt → inference pyt → export → gen_trt_engine → inference deploy → evaluate deploy) is supported.\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\nFFS shares the `depth_net_stereo` schema but its bp2 architecture widths are fixed invariants. For default AutoML, search only `train.optim.lr` and `train.optim.lr_decay` unless the user explicitly requests a wider search. Do not include FFS architecture fields such as `model.volume_dim`, `model.hidden_dims`, or other bp2 width settings in the default search space.\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## Workflow\n\n### Prerequisites — data accessibility\n\nYour dataset (left + right images + GT disparity for train \u002F evaluate, left + right only for inference) must be reachable from inside the container:\n- **SDK runner**: place files at the S3 paths the runner resolves (`S3_TRAIN` \u002F `S3_EVAL` placeholders shown in spec overrides).\n- **Direct `docker run`** (e.g. local testing): mount the host dataset root read-only at the same in-container path:\n\n```\ndocker run ... -v \u003Chost_data_root>:\u003Chost_data_root>:ro \u003Ccontainer> ...\n```\n\nThe same accessibility requirement applies to the `\u003Coutput_dir>` written by all actions, and to the bp2 checkpoint path.\n\n### Step 1 — Annotation file\n\nPer-line annotation file referenced by `data_sources[*].data_file`. Schema is identical to `depth-net-stereo`:\n\n| Columns | Format | Use |\n|---|---|---|\n| 2 | `\u003Cleft> \u003Cright>` | Stereo inference (no GT) |\n| 3 | `\u003Cleft> \u003Cright> \u003Cdisparity>` | Stereo with GT |\n| 4 | `\u003Cleft> \u003Cright> \u003Cdisparity> \u003Cocclusion_mask>` | Stereo with GT and occlusion mask |\n\nGenerate via `depth_net convert` if needed; see the `depth-net-stereo` skill for `convert_spec.yaml` template.\n\n### Step 2 — Pair `model_type` and `dataset_name` based on your data\n\nUse `model_type: FastFoundationStereo` for FFS. The `dataset_name` choice mirrors the stereo skill — pick the dataset-specific class when your layout matches a registered one, otherwise `GenericDataset`.\n\n| Data category | `model_type` | `dataset_name` |\n|---|---|---|\n| Middlebury | `FastFoundationStereo` | `Middlebury` |\n| KITTI | `FastFoundationStereo` | `Kitti` |\n| ETH3D | `FastFoundationStereo` | `Eth3d` |\n| FSD synthetic | `FastFoundationStereo` | `FSD` |\n| IsaacReal synthetic | `FastFoundationStereo` | `IsaacRealDataset` |\n| Crestereo synthetic | `FastFoundationStereo` | `Crestereo` |\n| Other \u002F non-canonical | `FastFoundationStereo` | `GenericDataset` |\n\nFor inference with 2-column annotations (left + right, no GT), use `dataset_name: GenericDataset` regardless of layout.\n\n### Step 3 — Set the bp2 distilled width overrides\n\nFFS requires 15 model-section width override fields whose values match the bp2 commercial checkpoint exactly. Omitting any field falls back to TAO defaults that do **not** match the bp2 ckpt and produce shape-mismatch errors at forward time. See `references\u002Fsetup-and-run.md` for the full copy-as-is `model:` block and notes. The spec templates at `references\u002Fspec_template_*.yaml` carry this block as the canonical source.\n\n### Step 4 — Write spec yaml from spec overrides\n\nCopy the action block from `references\u002Fspec-overrides.md`. Replace:\n- `model.model_type: FastFoundationStereo` (already set)\n- `dataset.\u003C...>.data_sources[*].dataset_name` from Step 2\n- `dataset.\u003C...>.data_sources[*].data_file` with the path from Step 1\n- For raw deploy use cases (no train): set `\u003Caction>.checkpoint` to the bp2 file path\n- For finetune use cases: set `train.pretrained_model_path` to the bp2 file path\n\nFor chained train → next-action checkpoint path resolution and shape-consistency notes, see `references\u002Fsetup-and-run.md`. SDK-runner deploys resolve handoff automatically via `parent_job_id` — see `references\u002Fparent-model-inference.md`.\n\n### Step 5 — Run\n\nCreate writable home\u002Fcache directories inside the mounted output path before using `--user`, then launch `docker run ... depth_net \u003Caction> -e \u003Cspec.yaml>`. See `references\u002Fsetup-and-run.md` for the full `mkdir` + `docker run` command, the `--user` rationale, and the local bind-mount `__pycache__` tip.\n\n### Step 6 — Verify\n\nCheck container exit code 0 and a populated `status.json` `kpi` block. For `train` inspect per-step `train_loss` directly (the entrypoint reports `Execution status: PASS` even when loss is NaN); for `evaluate` rely on `epe` \u002F `bp1` \u002F `bp2` \u002F `bp3` \u002F `d1` \u002F `rmse`; for `inference` check artifacts under `results_dir`. The pyt-vs-deploy KPI namespace difference and the expected deploy drift are detailed in `references\u002Fsetup-and-run.md`.\n\n### 7-action deploy flow\n\n```\ntrain (optional)            → finetuned ckpt\nevaluate (pyt)              → PyT eager EPE \u002F bp on val GT\ninference (pyt)             → PyT eager disparity samples (visual sanity)\nexport                      → static fp32 ONNX (recommended at 480×736 or 320×736)\ngen_trt_engine             → fp16 TRT engine on static ONNX path\ninference (deploy)         → TRT disparity samples\nevaluate (deploy)          → TRT EPE \u002F bp drift vs PyT eager fp32\n```\n\nSkip `train` for raw-bp2 deploy. The remaining 6 actions (or the 4 deploy-only verbs starting from `export`) cover both use cases.\n\n## Training Requirements\n\n- **Valid `dataset_name` values for stereo `data_sources`** (case-insensitive): `FSD`, `IsaacRealDataset`, `Crestereo`, `Middlebury`, `Eth3d`, `Kitti`, `GenericDataset`\n- **Monitoring metric:** val\u002Floss\n\n### Per-Action Dataset Requirements\n\n| Action | Spec Key | Source | Files | List? |\n|---|---|---|---|---|\n| evaluate | dataset.test_dataset.data_sources | eval_dataset | data_file: annotations.txt + dataset_name | Yes |\n| inference | dataset.infer_dataset.data_sources | inference_dataset | data_file: annotations.txt + dataset_name | Yes |\n| train | dataset.train_dataset.data_sources | train_datasets | data_file: annotations.txt + dataset_name | Yes |\n| train | dataset.val_dataset.data_sources | eval_dataset | data_file: annotations.txt + dataset_name | Yes |\n\n### Typical Spec Overrides\n\nData source overrides are **mandatory for every action**. Each `data_sources` entry is a dict with **two mandatory fields**: `data_file` and `dataset_name`. The `model.*` width fields are also mandatory — see Step 3. See `references\u002Fspec-overrides.md` for the `FFS_MODEL_BLOCK` and per-action (train \u002F evaluate \u002F inference \u002F export) Python override dicts.\n\n## Eval Dataset\n\nOptional. Val dataset configured via `dataset.val_dataset.data_sources` (each entry needs `data_file` and `dataset_name`).\n\n## Important Parameters\n\nKey knobs include `model.model_type` (`FastFoundationStereo`), `model.encoder` (`vitl`), `model.max_disparity` (set `192` explicitly — schema default `416` causes severe drift), `model.mixed_precision` (`false`), `model.gwc_feature_normalize` (`true`), `model.volume_dim` (`28`), `model.valid_iters` (`8`), and per-split `batch_size` \u002F `workers` \u002F `crop_size` \u002F `data_sources`. Full parameter reference, evaluation metrics, multi-GPU \u002F multi-node spec keys, export \u002F TRT defaults, the export use-case matrix, and hardware guidance are in `references\u002Fimportant-parameters.md`.