[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-train-depth-anything-v2":3,"mdc-ofg2pp-key":34,"related-repo-nvidia-tao-train-depth-anything-v2":2719,"related-org-nvidia-tao-train-depth-anything-v2":2821},{"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-depth-anything-v2","train monocular depth estimation models","Monocular depth estimation using Metric Depth Anything v2 or Relative Depth Anything architectures. Predicts per-pixel depth from single RGB images. Use when training, evaluating, exporting, or running inference for a TAO monocular depth model. Trigger phrases include \"train monocular depth\", \"DepthAnything v2\", \"metric depth from single image\", \"monocular depth estimation\".",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,17,20],{"name":13,"slug":14,"type":15},"Deep Learning","deep-learning","tag",{"name":9,"slug":8,"type":15},{"name":18,"slug":19,"type":15},"Imaging","imaging",{"name":21,"slug":22,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:27:12.716518","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-depth-anything-v2","---\nname: tao-train-depth-anything-v2\ndescription: Monocular depth estimation using Metric Depth Anything v2 or Relative Depth Anything architectures. Predicts\n  per-pixel depth from single RGB images. Use when training, evaluating, exporting, or running inference for a TAO\n  monocular depth model. Trigger phrases include \"train monocular depth\", \"DepthAnything v2\", \"metric depth from single\n  image\", \"monocular depth estimation\".\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- monocular\n- depth\n- estimation\n---\n\n# Depth Net Mono\n\nMonocular depth estimation using Metric Depth Anything v2 or Relative Depth Anything architectures. Predicts per-pixel depth from single RGB images.\n\nPretrained checkpoint loading varies by model variant and use case — see the **Pretrained checkpoint loading — use case matrix** in `references\u002Fparameters.md`.\n\nThe mono and stereo skills both invoke the unified TAO `depth_net` CLI inside the container; the mono\u002Fstereo family is selected via `model.model_type` (see `references\u002Fparameters.md`).\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference`), read `references\u002Ftao-deploy-depth-anything-v2.md` first. The deploy spec template lives in this skill's `references\u002Fspec_template_deploy.yaml`.\n\nPyT actions packaged by this model skill: `train`, `evaluate`, `inference`, `export`, and `quantize`. The PyT `depth_net` entrypoint does not accept a PyT-side `gen_trt_engine` action in the current TAO image. The `gen_trt_engine` action metadata must run with the TAO Deploy container, and the deploy workflow remains the deploy-specific entrypoint.\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## Workflow\n\n### Prerequisites — data accessibility\n\nYour dataset (RGB images + GT depth files) must be reachable from inside the container:\n- **SDK runner**: place files at the S3 paths the runner resolves (the `S3_TRAIN` \u002F `S3_EVAL` placeholders shown in **Typical Spec Overrides**). The runner handles S3 → container-path mounting transparently.\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.\n\n### Step 1 — Annotation file\n\nPer-line annotation file referenced by `data_sources[*].data_file`:\n\n| Columns | Format | Use |\n|---|---|---|\n| 1 | `\u003Cimage>` | Mono inference (no GT) |\n| 2 | `\u003Cimage> \u003Cgt_depth>` | Mono with GT |\n\nDo not pass stereo annotation rows such as `\u003Cleft_image> \u003Cright_image>\n\u003Cgt_depth>` directly to mono train\u002Fevaluate\u002Finference. If only a stereo depth\ndataset is available, derive a mono annotation file by keeping the left image\nand GT depth columns, then mount or stage the image\u002Fdepth archive at the same\ncontainer paths referenced by that derived annotation file.\n\nIf you already have one, point to it. Otherwise generate via `depth_net convert`:\n\n```\ndepth_net convert -e \u003Cconvert_spec.yaml>\n```\n\n`convert_spec.yaml` template:\n\n```yaml\nresults_dir: \u003Cdirectory where generated annotation files are written>\ndata_root: \u003Cdirectory whose immediate children are scene\u002Fsample folders that contain your image+depth files; convert walks data_root recursively but expects per-scene subdirectories at one level below>\nimage_dir_pattern: [\u003Csubstring matching left\u002FRGB image paths>]\ndepth_dir_pattern: [\u003Csubstring matching GT depth paths>]\nimage_extension: ''     # optional .endswith filter, e.g. '.jpg'\ndepth_extension: ''     # optional, swapped during depth derivation, e.g. '.png'\nsplit_ratio: 0.0        # 0.0\u002F1.0 = test-only; 0.8 = 80\u002F20 train+val\n```\n\n`convert` walks `data_root` recursively, selects paths whose path-string contains *all* substrings in `image_dir_pattern` (AND-filter), then derives the depth path by replacing `image_dir_pattern[0]` with `depth_dir_pattern[0]` and `image_extension` with `depth_extension`. Inspect your dataset's directory layout and identify the substring distinguishing RGB images from depth files (e.g. `rgb_` vs `sync_depth_`).\n\n`data_root` must point at the parent that contains the per-scene subdirectories (e.g. for NYU eval, use `\u002Fdata\u002Fnyu_v2\u002Feval\u002Ftest`, not `\u002Fdata\u002Fnyu_v2\u002Feval\u002Ftest\u002Fbathroom` — the latter limits the walk to a single scene). Always include the leading dot in `image_extension` \u002F `depth_extension` (e.g. `'.jpg'` not `'jpg'`); the substring swap is form-sensitive and a mismatch silently corrupts derived paths.\n\n### Step 2 — Pair `model_type` and `dataset_name` based on your data\n\nDefault — generic class for each task:\n\n| Data category | `model_type` | `dataset_name` |\n|---|---|---|\n| Disparity-encoded data (pixels) | `RelativeDepthAnything` | `RelativeMonoDataset` |\n| Metric depth (meters) | `MetricDepthAnything` | `MetricMonoDataset` |\n| Mono inference (no GT, any image) | matches train choice | `RelativeMonoDataset` or `MetricMonoDataset` |\n\nDataset-specific class — switch when the data needs preprocessing the generic class does not perform:\n\n| Special case | `model_type` | `dataset_name` | What the class adds |\n|---|---|---|---|\n| NYU `sync_depth_*.png` (raw uint16 millimetres) — relative | `RelativeDepthAnything` | `NYUDV2Relative` | mm→m unit conversion + Eigen evaluation crop |\n| NYU `sync_depth_*.png` (raw uint16 millimetres) — metric | `MetricDepthAnything` | `NYUDV2` | same |\n\nUsing a generic class on data that requires unit conversion (e.g. raw NYU uint16 PNGs) results in an empty valid mask and silent `train_loss = NaN`. Match the class to your data's encoding.