\n\n## Error Patterns\n\nFor `shape mismatch`, `gwc_feature_normalize` schema errors, `max_disparity` drift, negative disparity, `depth_net_stereo: not found`, the pyt-`evaluate` `crop_size` asymmetry, the `Failed to import SAM3` warning, and the dynamic-engine stride-incompatible silent failure, see `references\u002Ferror-patterns.md`.\n\n## Spec Param \u002F Parent Model Inference\n\nModel-specific inference mappings (per-action spec field → inference function) for train \u002F evaluate \u002F inference \u002F export \u002F gen_trt_engine, plus `parent_job_id` \u002F `parent_model` resolution and raw-bp2 explicit-checkpoint handling, are in `references\u002Fparent-model-inference.md`. Generated runners should read that section and apply the mappings with SDK helpers before `create_job()`.\n\n## Deployment\n\n- [tao-deploy-fast-foundation-stereo](references\u002Ftao-deploy-fast-foundation-stereo.md)\n",{"data":32,"body":44},{"name":4,"description":6,"license":23,"compatibility":33,"metadata":34,"allowed-tools":37,"tags":38},"Requires docker + nvidia-container-toolkit.",{"version":35,"author":36},"0.1.0","NVIDIA Corporation","Read Bash",[39,40,41,42,43],"stereo","depth","estimation","realtime","distilled",{"type":45,"children":46},"root",[47,56,70,108,151,158,163,176,244,250,363,432,438,445,450,495,507,520,526,546,645,673,694,721,928,941,947,983,989,1002,1061,1088,1094,1152,1158,1273,1279,1288,1307,1313,1390,1396,1534,1540,1603,1609,1634,1640,1789,1795,1859,1865,1898,1904],{"type":48,"tag":49,"props":50,"children":52},"element","h1",{"id":51},"depth-net-fast-stereo",[53],{"type":54,"value":55},"text","Depth Net Fast Stereo",{"type":48,"tag":57,"props":58,"children":59},"p",{},[60,62,68],{"type":54,"value":61},"Real-time stereo depth estimation using ",{"type":48,"tag":63,"props":64,"children":65},"strong",{},[66],{"type":54,"value":67},"FastFoundationStereo (FFS)",{"type":54,"value":69}," — the bp2 commercial distilled variant of FoundationStereo. Predicts disparity maps from rectified stereo image pairs with per-layer pruned widths for real-time inference.",{"type":48,"tag":57,"props":71,"children":72},{},[73,75,82,84,90,92,98,100,106],{"type":54,"value":74},"The mono \u002F stereo \u002F fast-stereo skills share the unified TAO ",{"type":48,"tag":76,"props":77,"children":79},"code",{"className":78},[],[80],{"type":54,"value":81},"depth_net",{"type":54,"value":83}," CLI; FFS is selected via ",{"type":48,"tag":76,"props":85,"children":87},{"className":86},[],[88],{"type":54,"value":89},"model.model_type: FastFoundationStereo",{"type":54,"value":91},". FFS differs from ",{"type":48,"tag":76,"props":93,"children":95},{"className":94},[],[96],{"type":54,"value":97},"FoundationStereo",{"type":54,"value":99}," only in pruned per-layer widths and a serialized forward path; everything else (entrypoint, action verbs, dataset classes, deploy chain) is identical to ",{"type":48,"tag":76,"props":101,"children":103},{"className":102},[],[104],{"type":54,"value":105},"depth-net-stereo",{"type":54,"value":107},".",{"type":48,"tag":57,"props":109,"children":110},{},[111,113,119,121,127,128,134,136,142,144,150],{"type":54,"value":112},"For TAO Deploy TensorRT actions (",{"type":48,"tag":76,"props":114,"children":116},{"className":115},[],[117],{"type":54,"value":118},"gen_trt_engine",{"type":54,"value":120},", TensorRT ",{"type":48,"tag":76,"props":122,"children":124},{"className":123},[],[125],{"type":54,"value":126},"evaluate",{"type":54,"value":120},{"type":48,"tag":76,"props":129,"children":131},{"className":130},[],[132],{"type":54,"value":133},"inference",{"type":54,"value":135},"), read ",{"type":48,"tag":76,"props":137,"children":139},{"className":138},[],[140],{"type":54,"value":141},"references\u002Ftao-deploy-fast-foundation-stereo.md",{"type":54,"value":143}," first. The deploy spec template lives at ",{"type":48,"tag":76,"props":145,"children":147},{"className":146},[],[148],{"type":54,"value":149},"references\u002Fspec_template_deploy.yaml",{"type":54,"value":107},{"type":48,"tag":152,"props":153,"children":155},"h2",{"id":154},"when-to-use",[156],{"type":54,"value":157},"When to Use",{"type":48,"tag":57,"props":159,"children":160},{},[161],{"type":54,"value":162},"Use this skill to train, evaluate, export, or run inference for a TAO FastFoundationStereo model. Two supported use cases:",{"type":48,"tag":57,"props":164,"children":165},{},[166,168,174],{"type":54,"value":167},"FFS raw-deploy and bp2-finetune flows require a pre-trained bp2 commercial checkpoint (",{"type":48,"tag":76,"props":169,"children":171},{"className":170},[],[172],{"type":54,"value":173},"model_best_bp2_serialize.pth",{"type":54,"value":175},"). The default PyT image does not guarantee that this file is present on disk, so treat the checkpoint path as a required user\u002Fregistry artifact. If no bp2 checkpoint is available, scratch training is still usable for workflow validation, but the resulting metrics are not representative of the bp2 model.",{"type":48,"tag":177,"props":178,"children":179},"ol",{},[180,226],{"type":48,"tag":181,"props":182,"children":183},"li",{},[184,189,191,197,199,204,206,211,212,218,219,224],{"type":48,"tag":63,"props":185,"children":186},{},[187],{"type":54,"value":188},"Raw deploy",{"type":54,"value":190}," — use the bp2 ckpt as-is. Skip ",{"type":48,"tag":76,"props":192,"children":194},{"className":193},[],[195],{"type":54,"value":196},"train",{"type":54,"value":198},"; run ",{"type":48,"tag":76,"props":200,"children":202},{"className":201},[],[203],{"type":54,"value":133},{"type":54,"value":205}," \u002F ",{"type":48,"tag":76,"props":207,"children":209},{"className":208},[],[210],{"type":54,"value":126},{"type":54,"value":205},{"type":48,"tag":76,"props":213,"children":215},{"className":214},[],[216],{"type":54,"value":217},"export",{"type":54,"value":205},{"type":48,"tag":76,"props":220,"children":222},{"className":221},[],[223],{"type":54,"value":118},{"type":54,"value":225}," directly with the bp2 file as the action's checkpoint.",{"type":48,"tag":181,"props":227,"children":228},{},[229,234,236,242],{"type":48,"tag":63,"props":230,"children":231},{},[232],{"type":54,"value":233},"Finetune on user data",{"type":54,"value":235}," — set ",{"type":48,"tag":76,"props":237,"children":239},{"className":238},[],[240],{"type":54,"value":241},"train.pretrained_model_path",{"type":54,"value":243}," to the bp2 file, train on user data, then verify + deploy on the resulting ckpt. The full 7-action sequence (train → evaluate pyt → inference pyt → export → gen_trt_engine → inference deploy → evaluate deploy) is supported.",{"type":48,"tag":152,"props":245,"children":247},{"id":246},"train-action-policy",[248],{"type":54,"value":249},"Train Action Policy",{"type":48,"tag":57,"props":251,"children":252},{},[253,255,261,263,269,271,277,279,285,286,292,294,300,302,307,309,315,317,323,325,331,333,339,341,347,349,354,356,361],{"type":54,"value":254},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":48,"tag":76,"props":256,"children":258},{"className":257},[],[259],{"type":54,"value":260},"references\u002Fskill_info.yaml",{"type":54,"value":262}," and resolve the run override from either an explicit ",{"type":48,"tag":76,"props":264,"children":266},{"className":265},[],[267],{"type":54,"value":268},"automl_policy",{"type":54,"value":270}," value or the user's workflow request. Use ",{"type":48,"tag":76,"props":272,"children":274},{"className":273},[],[275],{"type":54,"value":276},"automl_policy: on",{"type":54,"value":278}," by default and only expose ",{"type":48,"tag":76,"props":280,"children":282},{"className":281},[],[283],{"type":54,"value":284},"on",{"type":54,"value":205},{"type":48,"tag":76,"props":287,"children":289},{"className":288},[],[290],{"type":54,"value":291},"off",{"type":54,"value":293}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":48,"tag":76,"props":295,"children":297},{"className":296},[],[298],{"type":54,"value":299},"automl_policy: off",{"type":54,"value":301}," for this run only. When ",{"type":48,"tag":76,"props":303,"children":305},{"className":304},[],[306],{"type":54,"value":276},{"type":54,"value":308},", ",{"type":48,"tag":76,"props":310,"children":312},{"className":311},[],[313],{"type":54,"value":314},"automl_enabled: true",{"type":54,"value":316},", and both ",{"type":48,"tag":76,"props":318,"children":320},{"className":319},[],[321],{"type":54,"value":322},"schemas\u002Ftrain.schema.json",{"type":54,"value":324}," and ",{"type":48,"tag":76,"props":326,"children":328},{"className":327},[],[329],{"type":54,"value":330},"references\u002Fspec_template_train.yaml",{"type":54,"value":332}," are packaged, route the train action through ",{"type":48,"tag":76,"props":334,"children":336},{"className":335},[],[337],{"type":54,"value":338},"tao-skill-bank:tao-run-automl",{"type":54,"value":340}," by default with this model's ",{"type":48,"tag":76,"props":342,"children":344},{"className":343},[],[345],{"type":54,"value":346},"skill_dir",{"type":54,"value":348},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":48,"tag":76,"props":350,"children":352},{"className":351},[],[353],{"type":54,"value":268},{"type":54,"value":355},". Use direct model training only when ",{"type":48,"tag":76,"props":357,"children":359},{"className":358},[],[360],{"type":54,"value":299},{"type":54,"value":362}," 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":364,"children":365},{},[366,368,374,376,382,383,389,391,397,398,404,406,411,412,417,418,423,425,430],{"type":54,"value":367},"FFS shares the ",{"type":48,"tag":76,"props":369,"children":371},{"className":370},[],[372],{"type":54,"value":373},"depth_net_stereo",{"type":54,"value":375}," schema but its bp2 architecture widths are fixed invariants. For default AutoML, search only ",{"type":48,"tag":76,"props":377,"children":379},{"className":378},[],[380],{"type":54,"value":381},"train.optim.lr",{"type":54,"value":324},{"type":48,"tag":76,"props":384,"children":386},{"className":385},[],[387],{"type":54,"value":388},"train.optim.lr_decay",{"type":54,"value":390}," unless the user explicitly requests a wider search. Do not include FFS architecture fields such as ",{"type":48,"tag":76,"props":392,"children":394},{"className":393},[],[395],{"type":54,"value":396},"model.volume_dim",{"type":54,"value":308},{"type":48,"tag":76,"props":399,"children":401},{"className":400},[],[402],{"type":54,"value":403},"model.hidden_dims",{"type":54,"value":405},", or other bp2 width settings in the default search space.\nNon-train actions such as ",{"type":48,"tag":76,"props":407,"children":409},{"className":408},[],[410],{"type":54,"value":126},{"type":54,"value":308},{"type":48,"tag":76,"props":413,"children":415},{"className":414},[],[416],{"type":54,"value":133},{"type":54,"value":308},{"type":48,"tag":76,"props":419,"children":421},{"className":420},[],[422],{"type":54,"value":217},{"type":54,"value":424},", and deploy flows stay in this model skill. The per-run ",{"type":48,"tag":76,"props":426,"children":428},{"className":427},[],[429],{"type":54,"value":268},{"type":54,"value":431}," override does not change model metadata.",{"type":48,"tag":152,"props":433,"children":435},{"id":434},"workflow",[436],{"type":54,"value":437},"Workflow",{"type":48,"tag":439,"props":440,"children":442},"h3",{"id":441},"prerequisites-data-accessibility",[443],{"type":54,"value":444},"Prerequisites — data accessibility",{"type":48,"tag":57,"props":446,"children":447},{},[448],{"type":54,"value":449},"Your dataset (left + right images + GT disparity for train \u002F evaluate, left + right only for inference) must be reachable from inside the container:",{"type":48,"tag":451,"props":452,"children":453},"ul",{},[454,479],{"type":48,"tag":181,"props":455,"children":456},{},[457,462,464,470,471,477],{"type":48,"tag":63,"props":458,"children":459},{},[460],{"type":54,"value":461},"SDK runner",{"type":54,"value":463},": place files at the S3 paths the runner resolves (",{"type":48,"tag":76,"props":465,"children":467},{"className":466},[],[468],{"type":54,"value":469},"S3_TRAIN",{"type":54,"value":205},{"type":48,"tag":76,"props":472,"children":474},{"className":473},[],[475],{"type":54,"value":476},"S3_EVAL",{"type":54,"value":478}," placeholders shown in spec overrides).",{"type":48,"tag":181,"props":480,"children":481},{},[482,493],{"type":48,"tag":63,"props":483,"children":484},{},[485,487],{"type":54,"value":486},"Direct ",{"type":48,"tag":76,"props":488,"children":490},{"className":489},[],[491],{"type":54,"value":492},"docker run",{"type":54,"value":494}," (e.g. local testing): mount the host dataset root read-only at the same in-container path:",{"type":48,"tag":496,"props":497,"children":501},"pre",{"className":498,"code":500,"language":54},[499],"language-text","docker run ... -v \u003Chost_data_root>:\u003Chost_data_root>:ro \u003Ccontainer> ...\n",[502],{"type":48,"tag":76,"props":503,"children":505},{"__ignoreMap":504},"",[506],{"type":54,"value":500},{"type":48,"tag":57,"props":508,"children":509},{},[510,512,518],{"type":54,"value":511},"The same accessibility requirement applies to the ",{"type":48,"tag":76,"props":513,"children":515},{"className":514},[],[516],{"type":54,"value":517},"\u003Coutput_dir>",{"type":54,"value":519}," written by all actions, and to the bp2 checkpoint path.",{"type":48,"tag":439,"props":521,"children":523},{"id":522},"step-1-annotation-file",[524],{"type":54,"value":525},"Step 1 — Annotation file",{"type":48,"tag":57,"props":527,"children":528},{},[529,531,537,539,544],{"type":54,"value":530},"Per-line annotation file referenced by ",{"type":48,"tag":76,"props":532,"children":534},{"className":533},[],[535],{"type":54,"value":536},"data_sources[*].data_file",{"type":54,"value":538},". Schema is identical to ",{"type":48,"tag":76,"props":540,"children":542},{"className":541},[],[543],{"type":54,"value":105},{"type":54,"value":545},":",{"type":48,"tag":547,"props":548,"children":549},"table",{},[550,574],{"type":48,"tag":551,"props":552,"children":553},"thead",{},[554],{"type":48,"tag":555,"props":556,"children":557},"tr",{},[558,564,569],{"type":48,"tag":559,"props":560,"children":561},"th",{},[562],{"type":54,"value":563},"Columns",{"type":48,"tag":559,"props":565,"children":566},{},[567],{"type":54,"value":568},"Format",{"type":48,"tag":559,"props":570,"children":571},{},[572],{"type":54,"value":573},"Use",{"type":48,"tag":575,"props":576,"children":577},"tbody",{},[578,601,623],{"type":48,"tag":555,"props":579,"children":580},{},[581,587,596],{"type":48,"tag":582,"props":583,"children":584},"td",{},[585],{"type":54,"value":586},"2",{"type":48,"tag":582,"props":588,"children":589},{},[590],{"type":48,"tag":76,"props":591,"children":593},{"className":592},[],[594],{"type":54,"value":595},"\u003Cleft> \u003Cright>",{"type":48,"tag":582,"props":597,"children":598},{},[599],{"type":54,"value":600},"Stereo inference (no GT)",{"type":48,"tag":555,"props":602,"children":603},{},[604,609,618],{"type":48,"tag":582,"props":605,"children":606},{},[607],{"type":54,"value":608},"3",{"type":48,"tag":582,"props":610,"children":611},{},[612],{"type":48,"tag":76,"props":613,"children":615},{"className":614},[],[616],{"type":54,"value":617},"\u003Cleft> \u003Cright> \u003Cdisparity>",{"type":48,"tag":582,"props":619,"children":620},{},[621],{"type":54,"value":622},"Stereo with GT",{"type":48,"tag":555,"props":624,"children":625},{},[626,631,640],{"type":48,"tag":582,"props":627,"children":628},{},[629],{"type":54,"value":630},"4",{"type":48,"tag":582,"props":632,"children":633},{},[634],{"type":48,"tag":76,"props":635,"children":637},{"className":636},[],[638],{"type":54,"value":639},"\u003Cleft> \u003Cright> \u003Cdisparity> \u003Cocclusion_mask>",{"type":48,"tag":582,"props":641,"children":642},{},[643],{"type":54,"value":644},"Stereo with GT and occlusion mask",{"type":48,"tag":57,"props":646,"children":647},{},[648,650,656,658,663,665,671],{"type":54,"value":649},"Generate via ",{"type":48,"tag":76,"props":651,"children":653},{"className":652},[],[654],{"type":54,"value":655},"depth_net convert",{"type":54,"value":657}," if needed; see the ",{"type":48,"tag":76,"props":659,"children":661},{"className":660},[],[662],{"type":54,"value":105},{"type":54,"value":664}," skill for ",{"type":48,"tag":76,"props":666,"children":668},{"className":667},[],[669],{"type":54,"value":670},"convert_spec.yaml",{"type":54,"value":672}," template.",{"type":48,"tag":439,"props":674,"children":676},{"id":675},"step-2-pair-model_type-and-dataset_name-based-on-your-data",[677,679,685,686,692],{"type":54,"value":678},"Step 2 — Pair ",{"type":48,"tag":76,"props":680,"children":682},{"className":681},[],[683],{"type":54,"value":684},"model_type",{"type":54,"value":324},{"type":48,"tag":76,"props":687,"children":689},{"className":688},[],[690],{"type":54,"value":691},"dataset_name",{"type":54,"value":693}," based on your data",{"type":48,"tag":57,"props":695,"children":696},{},[697,699,705,707,712,714,720],{"type":54,"value":698},"Use ",{"type":48,"tag":76,"props":700,"children":702},{"className":701},[],[703],{"type":54,"value":704},"model_type: FastFoundationStereo",{"type":54,"value":706}," for FFS. The ",{"type":48,"tag":76,"props":708,"children":710},{"className":709},[],[711],{"type":54,"value":691},{"type":54,"value":713}," choice mirrors the stereo skill — pick the dataset-specific class when your layout matches a registered one, otherwise ",{"type":48,"tag":76,"props":715,"children":717},{"className":716},[],[718],{"type":54,"value":719},"GenericDataset",{"type":54,"value":107},{"type":48,"tag":547,"props":722,"children":723},{},[724,751],{"type":48,"tag":551,"props":725,"children":726},{},[727],{"type":48,"tag":555,"props":728,"children":729},{},[730,735,743],{"type":48,"tag":559,"props":731,"children":732},{},[733],{"type":54,"value":734},"Data category",{"type":48,"tag":559,"props":736,"children":737},{},[738],{"type":48,"tag":76,"props":739,"children":741},{"className":740},[],[742],{"type":54,"value":684},{"type":48,"tag":559,"props":744,"children":745},{},[746],{"type":48,"tag":76,"props":747,"children":749},{"className":748},[],[750],{"type":54,"value":691},{"type":48,"tag":575,"props":752,"children":753},{},[754,779,804,829,854,879,904],{"type":48,"tag":555,"props":755,"children":756},{},[757,762,771],{"type":48,"tag":582,"props":758,"children":759},{},[760],{"type":54,"value":761},"Middlebury",{"type":48,"tag":582,"props":763,"children":764},{},[765],{"type":48,"tag":76,"props":766,"children":768},{"className":767},[],[769],{"type":54,"value":770},"FastFoundationStereo",{"type":48,"tag":582,"props":772,"children":773},{},[774],{"type":48,"tag":76,"props":775,"children":777},{"className":776},[],[778],{"type":54,"value":761},{"type":48,"tag":555,"props":780,"children":781},{},[782,787,795],{"type":48,"tag":582,"props":783,"children":784},{},[785],{"type":54,"value":786},"KITTI",{"type":48,"tag":582,"props":788,"children":789},{},[790],{"type":48,"tag":76,"props":791,"children":793},{"className":792},[],[794],{"type":54,"value":770},{"type":48,"tag":582,"props":796,"children":797},{},[798],{"type":48,"tag":76,"props":799,"children":801},{"className":800},[],[802],{"type":54,"value":803},"Kitti",{"type":48,"tag":555,"props":805,"children":806},{},[807,812,820],{"type":48,"tag":582,"props":808,"children":809},{},[810],{"type":54,"value":811},"ETH3D",{"type":48,"tag":582,"props":813,"children":814},{},[815],{"type":48,"tag":76,"props":816,"children":818},{"className":817},[],[819],{"type":54,"value":770},{"type":48,"tag":582,"props":821,"children":822},{},[823],{"type":48,"tag":76,"props":824,"children":826},{"className":825},[],[827],{"type":54,"value":828},"Eth3d",{"type":48,"tag":555,"props":830,"children":831},{},[832,837,845],{"type":48,"tag":582,"props":833,"children":834},{},[835],{"type":54,"value":836},"FSD synthetic",{"type":48,"tag":582,"props":838,"children":839},{},[840],{"type":48,"tag":76,"props":841,"children":843},{"className":842},[],[844],{"type":54,"value":770},{"type":48,"tag":582,"props":846,"children":847},{},[848],{"type":48,"tag":76,"props":849,"children":851},{"className":850},[],[852],{"type":54,"value":853},"FSD",{"type":48,"tag":555,"props":855,"children":856},{},[857,862,870],{"type":48,"tag":582,"props":858,"children":859},{},[860],{"type":54,"value":861},"IsaacReal synthetic",{"type":48,"tag":582,"props":863,"children":864},{},[865],{"type":48,"tag":76,"props":866,"children":868},{"className":867},[],[869],{"type":54,"value":770},{"type":48,"tag":582,"props":871,"children":872},{},[873],{"type":48,"tag":76,"props":874,"children":876},{"className":875},[],[877],{"type":54,"value":878},"IsaacRealDataset",{"type":48,"tag":555,"props":880,"children":881},{},[882,887,895],{"type":48,"tag":582,"props":883,"children":884},{},[885],{"type":54,"value":886},"Crestereo synthetic",{"type":48,"tag":582,"props":888,"children":889},{},[890],{"type":48,"tag":76,"props":891,"children":893},{"className":892},[],[894],{"type":54,"value":770},{"type":48,"tag":582,"props":896,"children":897},{},[898],{"type":48,"tag":76,"props":899,"children":901},{"className":900},[],[902],{"type":54,"value":903},"Crestereo",{"type":48,"tag":555,"props":905,"children":906},{},[907,912,920],{"type":48,"tag":582,"props":908,"children":909},{},[910],{"type":54,"value":911},"Other \u002F non-canonical",{"type":48,"tag":582,"props":913,"children":914},{},[915],{"type":48,"tag":76,"props":916,"children":918},{"className":917},[],[919],{"type":54,"value":770},{"type":48,"tag":582,"props":921,"children":922},{},[923],{"type":48,"tag":76,"props":924,"children":926},{"className":925},[],[927],{"type":54,"value":719},{"type":48,"tag":57,"props":929,"children":930},{},[931,933,939],{"type":54,"value":932},"For inference with 2-column annotations (left + right, no GT), use ",{"type":48,"tag":76,"props":934,"children":936},{"className":935},[],[937],{"type":54,"value":938},"dataset_name: GenericDataset",{"type":54,"value":940}," regardless of layout.",{"type":48,"tag":439,"props":942,"children":944},{"id":943},"step-3-set-the-bp2-distilled-width-overrides",[945],{"type":54,"value":946},"Step 3 — Set the bp2 distilled width overrides",{"type":48,"tag":57,"props":948,"children":949},{},[950,952,957,959,965,967,973,975,981],{"type":54,"value":951},"FFS