\n\nFor relative mono data (`RelativeMonoDataset` or `NYUDV2Relative`), leave `dataset.min_depth` and `dataset.max_depth` unset or set both to `null`. Non-null metric depth ranges are passed into the relative dataset constructor and fail with `BaseRelativeMonoDataset.__init__() got an unexpected keyword argument 'min_depth'`.\n\n### Step 3 — Write spec yaml from Typical Spec Overrides\n\nCopy the action block from **Typical Spec Overrides** (`references\u002Fspec-overrides.md`). Replace:\n- `model.model_type` from Step 2\n- `dataset.\u003C...>.data_sources[*].dataset_name` from Step 2\n- `data_sources[*].data_file` with the path from Step 1 (S3 path under SDK runner, host path for direct docker)\n- For metric finetune: additionally apply the **Metric Variant Finetuning Recipe** in `references\u002Ffinetuning-recipes.md`.\n\nFor mono training set `train.precision: fp32` (recommended) or `bf16` (Ampere SM80+, alternative).\n\n### Step 4 — Run\n\nCreate writable home\u002Fcache directories inside the mounted output path before using\n`--user`. Some TAO containers do not have an `\u002Fetc\u002Fpasswd` entry for the host UID,\nand PyTorch \u002F matplotlib need writable cache paths when running as that UID.\n\n```bash\nmkdir -p \u003Coutput_dir>\u002Fhome \\\n         \u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n         \u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n         \u003Coutput_dir>\u002F.cache\u002Fxdg\n```\n\n```\ndocker run --gpus 'device=0' --shm-size 16G --ipc=host \\\n  --user \"$(id -u):$(id -g)\" \\\n  -e USER=\"$(id -un)\" \\\n  -e LOGNAME=\"$(id -un)\" \\\n  -e HOME=\u003Coutput_dir>\u002Fhome \\\n  -e MPLCONFIGDIR=\u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n  -e TORCHINDUCTOR_CACHE_DIR=\u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n  -e XDG_CACHE_HOME=\u003Coutput_dir>\u002F.cache\u002Fxdg \\\n  -v \u003Cdata_root>:\u003Cdata_root>:ro \\\n  -v \u003Coutput_dir>:\u003Coutput_dir> \\\n  \u003Ccontainer> \\\n  depth_net \u003Caction> -e \u003Cspec.yaml>\n```\n\nWithout `--user \"$(id -u):$(id -g)\"` the container writes outputs as `nobody:nogroup`, blocking host-side cleanup and retry.\n\n### Step 5 — Verify\n\n- Container exit code 0\n- `status.json` `kpi` block populated\n- For `train`: inspect per-step `train_loss` directly — the entrypoint reports `Execution status: PASS` even when `train_loss = NaN` (see the Metric Variant Finetuning Recipe → Sanity-run PASS criteria in `references\u002Ffinetuning-recipes.md`)\n- For `evaluate` \u002F `inference`: artifacts under `results_dir`\n\nFor TAO Deploy TensorRT actions (`gen_trt_engine`, TensorRT `evaluate`, and TensorRT `inference`), read `references\u002Ftao-deploy-depth-anything-v2.md` first. Deploy spec templates live in this skill's `references\u002F` folder with the `spec_template_deploy_*.yaml` prefix.\n\n## Training Requirements\n\n- **Valid `dataset_name` values for mono `data_sources`** (case-insensitive): `ThreeDVLM`, `FSD`, `NvCLIP`, `IssacStereo`, `Crestereo`, `Middlebury`, `NYUDV2`, `NYUDV2Relative`, `RelativeMonoDataset`, `MetricMonoDataset`. `NYUDV2` carries metric depth GT (meters) — pair with `MetricDepthAnything`; `NYUDV2Relative` is the same data with relative-depth conventions — pair with `RelativeDepthAnything`.\n- **Monitoring metric:** val\u002Fd1, val\u002Floss\n- For AutoML sanity runs on the packaged relative-depth smoke data, use `val\u002Fd1` as the primary monitor. `val\u002Floss` can be emitted as `NaN` even when the trainer exits successfully and writes a usable checkpoint, so it is not a reliable AutoML objective unless the run's status metrics show a finite value.\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| quantize | dataset.train_dataset.data_sources | train_datasets | data_file: annotations.txt + dataset_name | Yes |\n| quantize | dataset.val_dataset.data_sources | eval_dataset | data_file: annotations.txt + dataset_name | Yes |\n| quantize | dataset.quant_calibration_dataset.images_dir | train_datasets | images.tar.gz | No |\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** — construct data source paths from the Per-Action Dataset Requirements table above and include them in `spec_overrides`. Each `data_sources` entry is a dict with **two mandatory fields**: `data_file` and `dataset_name`. See `references\u002Fspec-overrides.md` for the full per-action override blocks (`train`, `evaluate`, `export`, `inference`, `quantize`), the `S3_TRAIN` \u002F `S3_EVAL` placeholders, the relative-variant precision recommendation, and the `quantize` known-issue note.\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\nSee `references\u002Fparameters.md` for the full parameter glossary (model, train, dataset, export, and inference keys with options, defaults, and sources) and the **Pretrained checkpoint loading — use case matrix**.\n\n## Finetuning Recipes\n\nSee `references\u002Ffinetuning-recipes.md` for:\n- **Relative Variant Finetuning Recipe** — finetune from a TAO-trained `RelativeDepthAnything` checkpoint (lr `5e-6`, `LambdaLR`, sanity-vs-convergent guidance, deploy LSQ alignment note).\n- **Metric Variant Finetuning Recipe** — checkpoint compatibility, required overrides, the dataset normalization block (`normalize_depth`\u002F`min_depth`\u002F`max_depth`) required in train AND export specs, trainer-enforced defaults, precision, the 1-epoch sanity-run override, and the Sanity-run PASS criteria with the NaN-mitigation order.\n\n## Multi-GPU \u002F Multi-Node\n\n**Launch method:** Lightning-managed (single `python` process, Lightning spawns workers).\n\n| Spec Key | Description | Default |\n|----------|-------------|---------|\n| `train.num_gpus` | Number of GPUs | 1 |\n| `train.gpu_ids` | GPU device indices | [0] |\n| `train.num_nodes` | Number of nodes | 1 |\n| `train.distributed_strategy` | `ddp` or `fsdp` | `ddp` |\n\n- `ddp` with activation checkpointing: `find_unused_parameters=False`\n- `ddp` without: `find_unused_parameters=True`\n- `fsdp` forces precision to FP16\n\n**Multi-node env vars** (set by orchestrator): `WORLD_SIZE`, `NODE_RANK`, `MASTER_ADDR`, `MASTER_PORT`, `NUM_GPU_PER_NODE`.\n\n## Export \u002F TRT Defaults\n\n- TRT data types: FP32, BF16 (Ampere SM80+). FP16 is not supported for the ViT-L mono backbone.\n- Fresh-install TRT precision: `fp32`. BF16 is supported on Ampere SM80+ hardware, but keep smoke tests on FP32 unless the user explicitly requests BF16.