requires 15 model-section width override fields whose values match the bp2 commercial checkpoint exactly. Omitting any field falls back to TAO defaults that do ",{"type":48,"tag":63,"props":953,"children":954},{},[955],{"type":54,"value":956},"not",{"type":54,"value":958}," match the bp2 ckpt and produce shape-mismatch errors at forward time. See ",{"type":48,"tag":76,"props":960,"children":962},{"className":961},[],[963],{"type":54,"value":964},"references\u002Fsetup-and-run.md",{"type":54,"value":966}," for the full copy-as-is ",{"type":48,"tag":76,"props":968,"children":970},{"className":969},[],[971],{"type":54,"value":972},"model:",{"type":54,"value":974}," block and notes. The spec templates at ",{"type":48,"tag":76,"props":976,"children":978},{"className":977},[],[979],{"type":54,"value":980},"references\u002Fspec_template_*.yaml",{"type":54,"value":982}," carry this block as the canonical source.",{"type":48,"tag":439,"props":984,"children":986},{"id":985},"step-4-write-spec-yaml-from-spec-overrides",[987],{"type":54,"value":988},"Step 4 — Write spec yaml from spec overrides",{"type":48,"tag":57,"props":990,"children":991},{},[992,994,1000],{"type":54,"value":993},"Copy the action block from ",{"type":48,"tag":76,"props":995,"children":997},{"className":996},[],[998],{"type":54,"value":999},"references\u002Fspec-overrides.md",{"type":54,"value":1001},". Replace:",{"type":48,"tag":451,"props":1003,"children":1004},{},[1005,1015,1026,1037,1050],{"type":48,"tag":181,"props":1006,"children":1007},{},[1008,1013],{"type":48,"tag":76,"props":1009,"children":1011},{"className":1010},[],[1012],{"type":54,"value":89},{"type":54,"value":1014}," (already set)",{"type":48,"tag":181,"props":1016,"children":1017},{},[1018,1024],{"type":48,"tag":76,"props":1019,"children":1021},{"className":1020},[],[1022],{"type":54,"value":1023},"dataset.\u003C...>.data_sources[*].dataset_name",{"type":54,"value":1025}," from Step 2",{"type":48,"tag":181,"props":1027,"children":1028},{},[1029,1035],{"type":48,"tag":76,"props":1030,"children":1032},{"className":1031},[],[1033],{"type":54,"value":1034},"dataset.\u003C...>.data_sources[*].data_file",{"type":54,"value":1036}," with the path from Step 1",{"type":48,"tag":181,"props":1038,"children":1039},{},[1040,1042,1048],{"type":54,"value":1041},"For raw deploy use cases (no train): set ",{"type":48,"tag":76,"props":1043,"children":1045},{"className":1044},[],[1046],{"type":54,"value":1047},"\u003Caction>.checkpoint",{"type":54,"value":1049}," to the bp2 file path",{"type":48,"tag":181,"props":1051,"children":1052},{},[1053,1055,1060],{"type":54,"value":1054},"For finetune use cases: set ",{"type":48,"tag":76,"props":1056,"children":1058},{"className":1057},[],[1059],{"type":54,"value":241},{"type":54,"value":1049},{"type":48,"tag":57,"props":1062,"children":1063},{},[1064,1066,1071,1073,1079,1081,1087],{"type":54,"value":1065},"For chained train → next-action checkpoint path resolution and shape-consistency notes, see ",{"type":48,"tag":76,"props":1067,"children":1069},{"className":1068},[],[1070],{"type":54,"value":964},{"type":54,"value":1072},". SDK-runner deploys resolve handoff automatically via ",{"type":48,"tag":76,"props":1074,"children":1076},{"className":1075},[],[1077],{"type":54,"value":1078},"parent_job_id",{"type":54,"value":1080}," — see ",{"type":48,"tag":76,"props":1082,"children":1084},{"className":1083},[],[1085],{"type":54,"value":1086},"references\u002Fparent-model-inference.md",{"type":54,"value":107},{"type":48,"tag":439,"props":1089,"children":1091},{"id":1090},"step-5-run",[1092],{"type":54,"value":1093},"Step 5 — Run",{"type":48,"tag":57,"props":1095,"children":1096},{},[1097,1099,1105,1107,1113,1115,1120,1122,1128,1130,1135,1137,1142,1144,1150],{"type":54,"value":1098},"Create writable home\u002Fcache directories inside the mounted output path before using ",{"type":48,"tag":76,"props":1100,"children":1102},{"className":1101},[],[1103],{"type":54,"value":1104},"--user",{"type":54,"value":1106},", then launch ",{"type":48,"tag":76,"props":1108,"children":1110},{"className":1109},[],[1111],{"type":54,"value":1112},"docker run ... depth_net \u003Caction> -e \u003Cspec.yaml>",{"type":54,"value":1114},". See ",{"type":48,"tag":76,"props":1116,"children":1118},{"className":1117},[],[1119],{"type":54,"value":964},{"type":54,"value":1121}," for the full ",{"type":48,"tag":76,"props":1123,"children":1125},{"className":1124},[],[1126],{"type":54,"value":1127},"mkdir",{"type":54,"value":1129}," + ",{"type":48,"tag":76,"props":1131,"children":1133},{"className":1132},[],[1134],{"type":54,"value":492},{"type":54,"value":1136}," command, the ",{"type":48,"tag":76,"props":1138,"children":1140},{"className":1139},[],[1141],{"type":54,"value":1104},{"type":54,"value":1143}," rationale, and the local bind-mount ",{"type":48,"tag":76,"props":1145,"children":1147},{"className":1146},[],[1148],{"type":54,"value":1149},"__pycache__",{"type":54,"value":1151}," tip.",{"type":48,"tag":439,"props":1153,"children":1155},{"id":1154},"step-6-verify",[1156],{"type":54,"value":1157},"Step 6 — Verify",{"type":48,"tag":57,"props":1159,"children":1160},{},[1161,1163,1169,1171,1177,1179,1184,1186,1192,1194,1200,1202,1207,1209,1215,1216,1222,1223,1229,1230,1236,1237,1243,1244,1250,1252,1257,1259,1265,1267,1272],{"type":54,"value":1162},"Check container exit code 0 and a populated ",{"type":48,"tag":76,"props":1164,"children":1166},{"className":1165},[],[1167],{"type":54,"value":1168},"status.json",{"type":54,"value":1170}," ",{"type":48,"tag":76,"props":1172,"children":1174},{"className":1173},[],[1175],{"type":54,"value":1176},"kpi",{"type":54,"value":1178}," block. For ",{"type":48,"tag":76,"props":1180,"children":1182},{"className":1181},[],[1183],{"type":54,"value":196},{"type":54,"value":1185}," inspect per-step ",{"type":48,"tag":76,"props":1187,"children":1189},{"className":1188},[],[1190],{"type":54,"value":1191},"train_loss",{"type":54,"value":1193}," directly (the entrypoint reports ",{"type":48,"tag":76,"props":1195,"children":1197},{"className":1196},[],[1198],{"type":54,"value":1199},"Execution status: PASS",{"type":54,"value":1201}," even when loss is NaN); for ",{"type":48,"tag":76,"props":1203,"children":1205},{"className":1204},[],[1206],{"type":54,"value":126},{"type":54,"value":1208}," rely on ",{"type":48,"tag":76,"props":1210,"children":1212},{"className":1211},[],[1213],{"type":54,"value":1214},"epe",{"type":54,"value":205},{"type":48,"tag":76,"props":1217,"children":1219},{"className":1218},[],[1220],{"type":54,"value":1221},"bp1",{"type":54,"value":205},{"type":48,"tag":76,"props":1224,"children":1226},{"className":1225},[],[1227],{"type":54,"value":1228},"bp2",{"type":54,"value":205},{"type":48,"tag":76,"props":1231,"children":1233},{"className":1232},[],[1234],{"type":54,"value":1235},"bp3",{"type":54,"value":205},{"type":48,"tag":76,"props":1238,"children":1240},{"className":1239},[],[1241],{"type":54,"value":1242},"d1",{"type":54,"value":205},{"type":48,"tag":76,"props":1245,"children":1247},{"className":1246},[],[1248],{"type":54,"value":1249},"rmse",{"type":54,"value":1251},"; for ",{"type":48,"tag":76,"props":1253,"children":1255},{"className":1254},[],[1256],{"type":54,"value":133},{"type":54,"value":1258}," check artifacts under ",{"type":48,"tag":76,"props":1260,"children":1262},{"className":1261},[],[1263],{"type":54,"value":1264},"results_dir",{"type":54,"value":1266},". The pyt-vs-deploy KPI namespace difference and the expected deploy drift are detailed in ",{"type":48,"tag":76,"props":1268,"children":1270},{"className":1269},[],[1271],{"type":54,"value":964},{"type":54,"value":107},{"type":48,"tag":439,"props":1274,"children":1276},{"id":1275},"_7-action-deploy-flow",[1277],{"type":54,"value":1278},"7-action deploy flow",{"type":48,"tag":496,"props":1280,"children":1283},{"className":1281,"code":1282,"language":54},[499],"train (optional)            → finetuned ckpt\nevaluate (pyt)              → PyT eager EPE \u002F bp on val GT\ninference (pyt)             → PyT eager disparity samples (visual sanity)\nexport                      → static fp32 ONNX (recommended at 480×736 or 320×736)\ngen_trt_engine             → fp16 TRT engine on static ONNX path\ninference (deploy)         → TRT disparity samples\nevaluate (deploy)          → TRT EPE \u002F bp drift vs PyT eager fp32\n",[1284],{"type":48,"tag":76,"props":1285,"children":1286},{"__ignoreMap":504},[1287],{"type":54,"value":1282},{"type":48,"tag":57,"props":1289,"children":1290},{},[1291,1293,1298,1300,1305],{"type":54,"value":1292},"Skip ",{"type":48,"tag":76,"props":1294,"children":1296},{"className":1295},[],[1297],{"type":54,"value":196},{"type":54,"value":1299}," for raw-bp2 deploy. The remaining 6 actions (or the 4 deploy-only verbs starting from ",{"type":48,"tag":76,"props":1301,"children":1303},{"className":1302},[],[1304],{"type":54,"value":217},{"type":54,"value":1306},") cover both use cases.",{"type":48,"tag":152,"props":1308,"children":1310},{"id":1309},"training-requirements",[1311],{"type":54,"value":1312},"Training Requirements",{"type":48,"tag":451,"props":1314,"children":1315},{},[1316,1380],{"type":48,"tag":181,"props":1317,"children":1318},{},[1319,1337,1339,1344,1345,1350,1351,1356,1357,1362,1363,1368,1369,1374,1375],{"type":48,"tag":63,"props":1320,"children":1321},{},[1322,1324,1329,1331],{"type":54,"value":1323},"Valid ",{"type":48,"tag":76,"props":1325,"children":1327},{"className":1326},[],[1328],{"type":54,"value":691},{"type":54,"value":1330}," values for stereo ",{"type":48,"tag":76,"props":1332,"children":1334},{"className":1333},[],[1335],{"type":54,"value":1336},"data_sources",{"type":54,"value":1338}," (case-insensitive): ",{"type":48,"tag":76,"props":1340,"children":1342},{"className":1341},[],[1343],{"type":54,"value":853},{"type":54,"value":308},{"type":48,"tag":76,"props":1346,"children":1348},{"className":1347},[],[1349],{"type":54,"value":878},{"type":54,"value":308},{"type":48,"tag":76,"props":1352,"children":1354},{"className":1353},[],[1355],{"type":54,"value":903},{"type":54,"value":308},{"type":48,"tag":76,"props":1358,"children":1360},{"className":1359},[],[1361],{"type":54,"value":761},{"type":54,"value":308},{"type":48,"tag":76,"props":1364,"children":1366},{"className":1365},[],[1367],{"type":54,"value":828},{"type":54,"value":308},{"type":48,"tag":76,"props":1370,"children":1372},{"className":1371},[],[1373],{"type":54,"value":803},{"type":54,"value":308},{"type":48,"tag":76,"props":1376,"children":1378},{"className":1377},[],[1379],{"type":54,"value":719},{"type":48,"tag":181,"props":1381,"children":1382},{},[1383,1388],{"type":48,"tag":63,"props":1384,"children":1385},{},[1386],{"type":54,"value":1387},"Monitoring metric:",{"type":54,"value":1389}," val\u002Floss",{"type":48,"tag":439,"props":1391,"children":1393},{"id":1392},"per-action-dataset-requirements",[1394],{"type":54,"value":1395},"Per-Action Dataset Requirements",{"type":48,"tag":547,"props":1397,"children":1398},{},[1399,1430],{"type":48,"tag":551,"props":1400,"children":1401},{},[1402],{"type":48,"tag":555,"props":1403,"children":1404},{},[1405,1410,1415,1420,1425],{"type":48,"tag":559,"props":1406,"children":1407},{},[1408],{"type":54,"value":1409},"Action",{"type":48,"tag":559,"props":1411,"children":1412},{},[1413],{"type":54,"value":1414},"Spec Key",{"type":48,"tag":559,"props":1416,"children":1417},{},[1418],{"type":54,"value":1419},"Source",{"type":48,"tag":559,"props":1421,"children":1422},{},[1423],{"type":54,"value":1424},"Files",{"type":48,"tag":559,"props":1426,"children":1427},{},[1428],{"type":54,"value":1429},"List?",{"type":48,"tag":575,"props":1431,"children":1432},{},[1433,1460,1485,1510],{"type":48,"tag":555,"props":1434,"children":1435},{},[1436,1440,1445,1450,1455],{"type":48,"tag":582,"props":1437,"children":1438},{},[1439],{"type":54,"value":126},{"type":48,"tag":582,"props":1441,"children":1442},{},[1443],{"type":54,"value":1444},"dataset.test_dataset.data_sources",{"type":48,"tag":582,"props":1446,"children":1447},{},[1448],{"type":54,"value":1449},"eval_dataset",{"type":48,"tag":582,"props":1451,"children":1452},{},[1453],{"type":54,"value":1454},"data_file: annotations.txt + dataset_name",{"type":48,"tag":582,"props":1456,"children":1457},{},[1458],{"type":54,"value":1459},"Yes",{"type":48,"tag":555,"props":1461,"children":1462},{},[1463,1467,1472,1477,1481],{"type":48,"tag":582,"props":1464,"children":1465},{},[1466],{"type":54,"value":133},{"type":48,"tag":582,"props":1468,"children":1469},{},[1470],{"type":54,"value":1471},"dataset.infer_dataset.data_sources",{"type":48,"tag":582,"props":1473,"children":1474},{},[1475],{"type":54,"value":1476},"inference_dataset",{"type":48,"tag":582,"props":1478,"children":1479},{},[1480],{"type":54,"value":1454},{"type":48,"tag":582,"props":1482,"children":1483},{},[1484],{"type":54,"value":1459},{"type":48,"tag":555,"props":1486,"children":1487},{},[1488,1492,1497,1502,1506],{"type":48,"tag":582,"props":1489,"children":1490},{},[1491],{"type":54,"value":196},{"type":48,"tag":582,"props":1493,"children":1494},{},[1495],{"type":54,"value":1496},"dataset.train_dataset.data_sources",{"type":48,"tag":582,"props":1498,"children":1499},{},[1500],{"type":54,"value":1501},"train_datasets",{"type":48,"tag":582,"props":1503,"children":1504},{},[1505],{"type":54,"value":1454},{"type":48,"tag":582,"props":1507,"children":1508},{},[1509],{"type":54,"value":1459},{"type":48,"tag":555,"props":1511,"children":1512},{},[1513,1517,1522,1526,1530],{"type":48,"tag":582,"props":1514,"children":1515},{},[1516],{"type":54,"value":196},{"type":48,"tag":582,"props":1518,"children":1519},{},[1520],{"type":54,"value":1521},"dataset.val_dataset.data_sources",{"type":48,"tag":582,"props":1523,"children":1524},{},[1525],{"type":54,"value":1449},{"type":48,"tag":582,"props":1527,"children":1528},{},[1529],{"type":54,"value":1454},{"type":48,"tag":582,"props":1531,"children":1532},{},[1533],{"type":54,"value":1459},{"type":48,"tag":439,"props":1535,"children":1537},{"id":1536},"typical-spec-overrides",[1538],{"type":54,"value":1539},"Typical Spec Overrides",{"type":48,"tag":57,"props":1541,"children":1542},{},[1543,1545,1550,1552,1557,1559,1564,1566,1572,1573,1578,1580,1586,1588,1593,1595,1601],{"type":54,"value":1544},"Data source overrides are ",{"type":48,"tag":63,"props":1546,"children":1547},{},[1548],{"type":54,"value":1549},"mandatory for every action",{"type":54,"value":1551},". Each ",{"type":48,"tag":76,"props":1553,"children":1555},{"className":1554},[],[1556],{"type":54,"value":1336},{"type":54,"value":1558}," entry is a dict with ",{"type":48,"tag":63,"props":1560,"children":1561},{},[1562],{"type":54,"value":1563},"two mandatory fields",{"type":54,"value":1565},": ",{"type":48,"tag":76,"props":1567,"children":1569},{"className":1568},[],[1570],{"type":54,"value":1571},"data_file",{"type":54,"value":324},{"type":48,"tag":76,"props":1574,"children":1576},{"className":1575},[],[1577],{"type":54,"value":691},{"type":54,"value":1579},". The ",{"type":48,"tag":76,"props":1581,"children":1583},{"className":1582},[],[1584],{"type":54,"value":1585},"model.