\n\n## Hardware\n\nMinimum 1 GPU(s), recommended 2 GPU(s). 24GB+ VRAM per GPU. ViT-Large encoder is memory intensive. Use `fp32` (recommended) or `bf16` (Ampere SM80+, alternative) for training. Activation checkpointing is available for larger inputs.\n\n## Error Patterns\n\nSee `references\u002Ftroubleshooting.md` for the full error-pattern catalog (depth range mismatch, relative dataset rejecting `min_depth`, missing pretrained weights, `encoder` key location, `dataset_name` not in struct, `depth_net_mono` not found, metric variant hyperparameter sourcing, and export ONNX overwrite).\n\n## Spec Param \u002F Parent Model Inference\n\nSee `references\u002Fspec-param-inference.md` for the model-specific inference mappings (the TAO Core `depth_net_mono.config.json` action table), checkpoint-file naming under `\u003Cresults_dir>\u002Ftrain\u002F`, the `dn_model_latest.pth` policy, the parent-`gen_trt_engine` rationale, and the `parent_model` \u002F `parent_job_id` resolution rules.\n\n## Deployment\n\n- [tao-deploy-depth-anything-v2](references\u002Ftao-deploy-depth-anything-v2.md)\n",{"data":35,"body":45},{"name":4,"description":6,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"tags":41},"Requires docker + nvidia-container-toolkit.",{"version":38,"author":39},"0.1.0","NVIDIA Corporation","Read Bash",[42,43,44],"monocular","depth","estimation",{"type":46,"children":47},"root",[48,57,63,85,113,157,219,226,339,370,376,383,388,441,453,466,472,485,562,575,587,596,607,781,857,912,933,938,1051,1056,1168,1181,1229,1235,1254,1305,1326,1332,1353,1488,1497,1518,1524,1615,1660,1666,1825,1831,2041,2046,2152,2158,2182,2188,2205,2211,2222,2289,2295,2313,2442,2487,2532,2538,2559,2565,2583,2589,2631,2637,2695,2701,2713],{"type":49,"tag":50,"props":51,"children":53},"element","h1",{"id":52},"depth-net-mono",[54],{"type":55,"value":56},"text","Depth Net Mono",{"type":49,"tag":58,"props":59,"children":60},"p",{},[61],{"type":55,"value":62},"Monocular depth estimation using Metric Depth Anything v2 or Relative Depth Anything architectures. Predicts per-pixel depth from single RGB images.",{"type":49,"tag":58,"props":64,"children":65},{},[66,68,74,76,83],{"type":55,"value":67},"Pretrained checkpoint loading varies by model variant and use case — see the ",{"type":49,"tag":69,"props":70,"children":71},"strong",{},[72],{"type":55,"value":73},"Pretrained checkpoint loading — use case matrix",{"type":55,"value":75}," in ",{"type":49,"tag":77,"props":78,"children":80},"code",{"className":79},[],[81],{"type":55,"value":82},"references\u002Fparameters.md",{"type":55,"value":84},".",{"type":49,"tag":58,"props":86,"children":87},{},[88,90,96,98,104,106,111],{"type":55,"value":89},"The mono and stereo skills both invoke the unified TAO ",{"type":49,"tag":77,"props":91,"children":93},{"className":92},[],[94],{"type":55,"value":95},"depth_net",{"type":55,"value":97}," CLI inside the container; the mono\u002Fstereo family is selected via ",{"type":49,"tag":77,"props":99,"children":101},{"className":100},[],[102],{"type":55,"value":103},"model.model_type",{"type":55,"value":105}," (see ",{"type":49,"tag":77,"props":107,"children":109},{"className":108},[],[110],{"type":55,"value":82},{"type":55,"value":112},").",{"type":49,"tag":58,"props":114,"children":115},{},[116,118,124,126,132,134,140,142,148,150,156],{"type":55,"value":117},"For TAO Deploy TensorRT actions (",{"type":49,"tag":77,"props":119,"children":121},{"className":120},[],[122],{"type":55,"value":123},"gen_trt_engine",{"type":55,"value":125},", TensorRT ",{"type":49,"tag":77,"props":127,"children":129},{"className":128},[],[130],{"type":55,"value":131},"evaluate",{"type":55,"value":133},", and TensorRT ",{"type":49,"tag":77,"props":135,"children":137},{"className":136},[],[138],{"type":55,"value":139},"inference",{"type":55,"value":141},"), read ",{"type":49,"tag":77,"props":143,"children":145},{"className":144},[],[146],{"type":55,"value":147},"references\u002Ftao-deploy-depth-anything-v2.md",{"type":55,"value":149}," first. The deploy spec template lives in this skill's ",{"type":49,"tag":77,"props":151,"children":153},{"className":152},[],[154],{"type":55,"value":155},"references\u002Fspec_template_deploy.yaml",{"type":55,"value":84},{"type":49,"tag":58,"props":158,"children":159},{},[160,162,168,170,175,176,181,182,188,190,196,198,203,205,210,212,217],{"type":55,"value":161},"PyT actions packaged by this model skill: ",{"type":49,"tag":77,"props":163,"children":165},{"className":164},[],[166],{"type":55,"value":167},"train",{"type":55,"value":169},", ",{"type":49,"tag":77,"props":171,"children":173},{"className":172},[],[174],{"type":55,"value":131},{"type":55,"value":169},{"type":49,"tag":77,"props":177,"children":179},{"className":178},[],[180],{"type":55,"value":139},{"type":55,"value":169},{"type":49,"tag":77,"props":183,"children":185},{"className":184},[],[186],{"type":55,"value":187},"export",{"type":55,"value":189},", and ",{"type":49,"tag":77,"props":191,"children":193},{"className":192},[],[194],{"type":55,"value":195},"quantize",{"type":55,"value":197},". The PyT ",{"type":49,"tag":77,"props":199,"children":201},{"className":200},[],[202],{"type":55,"value":95},{"type":55,"value":204}," entrypoint does not accept a PyT-side ",{"type":49,"tag":77,"props":206,"children":208},{"className":207},[],[209],{"type":55,"value":123},{"type":55,"value":211}," action in the current TAO image. The ",{"type":49,"tag":77,"props":213,"children":215},{"className":214},[],[216],{"type":55,"value":123},{"type":55,"value":218}," action metadata must run with the TAO Deploy container, and the deploy workflow remains the deploy-specific entrypoint.",{"type":49,"tag":220,"props":221,"children":223},"h2",{"id":222},"train-action-policy",[224],{"type":55,"value":225},"Train Action Policy",{"type":49,"tag":58,"props":227,"children":228},{},[229,231,237,239,245,247,253,255,261,263,269,271,277,279,284,285,291,293,299,301,307,309,315,317,323,325,330,332,337],{"type":55,"value":230},"This model is AutoML-enabled at the model layer. Before handling any train-stage request, read ",{"type":49,"tag":77,"props":232,"children":234},{"className":233},[],[235],{"type":55,"value":236},"references\u002Fskill_info.yaml",{"type":55,"value":238}," and resolve the run override from either an explicit ",{"type":49,"tag":77,"props":240,"children":242},{"className":241},[],[243],{"type":55,"value":244},"automl_policy",{"type":55,"value":246}," value or the user's workflow request. Use ",{"type":49,"tag":77,"props":248,"children":250},{"className":249},[],[251],{"type":55,"value":252},"automl_policy: on",{"type":55,"value":254}," by default and only expose ",{"type":49,"tag":77,"props":256,"children":258},{"className":257},[],[259],{"type":55,"value":260},"on",{"type":55,"value":262}," \u002F ",{"type":49,"tag":77,"props":264,"children":266},{"className":265},[],[267],{"type":55,"value":268},"off",{"type":55,"value":270}," in new launch prompts. Treat phrases like \"turn off AutoML\", \"disable AutoML\", \"no HPO\", or \"plain training\" as ",{"type":49,"tag":77,"props":272,"children":274},{"className":273},[],[275],{"type":55,"value":276},"automl_policy: off",{"type":55,"value":278}," for this run only. When ",{"type":49,"tag":77,"props":280,"children":282},{"className":281},[],[283],{"type":55,"value":252},{"type":55,"value":169},{"type":49,"tag":77,"props":286,"children":288},{"className":287},[],[289],{"type":55,"value":290},"automl_enabled: true",{"type":55,"value":292},", and both ",{"type":49,"tag":77,"props":294,"children":296},{"className":295},[],[297],{"type":55,"value":298},"schemas\u002Ftrain.schema.json",{"type":55,"value":300}," and ",{"type":49,"tag":77,"props":302,"children":304},{"className":303},[],[305],{"type":55,"value":306},"references\u002Fspec_template_train.yaml",{"type":55,"value":308}," are packaged, route the train action through ",{"type":49,"tag":77,"props":310,"children":312},{"className":311},[],[313],{"type":55,"value":314},"tao-skill-bank:tao-run-automl",{"type":55,"value":316}," by default with this model's ",{"type":49,"tag":77,"props":318,"children":320},{"className":319},[],[321],{"type":55,"value":322},"skill_dir",{"type":55,"value":324},". Preserve workflow\u002Fapplication overrides for datasets, specs, output directories, GPU\u002Fplatform settings, parent checkpoints, and ",{"type":49,"tag":77,"props":326,"children":328},{"className":327},[],[329],{"type":55,"value":244},{"type":55,"value":331},". Use direct model training only when ",{"type":49,"tag":77,"props":333,"children":335},{"className":334},[],[336],{"type":55,"value":276},{"type":55,"value":338}," or the packaged train schema\u002Ftemplate is missing; in the missing-schema case, report that AutoML is enabled but not runnable for this model until schemas are generated.",{"type":49,"tag":58,"props":340,"children":341},{},[342,344,349,350,355,356,361,363,368],{"type":55,"value":343},"Non-train actions such as ",{"type":49,"tag":77,"props":345,"children":347},{"className":346},[],[348],{"type":55,"value":131},{"type":55,"value":169},{"type":49,"tag":77,"props":351,"children":353},{"className":352},[],[354],{"type":55,"value":139},{"type":55,"value":169},{"type":49,"tag":77,"props":357,"children":359},{"className":358},[],[360],{"type":55,"value":187},{"type":55,"value":362},", and deploy flows stay in this model skill. The per-run ",{"type":49,"tag":77,"props":364,"children":366},{"className":365},[],[367],{"type":55,"value":244},{"type":55,"value":369}," override does not change model metadata.",{"type":49,"tag":220,"props":371,"children":373},{"id":372},"workflow",[374],{"type":55,"value":375},"Workflow",{"type":49,"tag":377,"props":378,"children":380},"h3",{"id":379},"prerequisites-data-accessibility",[381],{"type":55,"value":382},"Prerequisites — data accessibility",{"type":49,"tag":58,"props":384,"children":385},{},[386],{"type":55,"value":387},"Your dataset (RGB images + GT depth files) must be reachable from inside the container:",{"type":49,"tag":389,"props":390,"children":391},"ul",{},[392,425],{"type":49,"tag":393,"props":394,"children":395},"li",{},[396,401,403,409,410,416,418,423],{"type":49,"tag":69,"props":397,"children":398},{},[399],{"type":55,"value":400},"SDK runner",{"type":55,"value":402},": place files at the S3 paths the runner resolves (the ",{"type":49,"tag":77,"props":404,"children":406},{"className":405},[],[407],{"type":55,"value":408},"S3_TRAIN",{"type":55,"value":262},{"type":49,"tag":77,"props":411,"children":413},{"className":412},[],[414],{"type":55,"value":415},"S3_EVAL",{"type":55,"value":417}," placeholders shown in ",{"type":49,"tag":69,"props":419,"children":420},{},[421],{"type":55,"value":422},"Typical Spec Overrides",{"type":55,"value":424},"). The runner handles S3 → container-path mounting transparently.",{"type":49,"tag":393,"props":426,"children":427},{},[428,439],{"type":49,"tag":69,"props":429,"children":430},{},[431,433],{"type":55,"value":432},"Direct ",{"type":49,"tag":77,"props":434,"children":436},{"className":435},[],[437],{"type":55,"value":438},"docker run",{"type":55,"value":440}," (e.g. local testing): mount the host dataset root read-only at the same in-container path:",{"type":49,"tag":442,"props":443,"children":447},"pre",{"className":444,"code":446,"language":55},[445],"language-text","docker run ... -v \u003Chost_data_root>:\u003Chost_data_root>:ro \u003Ccontainer> ...\n",[448],{"type":49,"tag":77,"props":449,"children":451},{"__ignoreMap":450},"",[452],{"type":55,"value":446},{"type":49,"tag":58,"props":454,"children":455},{},[456,458,464],{"type":55,"value":457},"The same accessibility requirement applies to the ",{"type":49,"tag":77,"props":459,"children":461},{"className":460},[],[462],{"type":55,"value":463},"\u003Coutput_dir>",{"type":55,"value":465}," written by all actions.",{"type":49,"tag":377,"props":467,"children":469},{"id":468},"step-1-annotation-file",[470],{"type":55,"value":471},"Step 1 — Annotation file",{"type":49,"tag":58,"props":473,"children":474},{},[475,477,483],{"type":55,"value":476},"Per-line annotation file referenced by ",{"type":49,"tag":77,"props":478,"children":480},{"className":479},[],[481],{"type":55,"value":482},"data_sources[*].data_file",{"type":55,"value":484},":",{"type":49,"tag":486,"props":487,"children":488},"table",{},[489,513],{"type":49,"tag":490,"props":491,"children":492},"thead",{},[493],{"type":49,"tag":494,"props":495,"children":496},"tr",{},[497,503,508],{"type":49,"tag":498,"props":499,"children":500},"th",{},[501],{"type":55,"value":502},"Columns",{"type":49,"tag":498,"props":504,"children":505},{},[506],{"type":55,"value":507},"Format",{"type":49,"tag":498,"props":509,"children":510},{},[511],{"type":55,"value":512},"Use",{"type":49,"tag":514,"props":515,"children":516},"tbody",{},[517,540],{"type":49,"tag":494,"props":518,"children":519},{},[520,526,535],{"type":49,"tag":521,"props":522,"children":523},"td",{},[524],{"type":55,"value":525},"1",{"type":49,"tag":521,"props":527,"children":528},{},[529],{"type":49,"tag":77,"props":530,"children":532},{"className":531},[],[533],{"type":55,"value":534},"\u003Cimage>",{"type":49,"tag":521,"props":536,"children":537},{},[538],{"type":55,"value":539},"Mono