*",{"type":54,"value":1587}," width fields are also mandatory — see Step 3. See ",{"type":48,"tag":76,"props":1589,"children":1591},{"className":1590},[],[1592],{"type":54,"value":999},{"type":54,"value":1594}," for the ",{"type":48,"tag":76,"props":1596,"children":1598},{"className":1597},[],[1599],{"type":54,"value":1600},"FFS_MODEL_BLOCK",{"type":54,"value":1602}," and per-action (train \u002F evaluate \u002F inference \u002F export) Python override dicts.",{"type":48,"tag":152,"props":1604,"children":1606},{"id":1605},"eval-dataset",[1607],{"type":54,"value":1608},"Eval Dataset",{"type":48,"tag":57,"props":1610,"children":1611},{},[1612,1614,1619,1621,1626,1627,1632],{"type":54,"value":1613},"Optional. Val dataset configured via ",{"type":48,"tag":76,"props":1615,"children":1617},{"className":1616},[],[1618],{"type":54,"value":1521},{"type":54,"value":1620}," (each entry needs ",{"type":48,"tag":76,"props":1622,"children":1624},{"className":1623},[],[1625],{"type":54,"value":1571},{"type":54,"value":324},{"type":48,"tag":76,"props":1628,"children":1630},{"className":1629},[],[1631],{"type":54,"value":691},{"type":54,"value":1633},").",{"type":48,"tag":152,"props":1635,"children":1637},{"id":1636},"important-parameters",[1638],{"type":54,"value":1639},"Important Parameters",{"type":48,"tag":57,"props":1641,"children":1642},{},[1643,1645,1651,1653,1658,1660,1666,1667,1673,1674,1680,1682,1688,1690,1696,1698,1704,1705,1711,1712,1718,1719,1725,1726,1731,1732,1738,1739,1745,1746,1752,1754,1760,1761,1767,1768,1774,1775,1780,1782,1788],{"type":54,"value":1644},"Key knobs include ",{"type":48,"tag":76,"props":1646,"children":1648},{"className":1647},[],[1649],{"type":54,"value":1650},"model.model_type",{"type":54,"value":1652}," (",{"type":48,"tag":76,"props":1654,"children":1656},{"className":1655},[],[1657],{"type":54,"value":770},{"type":54,"value":1659},"), ",{"type":48,"tag":76,"props":1661,"children":1663},{"className":1662},[],[1664],{"type":54,"value":1665},"model.encoder",{"type":54,"value":1652},{"type":48,"tag":76,"props":1668,"children":1670},{"className":1669},[],[1671],{"type":54,"value":1672},"vitl",{"type":54,"value":1659},{"type":48,"tag":76,"props":1675,"children":1677},{"className":1676},[],[1678],{"type":54,"value":1679},"model.max_disparity",{"type":54,"value":1681}," (set ",{"type":48,"tag":76,"props":1683,"children":1685},{"className":1684},[],[1686],{"type":54,"value":1687},"192",{"type":54,"value":1689}," explicitly — schema default ",{"type":48,"tag":76,"props":1691,"children":1693},{"className":1692},[],[1694],{"type":54,"value":1695},"416",{"type":54,"value":1697}," causes severe drift), ",{"type":48,"tag":76,"props":1699,"children":1701},{"className":1700},[],[1702],{"type":54,"value":1703},"model.mixed_precision",{"type":54,"value":1652},{"type":48,"tag":76,"props":1706,"children":1708},{"className":1707},[],[1709],{"type":54,"value":1710},"false",{"type":54,"value":1659},{"type":48,"tag":76,"props":1713,"children":1715},{"className":1714},[],[1716],{"type":54,"value":1717},"model.gwc_feature_normalize",{"type":54,"value":1652},{"type":48,"tag":76,"props":1720,"children":1722},{"className":1721},[],[1723],{"type":54,"value":1724},"true",{"type":54,"value":1659},{"type":48,"tag":76,"props":1727,"children":1729},{"className":1728},[],[1730],{"type":54,"value":396},{"type":54,"value":1652},{"type":48,"tag":76,"props":1733,"children":1735},{"className":1734},[],[1736],{"type":54,"value":1737},"28",{"type":54,"value":1659},{"type":48,"tag":76,"props":1740,"children":1742},{"className":1741},[],[1743],{"type":54,"value":1744},"model.valid_iters",{"type":54,"value":1652},{"type":48,"tag":76,"props":1747,"children":1749},{"className":1748},[],[1750],{"type":54,"value":1751},"8",{"type":54,"value":1753},"), and per-split ",{"type":48,"tag":76,"props":1755,"children":1757},{"className":1756},[],[1758],{"type":54,"value":1759},"batch_size",{"type":54,"value":205},{"type":48,"tag":76,"props":1762,"children":1764},{"className":1763},[],[1765],{"type":54,"value":1766},"workers",{"type":54,"value":205},{"type":48,"tag":76,"props":1769,"children":1771},{"className":1770},[],[1772],{"type":54,"value":1773},"crop_size",{"type":54,"value":205},{"type":48,"tag":76,"props":1776,"children":1778},{"className":1777},[],[1779],{"type":54,"value":1336},{"type":54,"value":1781},". Full parameter reference, evaluation metrics, multi-GPU \u002F multi-node spec keys, export \u002F TRT defaults, the export use-case matrix, and hardware guidance are in ",{"type":48,"tag":76,"props":1783,"children":1785},{"className":1784},[],[1786],{"type":54,"value":1787},"references\u002Fimportant-parameters.md",{"type":54,"value":107},{"type":48,"tag":152,"props":1790,"children":1792},{"id":1791},"error-patterns",[1793],{"type":54,"value":1794},"Error Patterns",{"type":48,"tag":57,"props":1796,"children":1797},{},[1798,1800,1806,1807,1813,1815,1821,1823,1829,1831,1836,1837,1842,1844,1850,1852,1858],{"type":54,"value":1799},"For ",{"type":48,"tag":76,"props":1801,"children":1803},{"className":1802},[],[1804],{"type":54,"value":1805},"shape mismatch",{"type":54,"value":308},{"type":48,"tag":76,"props":1808,"children":1810},{"className":1809},[],[1811],{"type":54,"value":1812},"gwc_feature_normalize",{"type":54,"value":1814}," schema errors, ",{"type":48,"tag":76,"props":1816,"children":1818},{"className":1817},[],[1819],{"type":54,"value":1820},"max_disparity",{"type":54,"value":1822}," drift, negative disparity, ",{"type":48,"tag":76,"props":1824,"children":1826},{"className":1825},[],[1827],{"type":54,"value":1828},"depth_net_stereo: not found",{"type":54,"value":1830},", the pyt-",{"type":48,"tag":76,"props":1832,"children":1834},{"className":1833},[],[1835],{"type":54,"value":126},{"type":54,"value":1170},{"type":48,"tag":76,"props":1838,"children":1840},{"className":1839},[],[1841],{"type":54,"value":1773},{"type":54,"value":1843}," asymmetry, the ",{"type":48,"tag":76,"props":1845,"children":1847},{"className":1846},[],[1848],{"type":54,"value":1849},"Failed to import SAM3",{"type":54,"value":1851}," warning, and the dynamic-engine stride-incompatible silent failure, see ",{"type":48,"tag":76,"props":1853,"children":1855},{"className":1854},[],[1856],{"type":54,"value":1857},"references\u002Ferror-patterns.md",{"type":54,"value":107},{"type":48,"tag":152,"props":1860,"children":1862},{"id":1861},"spec-param-parent-model-inference",[1863],{"type":54,"value":1864},"Spec Param \u002F Parent Model Inference",{"type":48,"tag":57,"props":1866,"children":1867},{},[1868,1870,1875,1876,1882,1884,1889,1891,1897],{"type":54,"value":1869},"Model-specific inference mappings (per-action spec field → inference function) for train \u002F evaluate \u002F inference \u002F export \u002F gen_trt_engine, plus ",{"type":48,"tag":76,"props":1871,"children":1873},{"className":1872},[],[1874],{"type":54,"value":1078},{"type":54,"value":205},{"type":48,"tag":76,"props":1877,"children":1879},{"className":1878},[],[1880],{"type":54,"value":1881},"parent_model",{"type":54,"value":1883}," resolution and raw-bp2 explicit-checkpoint handling, are in ",{"type":48,"tag":76,"props":1885,"children":1887},{"className":1886},[],[1888],{"type":54,"value":1086},{"type":54,"value":1890},". Generated runners should read that section and apply the mappings with SDK helpers before ",{"type":48,"tag":76,"props":1892,"children":1894},{"className":1893},[],[1895],{"type":54,"value":1896},"create_job()",{"type":54,"value":107},{"type":48,"tag":152,"props":1899,"children":1901},{"id":1900},"deployment",[1902],{"type":54,"value":1903},"Deployment",{"type":48,"tag":451,"props":1905,"children":1906},{},[1907],{"type":48,"tag":181,"props":1908,"children":1909},{},[1910],{"type":48,"tag":1911,"props":1912,"children":1913},"a",{"href":141},[1914],{"type":54,"value":1915},"tao-deploy-fast-foundation-stereo",{"items":1917,"total":2017},[1918,1933,1945,1959,1971,1988,2003],{"slug":1919,"name":1919,"fn":1920,"description":1921,"org":1922,"tags":1923,"stars":20,"repoUrl":21,"updatedAt":1932},"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},[1924,1927,1930,1931],{"name":1925,"slug":1926,"type":15},"Data Analysis","data-analysis",{"name":1928,"slug":1929,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":13,"slug":14,"type":15},"2026-07-14T05:28:43.176466",{"slug":1934,"name":1934,"fn":1935,"description":1936,"org":1937,"tags":1938,"stars":20,"repoUrl":21,"updatedAt":1944},"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},[1939,1940,1943],{"name":1903,"slug":1900,"type":15},{"name":1941,"slug":1942,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1946,"name":1946,"fn":1947,"description":1948,"org":1949,"tags":1950,"stars":20,"repoUrl":21,"updatedAt":1958},"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},[1951,1954,1955],{"name":1952,"slug":1953,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1956,"slug":1957,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1960,"name":1960,"fn":1961,"description":1962,"org":1963,"tags":1964,"stars":20,"repoUrl":21,"updatedAt":1970},"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},[1965,1966,1967],{"name":1925,"slug":1926,"type":15},{"name":9,"slug":8,"type":15},{"name":1968,"slug":1969,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1972,"name":1972,"fn":1973,"description":1974,"org":1975,"tags":1976,"stars":20,"repoUrl":21,"updatedAt":1987},"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},[1977,1980,1983,1984],{"name":1978,"slug":1979,"type":15},"Automation","automation",{"name":1981,"slug":1982,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1985,"slug":1986,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1989,"name":1989,"fn":1990,"description":1991,"org":1992,"tags":1993,"stars":20,"repoUrl":21,"updatedAt":2002},"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},[1994,1995,1998,1999],{"name":1903,"slug":1900,"type":15},{"name":1996,"slug":1997,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2000,"slug":2001,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2004,"name":2004,"fn":2005,"description":2006,"org":2007,"tags":2008,"stars":20,"repoUrl":21,"updatedAt":2016},"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},[2009,2010,2013],{"name":9,"slug":8,"type":15},{"name":2011,"slug":2012,"type":15},"Quantum Computing","quantum-computing",{"name":2014,"slug":2015,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":2019,"total":2170},[2020,2038,2054,2065,2077,2091,2104,2118,2129,2138,2152,2161],{"slug":2021,"name":2021,"fn":2022,"description":2023,"org":2024,"tags":2025,"stars":2035,"repoUrl":2036,"updatedAt":2037},"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},[2026,2029,2032],{"name":2027,"slug":2028,"type":15},"Documentation","documentation",{"name":2030,"slug":2031,"type":15},"MCP","mcp",{"name":2033,"slug":2034,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":2039,"name":2039,"fn":2040,"description":2041,"org":2042,"tags":2043,"stars":2051,"repoUrl":2052,"updatedAt":2053},"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},[2044,2047,2048],{"name":2045,"slug":2046,"type":15},"Containers","containers",{"name":1903,"slug":1900,"type":15},{"name":2049,"slug":2050,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":2055,"name":2055,"fn":2056,"description":2057,"org":2058,"tags":2059,"stars":2051,"repoUrl":2052,"updatedAt":2064},"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},[2060,2063],{"name":2061,"slug":2062,"type":15},"CI\u002FCD","ci-cd",{"name":1903,"slug":1900,"type":15},"2026-07-14T05:25:59.97109",{"slug":2066,"name":2066,"fn":2067,"description":2068,"org":2069,"tags":2070,"stars":2051,"repoUrl":2052,"updatedAt":2076},"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},[2071,2072,2073],{"name":2061,"slug":2062,"type":15},{"name":1903,"slug":1900,"type":15},{"name":2074,"slug":2075,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":2078,"name":2078,"fn":2079,"description":2080,"org":2081,"tags":2082,"stars":2051,"repoUrl":2052,"updatedAt":2090},"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},[2083,2086,2087],{"name":2084,"slug":2085,"type":15},"Debugging","debugging",{"name":2074,"slug":2075,"type":15},{"name":2088,"slug":2089,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":2092,"name":2092,"fn":2093,"description":2094,"org":2095,"tags":2096,"stars":2051,"repoUrl":2052,"updatedAt":2103},"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},[2097,2100],{"name":2098,"slug":2099,"type":15},"Best Practices","best-practices",{"name":2101,"slug":2102,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":2105,"name":2105,"fn":2106,"description":2107,"org":2108,"tags":2109,"stars":2051,"repoUrl":2052,"updatedAt":2117},"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},[2110,2113,2116],{"name":2111,"slug":2112,"type":15},"Machine Learning","machine-learning",{"name":2114,"slug":2115,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":2119,"name":2119,"fn":2120,"description":2121,"org":2122,"tags":2123,"stars":2051,"repoUrl":2052,"updatedAt":2128},"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},[2124,2127],{"name":2125,"slug":2126,"type":15},"QA","qa",{"name":1968,"slug":1969,"type":15},"2026-07-14T05:25:53.673039",{"slug":2130,"name":2130,"fn":2131,"description":2132,"org":2133,"tags":2134,"stars":2051,"repoUrl":2052,"updatedAt":2137},"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},[2135,2136],{"name":1903,"slug":1900,"type":15},{"name":1941,"slug":1942,"type":15},"2026-07-14T05:25:49.362534",{"slug":2139,"name":2139,"fn":2140,"description":2141,"org":2142,"tags":2143,"stars":2051,"repoUrl":2052,"updatedAt":2151},"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},[2144,2147,2148],{"name":2145,"slug":2146,"type":15},"Code Review","code-review",{"name":2074,"slug":2075,"type":15},{"name":2149,"slug":2150,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":2153,"name":2153,"fn":2154,"description":2155,"org":2156,"tags":2157,"stars":2051,"repoUrl":2052,"updatedAt":2160},"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},[2158,2159],{"name":2125,"slug":2126,"type":15},{"name":1968,"slug":1969,"type":15},"2026-07-14T05:25:54.928983",{"slug":2162,"name":2162,"fn":2163,"description":2164,"org":2165,"tags":2166,"stars":2051,"repoUrl":2052,"updatedAt":2169},"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},[2167,2168],{"name":1978,"slug":1979,"type":15},{"name":2061,"slug":2062,"type":15},"2026-07-30T05:29:03.275638",496]