inference (no GT)",{"type":49,"tag":494,"props":541,"children":542},{},[543,548,557],{"type":49,"tag":521,"props":544,"children":545},{},[546],{"type":55,"value":547},"2",{"type":49,"tag":521,"props":549,"children":550},{},[551],{"type":49,"tag":77,"props":552,"children":554},{"className":553},[],[555],{"type":55,"value":556},"\u003Cimage> \u003Cgt_depth>",{"type":49,"tag":521,"props":558,"children":559},{},[560],{"type":55,"value":561},"Mono with GT",{"type":49,"tag":58,"props":563,"children":564},{},[565,567,573],{"type":55,"value":566},"Do not pass stereo annotation rows such as ",{"type":49,"tag":77,"props":568,"children":570},{"className":569},[],[571],{"type":55,"value":572},"\u003Cleft_image> \u003Cright_image> \u003Cgt_depth>",{"type":55,"value":574}," directly to mono train\u002Fevaluate\u002Finference. If only a stereo depth\ndataset is available, derive a mono annotation file by keeping the left image\nand GT depth columns, then mount or stage the image\u002Fdepth archive at the same\ncontainer paths referenced by that derived annotation file.",{"type":49,"tag":58,"props":576,"children":577},{},[578,580,586],{"type":55,"value":579},"If you already have one, point to it. Otherwise generate via ",{"type":49,"tag":77,"props":581,"children":583},{"className":582},[],[584],{"type":55,"value":585},"depth_net convert",{"type":55,"value":484},{"type":49,"tag":442,"props":588,"children":591},{"className":589,"code":590,"language":55},[445],"depth_net convert -e \u003Cconvert_spec.yaml>\n",[592],{"type":49,"tag":77,"props":593,"children":594},{"__ignoreMap":450},[595],{"type":55,"value":590},{"type":49,"tag":58,"props":597,"children":598},{},[599,605],{"type":49,"tag":77,"props":600,"children":602},{"className":601},[],[603],{"type":55,"value":604},"convert_spec.yaml",{"type":55,"value":606}," template:",{"type":49,"tag":442,"props":608,"children":612},{"className":609,"code":610,"language":611,"meta":450,"style":450},"language-yaml shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","results_dir: \u003Cdirectory where generated annotation files are written>\ndata_root: \u003Cdirectory whose immediate children are scene\u002Fsample folders that contain your image+depth files; convert walks data_root recursively but expects per-scene subdirectories at one level below>\nimage_dir_pattern: [\u003Csubstring matching left\u002FRGB image paths>]\ndepth_dir_pattern: [\u003Csubstring matching GT depth paths>]\nimage_extension: ''     # optional .endswith filter, e.g. '.jpg'\ndepth_extension: ''     # optional, swapped during depth derivation, e.g. '.png'\nsplit_ratio: 0.0        # 0.0\u002F1.0 = test-only; 0.8 = 80\u002F20 train+val\n","yaml",[613],{"type":49,"tag":77,"props":614,"children":615},{"__ignoreMap":450},[616,639,657,685,711,735,757],{"type":49,"tag":617,"props":618,"children":621},"span",{"class":619,"line":620},"line",1,[622,628,633],{"type":49,"tag":617,"props":623,"children":625},{"style":624},"--shiki-light:#E53935;--shiki-default:#F07178;--shiki-dark:#F07178",[626],{"type":55,"value":627},"results_dir",{"type":49,"tag":617,"props":629,"children":631},{"style":630},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[632],{"type":55,"value":484},{"type":49,"tag":617,"props":634,"children":636},{"style":635},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[637],{"type":55,"value":638}," \u003Cdirectory where generated annotation files are written>\n",{"type":49,"tag":617,"props":640,"children":642},{"class":619,"line":641},2,[643,648,652],{"type":49,"tag":617,"props":644,"children":645},{"style":624},[646],{"type":55,"value":647},"data_root",{"type":49,"tag":617,"props":649,"children":650},{"style":630},[651],{"type":55,"value":484},{"type":49,"tag":617,"props":653,"children":654},{"style":635},[655],{"type":55,"value":656}," \u003Cdirectory whose immediate children are scene\u002Fsample folders that contain your image+depth files; convert walks data_root recursively but expects per-scene subdirectories at one level below>\n",{"type":49,"tag":617,"props":658,"children":660},{"class":619,"line":659},3,[661,666,670,675,680],{"type":49,"tag":617,"props":662,"children":663},{"style":624},[664],{"type":55,"value":665},"image_dir_pattern",{"type":49,"tag":617,"props":667,"children":668},{"style":630},[669],{"type":55,"value":484},{"type":49,"tag":617,"props":671,"children":672},{"style":630},[673],{"type":55,"value":674}," [",{"type":49,"tag":617,"props":676,"children":677},{"style":635},[678],{"type":55,"value":679},"\u003Csubstring matching left\u002FRGB image paths>",{"type":49,"tag":617,"props":681,"children":682},{"style":630},[683],{"type":55,"value":684},"]\n",{"type":49,"tag":617,"props":686,"children":688},{"class":619,"line":687},4,[689,694,698,702,707],{"type":49,"tag":617,"props":690,"children":691},{"style":624},[692],{"type":55,"value":693},"depth_dir_pattern",{"type":49,"tag":617,"props":695,"children":696},{"style":630},[697],{"type":55,"value":484},{"type":49,"tag":617,"props":699,"children":700},{"style":630},[701],{"type":55,"value":674},{"type":49,"tag":617,"props":703,"children":704},{"style":635},[705],{"type":55,"value":706},"\u003Csubstring matching GT depth paths>",{"type":49,"tag":617,"props":708,"children":709},{"style":630},[710],{"type":55,"value":684},{"type":49,"tag":617,"props":712,"children":714},{"class":619,"line":713},5,[715,720,724,729],{"type":49,"tag":617,"props":716,"children":717},{"style":624},[718],{"type":55,"value":719},"image_extension",{"type":49,"tag":617,"props":721,"children":722},{"style":630},[723],{"type":55,"value":484},{"type":49,"tag":617,"props":725,"children":726},{"style":630},[727],{"type":55,"value":728}," ''",{"type":49,"tag":617,"props":730,"children":732},{"style":731},"--shiki-light:#90A4AE;--shiki-light-font-style:italic;--shiki-default:#546E7A;--shiki-default-font-style:italic;--shiki-dark:#676E95;--shiki-dark-font-style:italic",[733],{"type":55,"value":734},"     # optional .endswith filter, e.g. '.jpg'\n",{"type":49,"tag":617,"props":736,"children":738},{"class":619,"line":737},6,[739,744,748,752],{"type":49,"tag":617,"props":740,"children":741},{"style":624},[742],{"type":55,"value":743},"depth_extension",{"type":49,"tag":617,"props":745,"children":746},{"style":630},[747],{"type":55,"value":484},{"type":49,"tag":617,"props":749,"children":750},{"style":630},[751],{"type":55,"value":728},{"type":49,"tag":617,"props":753,"children":754},{"style":731},[755],{"type":55,"value":756},"     # optional, swapped during depth derivation, e.g. '.png'\n",{"type":49,"tag":617,"props":758,"children":760},{"class":619,"line":759},7,[761,766,770,776],{"type":49,"tag":617,"props":762,"children":763},{"style":624},[764],{"type":55,"value":765},"split_ratio",{"type":49,"tag":617,"props":767,"children":768},{"style":630},[769],{"type":55,"value":484},{"type":49,"tag":617,"props":771,"children":773},{"style":772},"--shiki-light:#F76D47;--shiki-default:#F78C6C;--shiki-dark:#F78C6C",[774],{"type":55,"value":775}," 0.0",{"type":49,"tag":617,"props":777,"children":778},{"style":731},[779],{"type":55,"value":780},"        # 0.0\u002F1.0 = test-only; 0.8 = 80\u002F20 train+val\n",{"type":49,"tag":58,"props":782,"children":783},{},[784,790,792,797,799,805,807,812,814,820,822,828,829,834,835,840,842,848,850,856],{"type":49,"tag":77,"props":785,"children":787},{"className":786},[],[788],{"type":55,"value":789},"convert",{"type":55,"value":791}," walks ",{"type":49,"tag":77,"props":793,"children":795},{"className":794},[],[796],{"type":55,"value":647},{"type":55,"value":798}," recursively, selects paths whose path-string contains ",{"type":49,"tag":800,"props":801,"children":802},"em",{},[803],{"type":55,"value":804},"all",{"type":55,"value":806}," substrings in ",{"type":49,"tag":77,"props":808,"children":810},{"className":809},[],[811],{"type":55,"value":665},{"type":55,"value":813}," (AND-filter), then derives the depth path by replacing ",{"type":49,"tag":77,"props":815,"children":817},{"className":816},[],[818],{"type":55,"value":819},"image_dir_pattern[0]",{"type":55,"value":821}," with ",{"type":49,"tag":77,"props":823,"children":825},{"className":824},[],[826],{"type":55,"value":827},"depth_dir_pattern[0]",{"type":55,"value":300},{"type":49,"tag":77,"props":830,"children":832},{"className":831},[],[833],{"type":55,"value":719},{"type":55,"value":821},{"type":49,"tag":77,"props":836,"children":838},{"className":837},[],[839],{"type":55,"value":743},{"type":55,"value":841},". Inspect your dataset's directory layout and identify the substring distinguishing RGB images from depth files (e.g. ",{"type":49,"tag":77,"props":843,"children":845},{"className":844},[],[846],{"type":55,"value":847},"rgb_",{"type":55,"value":849}," vs ",{"type":49,"tag":77,"props":851,"children":853},{"className":852},[],[854],{"type":55,"value":855},"sync_depth_",{"type":55,"value":112},{"type":49,"tag":58,"props":858,"children":859},{},[860,865,867,873,875,881,883,888,889,894,896,902,904,910],{"type":49,"tag":77,"props":861,"children":863},{"className":862},[],[864],{"type":55,"value":647},{"type":55,"value":866}," must point at the parent that contains the per-scene subdirectories (e.g. for NYU eval, use ",{"type":49,"tag":77,"props":868,"children":870},{"className":869},[],[871],{"type":55,"value":872},"\u002Fdata\u002Fnyu_v2\u002Feval\u002Ftest",{"type":55,"value":874},", not ",{"type":49,"tag":77,"props":876,"children":878},{"className":877},[],[879],{"type":55,"value":880},"\u002Fdata\u002Fnyu_v2\u002Feval\u002Ftest\u002Fbathroom",{"type":55,"value":882}," — the latter limits the walk to a single scene). Always include the leading dot in ",{"type":49,"tag":77,"props":884,"children":886},{"className":885},[],[887],{"type":55,"value":719},{"type":55,"value":262},{"type":49,"tag":77,"props":890,"children":892},{"className":891},[],[893],{"type":55,"value":743},{"type":55,"value":895}," (e.g. ",{"type":49,"tag":77,"props":897,"children":899},{"className":898},[],[900],{"type":55,"value":901},"'.jpg'",{"type":55,"value":903}," not ",{"type":49,"tag":77,"props":905,"children":907},{"className":906},[],[908],{"type":55,"value":909},"'jpg'",{"type":55,"value":911},"); the substring swap is form-sensitive and a mismatch silently corrupts derived paths.",{"type":49,"tag":377,"props":913,"children":915},{"id":914},"step-2-pair-model_type-and-dataset_name-based-on-your-data",[916,918,924,925,931],{"type":55,"value":917},"Step 2 — Pair ",{"type":49,"tag":77,"props":919,"children":921},{"className":920},[],[922],{"type":55,"value":923},"model_type",{"type":55,"value":300},{"type":49,"tag":77,"props":926,"children":928},{"className":927},[],[929],{"type":55,"value":930},"dataset_name",{"type":55,"value":932}," based on your data",{"type":49,"tag":58,"props":934,"children":935},{},[936],{"type":55,"value":937},"Default — generic class for each task:",{"type":49,"tag":486,"props":939,"children":940},{},[941,968],{"type":49,"tag":490,"props":942,"children":943},{},[944],{"type":49,"tag":494,"props":945,"children":946},{},[947,952,960],{"type":49,"tag":498,"props":948,"children":949},{},[950],{"type":55,"value":951},"Data category",{"type":49,"tag":498,"props":953,"children":954},{},[955],{"type":49,"tag":77,"props":956,"children":958},{"className":957},[],[959],{"type":55,"value":923},{"type":49,"tag":498,"props":961,"children":962},{},[963],{"type":49,"tag":77,"props":964,"children":966},{"className":965},[],[967],{"type":55,"value":930},{"type":49,"tag":514,"props":969,"children":970},{},[971,997,1023],{"type":49,"tag":494,"props":972,"children":973},{},[974,979,988],{"type":49,"tag":521,"props":975,"children":976},{},[977],{"type":55,"value":978},"Disparity-encoded data (pixels)",{"type":49,"tag":521,"props":980,"children":981},{},[982],{"type":49,"tag":77,"props":983,"children":985},{"className":984},[],[986],{"type":55,"value":987},"RelativeDepthAnything",{"type":49,"tag":521,"props":989,"children":990},{},[991],{"type":49,"tag":77,"props":992,"children":994},{"className":993},[],[995],{"type":55,"value":996},"RelativeMonoDataset",{"type":49,"tag":494,"props":998,"children":999},{},[1000,1005,1014],{"type":49,"tag":521,"props":1001,"children":1002},{},[1003],{"type":55,"value":1004},"Metric depth (meters)",{"type":49,"tag":521,"props":1006,"children":1007},{},[1008],{"type":49,"tag":77,"props":1009,"children":1011},{"className":1010},[],[1012],{"type":55,"value":1013},"MetricDepthAnything",{"type":49,"tag":521,"props":1015,"children":1016},{},[1017],{"type":49,"tag":77,"props":1018,"children":1020},{"className":1019},[],[1021],{"type":55,"value":1022},"MetricMonoDataset",{"type":49,"tag":494,"props":1024,"children":1025},{},[1026,1031,1036],{"type":49,"tag":521,"props":1027,"children":1028},{},[1029],{"type":55,"value":1030},"Mono inference (no GT, any image)",{"type":49,"tag":521,"props":1032,"children":1033},{},[1034],{"type":55,"value":1035},"matches train choice",{"type":49,"tag":521,"props":1037,"children":1038},{},[1039,1044,1046],{"type":49,"tag":77,"props":1040,"children":1042},{"className":1041},[],[1043],{"type":55,"value":996},{"type":55,"value":1045}," or ",{"type":49,"tag":77,"props":1047,"children":1049},{"className":1048},[],[1050],{"type":55,"value":1022},{"type":49,"tag":58,"props":1052,"children":1053},{},[1054],{"type":55,"value":1055},"Dataset-specific class — switch when the data needs preprocessing the generic class does not perform:",{"type":49,"tag":486,"props":1057,"children":1058},{},[1059,1091],{"type":49,"tag":490,"props":1060,"children":1061},{},[1062],{"type":49,"tag":494,"props":1063,"children":1064},{},[1065,1070,1078,1086],{"type":49,"tag":498,"props":1066,"children":1067},{},[1068],{"type":55,"value":1069},"Special case",{"type":49,"tag":498,"props":1071,"children":1072},{},[1073],{"type":49,"tag":77,"props":1074,"children":1076},{"className":1075},[],[1077],{"type":55,"value":923},{"type":49,"tag":498,"props":1079,"children":1080},{},[1081],{"type":49,"tag":77,"props":1082,"children":1084},{"className":1083},[],[1085],{"type":55,"value":930},{"type":49,"tag":498,"props":1087,"children":1088},{},[1089],{"type":55,"value":1090},"What the class adds",{"type":49,"tag":514,"props":1092,"children":1093},{},[1094,1132],{"type":49,"tag":494,"props":1095,"children":1096},{},[1097,1110,1118,1127],{"type":49,"tag":521,"props":1098,"children":1099},{},[1100,1102,1108],{"type":55,"value":1101},"NYU ",{"type":49,"tag":77,"props":1103,"children":1105},{"className":1104},[],[1106],{"type":55,"value":1107},"sync_depth_*.png",{"type":55,"value":1109}," (raw uint16 millimetres) — relative",{"type":49,"tag":521,"props":1111,"children":1112},{},[1113],{"type":49,"tag":77,"props":1114,"children":1116},{"className":1115},[],[1117],{"type":55,"value":987},{"type":49,"tag":521,"props":1119,"children":1120},{},[1121],{"type":49,"tag":77,"props":1122,"children":1124},{"className":1123},[],[1125],{"type":55,"value":1126},"NYUDV2Relative",{"type":49,"tag":521,"props":1128,"children":1129},{},[1130],{"type":55,"value":1131},"mm→m unit conversion + Eigen evaluation crop",{"type":49,"tag":494,"props":1133,"children":1134},{},[1135,1146,1154,1163],{"type":49,"tag":521,"props":1136,"children":1137},{},[1138,1139,1144],{"type":55,"value":1101},{"type":49,"tag":77,"props":1140,"children":1142},{"className":1141},[],[1143],{"type":55,"value":1107},{"type":55,"value":1145}," (raw uint16 millimetres) — metric",{"type":49,"tag":521,"props":1147,"children":1148},{},[1149],{"type":49,"tag":77,"props":1150,"children":1152},{"className":1151},[],[1153],{"type":55,"value":1013},{"type":49,"tag":521,"props":1155,"children":1156},{},[1157],{"type":49,"tag":77,"props":1158,"children":1160},{"className":1159},[],[1161],{"type":55,"value":1162},"NYUDV2",{"type":49,"tag":521,"props":1164,"children":1165},{},[1166],{"type":55,"value":1167},"same",{"type":49,"tag":58,"props":1169,"children":1170},{},[1171,1173,1179],{"type":55,"value":1172},"Using a generic class on data that requires unit conversion (e.g. raw NYU uint16 PNGs) results in an empty valid mask and silent ",{"type":49,"tag":77,"props":1174,"children":1176},{"className":1175},[],[1177],{"type":55,"value":1178},"train_loss = NaN",{"type":55,"value":1180},". Match the class to your data's encoding.",{"type":49,"tag":58,"props":1182,"children":1183},{},[1184,1186,1191,1192,1197,1199,1205,1206,1212,1214,1220,1222,1228],{"type":55,"value":1185},"For relative mono data (",{"type":49,"tag":77,"props":1187,"children":1189},{"className":1188},[],[1190],{"type":55,"value":996},{"type":55,"value":1045},{"type":49,"tag":77,"props":1193,"children":1195},{"className":1194},[],[1196],{"type":55,"value":1126},{"type":55,"value":1198},"), leave ",{"type":49,"tag":77,"props":1200,"children":1202},{"className":1201},[],[1203],{"type":55,"value":1204},"dataset.min_depth",{"type":55,"value":300},{"type":49,"tag":77,"props":1207,"children":1209},{"className":1208},[],[1210],{"type":55,"value":1211},"dataset.max_depth",{"type":55,"value":1213}," unset or set both to ",{"type":49,"tag":77,"props":1215,"children":1217},{"className":1216},[],[1218],{"type":55,"value":1219},"null",{"type":55,"value":1221},". Non-null metric depth ranges are passed into the relative dataset constructor and fail with ",{"type":49,"tag":77,"props":1223,"children":1225},{"className":1224},[],[1226],{"type":55,"value":1227},"BaseRelativeMonoDataset.__init__() got an unexpected keyword argument 'min_depth'",{"type":55,"value":84},{"type":49,"tag":377,"props":1230,"children":1232},{"id":1231},"step-3-write-spec-yaml-from-typical-spec-overrides",[1233],{"type":55,"value":1234},"Step 3 — Write spec yaml from Typical Spec Overrides",{"type":49,"tag":58,"props":1236,"children":1237},{},[1238,1240,1244,1246,1252],{"type":55,"value":1239},"Copy the action block from ",{"type":49,"tag":69,"props":1241,"children":1242},{},[1243],{"type":55,"value":422},{"type":55,"value":1245}," (",{"type":49,"tag":77,"props":1247,"children":1249},{"className":1248},[],[1250],{"type":55,"value":1251},"references\u002Fspec-overrides.md",{"type":55,"value":1253},"). Replace:",{"type":49,"tag":389,"props":1255,"children":1256},{},[1257,1267,1277,1287],{"type":49,"tag":393,"props":1258,"children":1259},{},[1260,1265],{"type":49,"tag":77,"props":1261,"children":1263},{"className":1262},[],[1264],{"type":55,"value":103},{"type":55,"value":1266}," from Step 2",{"type":49,"tag":393,"props":1268,"children":1269},{},[1270,1276],{"type":49,"tag":77,"props":1271,"children":1273},{"className":1272},[],[1274],{"type":55,"value":1275},"dataset.\u003C...>.data_sources[*].dataset_name",{"type":55,"value":1266},{"type":49,"tag":393,"props":1278,"children":1279},{},[1280,1285],{"type":49,"tag":77,"props":1281,"children":1283},{"className":1282},[],[1284],{"type":55,"value":482},{"type":55,"value":1286}," with the path from Step 1 (S3 path under SDK runner, host path for direct docker)",{"type":49,"tag":393,"props":1288,"children":1289},{},[1290,1292,1297,1298,1304],{"type":55,"value":1291},"For metric finetune: additionally apply the ",{"type":49,"tag":69,"props":1293,"children":1294},{},[1295],{"type":55,"value":1296},"Metric Variant Finetuning Recipe",{"type":55,"value":75},{"type":49,"tag":77,"props":1299,"children":1301},{"className":1300},[],[1302],{"type":55,"value":1303},"references\u002Ffinetuning-recipes.md",{"type":55,"value":84},{"type":49,"tag":58,"props":1306,"children":1307},{},[1308,1310,1316,1318,1324],{"type":55,"value":1309},"For mono training set ",{"type":49,"tag":77,"props":1311,"children":1313},{"className":1312},[],[1314],{"type":55,"value":1315},"train.precision: fp32",{"type":55,"value":1317}," (recommended) or ",{"type":49,"tag":77,"props":1319,"children":1321},{"className":1320},[],[1322],{"type":55,"value":1323},"bf16",{"type":55,"value":1325}," (Ampere SM80+, alternative).",{"type":49,"tag":377,"props":1327,"children":1329},{"id":1328},"step-4-run",[1330],{"type":55,"value":1331},"Step 4 — Run",{"type":49,"tag":58,"props":1333,"children":1334},{},[1335,1337,1343,1345,1351],{"type":55,"value":1336},"Create writable home\u002Fcache directories inside the mounted output path before using\n",{"type":49,"tag":77,"props":1338,"children":1340},{"className":1339},[],[1341],{"type":55,"value":1342},"--user",{"type":55,"value":1344},". Some TAO containers do not have an ",{"type":49,"tag":77,"props":1346,"children":1348},{"className":1347},[],[1349],{"type":55,"value":1350},"\u002Fetc\u002Fpasswd",{"type":55,"value":1352}," entry for the host UID,\nand PyTorch \u002F matplotlib need writable cache paths when running as that UID.",{"type":49,"tag":442,"props":1354,"children":1358},{"className":1355,"code":1356,"language":1357,"meta":450,"style":450},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","mkdir -p \u003Coutput_dir>\u002Fhome \\\n         \u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n         \u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n         \u003Coutput_dir>\u002F.cache\u002Fxdg\n","bash",[1359],{"type":49,"tag":77,"props":1360,"children":1361},{"__ignoreMap":450},[1362,1407,1436,1464],{"type":49,"tag":617,"props":1363,"children":1364},{"class":619,"line":620},[1365,1371,1376,1381,1386,1392,1397,1402],{"type":49,"tag":617,"props":1366,"children":1368},{"style":1367},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[1369],{"type":55,"value":1370},"mkdir",{"type":49,"tag":617,"props":1372,"children":1373},{"style":635},[1374],{"type":55,"value":1375}," -p",{"type":49,"tag":617,"props":1377,"children":1378},{"style":630},[1379],{"type":55,"value":1380}," \u003C",{"type":49,"tag":617,"props":1382,"children":1383},{"style":635},[1384],{"type":55,"value":1385},"output_di",{"type":49,"tag":617,"props":1387,"children":1389},{"style":1388},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[1390],{"type":55,"value":1391},"r",{"type":49,"tag":617,"props":1393,"children":1394},{"style":630},[1395],{"type":55,"value":1396},">",{"type":49,"tag":617,"props":1398,"children":1399},{"style":635},[1400],{"type":55,"value":1401},"\u002Fhome",{"type":49,"tag":617,"props":1403,"children":1404},{"style":1388},[1405],{"type":55,"value":1406}," \\\n",{"type":49,"tag":617,"props":1408,"children":1409},{"class":619,"line":641},[1410,1415,1419,1423,1427,1432],{"type":49,"tag":617,"props":1411,"children":1412},{"style":630},[1413],{"type":55,"value":1414},"         \u003C",{"type":49,"tag":617,"props":1416,"children":1417},{"style":635},[1418],{"type":55,"value":1385},{"type":49,"tag":617,"props":1420,"children":1421},{"style":1388},[1422],{"type":55,"value":1391},{"type":49,"tag":617,"props":1424,"children":1425},{"style":630},[1426],{"type":55,"value":1396},{"type":49,"tag":617,"props":1428,"children":1429},{"style":635},[1430],{"type":55,"value":1431},"\u002F.cache\u002Fmatplotlib",{"type":49,"tag":617,"props":1433,"children":1434},{"style":1388},[1435],{"type":55,"value":1406},{"type":49,"tag":617,"props":1437,"children":1438},{"class":619,"line":659},[1439,1443,1447,1451,1455,1460],{"type":49,"tag":617,"props":1440,"children":1441},{"style":630},[1442],{"type":55,"value":1414},{"type":49,"tag":617,"props":1444,"children":1445},{"style":635},[1446],{"type":55,"value":1385},{"type":49,"tag":617,"props":1448,"children":1449},{"style":1388},[1450],{"type":55,"value":1391},{"type":49,"tag":617,"props":1452,"children":1453},{"style":630},[1454],{"type":55,"value":1396},{"type":49,"tag":617,"props":1456,"children":1457},{"style":635},[1458],{"type":55,"value":1459},"\u002F.cache\u002Ftorchinductor",{"type":49,"tag":617,"props":1461,"children":1462},{"style":1388},[1463],{"type":55,"value":1406},{"type":49,"tag":617,"props":1465,"children":1466},{"class":619,"line":687},[1467,1471,1475,1479,1483],{"type":49,"tag":617,"props":1468,"children":1469},{"style":630},[1470],{"type":55,"value":1414},{"type":49,"tag":617,"props":1472,"children":1473},{"style":635},[1474],{"type":55,"value":1385},{"type":49,"tag":617,"props":1476,"children":1477},{"style":1388},[1478],{"type":55,"value":1391},{"type":49,"tag":617,"props":1480,"children":1481},{"style":630},[1482],{"type":55,"value":1396},{"type":49,"tag":617,"props":1484,"children":1485},{"style":635},[1486],{"type":55,"value":1487},"\u002F.cache\u002Fxdg\n",{"type":49,"tag":442,"props":1489,"children":1492},{"className":1490,"code":1491,"language":55},[445],"docker run --gpus 'device=0' --shm-size 16G --ipc=host \\\n  --user \"$(id -u):$(id -g)\" \\\n  -e USER=\"$(id -un)\" \\\n  -e LOGNAME=\"$(id -un)\" \\\n  -e HOME=\u003Coutput_dir>\u002Fhome \\\n  -e MPLCONFIGDIR=\u003Coutput_dir>\u002F.cache\u002Fmatplotlib \\\n  -e TORCHINDUCTOR_CACHE_DIR=\u003Coutput_dir>\u002F.cache\u002Ftorchinductor \\\n  -e XDG_CACHE_HOME=\u003Coutput_dir>\u002F.cache\u002Fxdg \\\n  -v \u003Cdata_root>:\u003Cdata_root>:ro \\\n  -v \u003Coutput_dir>:\u003Coutput_dir> \\\n  \u003Ccontainer> \\\n  depth_net \u003Caction> -e \u003Cspec.yaml>\n",[1493],{"type":49,"tag":77,"props":1494,"children":1495},{"__ignoreMap":450},[1496],{"type":55,"value":1491},{"type":49,"tag":58,"props":1498,"children":1499},{},[1500,1502,1508,1510,1516],{"type":55,"value":1501},"Without ",{"type":49,"tag":77,"props":1503,"children":1505},{"className":1504},[],[1506],{"type":55,"value":1507},"--user \"$(id -u):$(id -g)\"",{"type":55,"value":1509}," the container writes outputs as ",{"type":49,"tag":77,"props":1511,"children":1513},{"className":1512},[],[1514],{"type":55,"value":1515},"nobody:nogroup",{"type":55,"value":1517},", blocking host-side cleanup and retry.",{"type":49,"tag":377,"props":1519,"children":1521},{"id":1520},"step-5-verify",[1522],{"type":55,"value":1523},"Step 5 — Verify",{"type":49,"tag":389,"props":1525,"children":1526},{},[1527,1532,1551,1593],{"type":49,"tag":393,"props":1528,"children":1529},{},[1530],{"type":55,"value":1531},"Container exit code 0",{"type":49,"tag":393,"props":1533,"children":1534},{},[1535,1541,1543,1549],{"type":49,"tag":77,"props":1536,"children":1538},{"className":1537},[],[1539],{"type":55,"value":1540},"status.json",{"type":55,"value":1542}," ",{"type":49,"tag":77,"props":1544,"children":1546},{"className":1545},[],[1547],{"type":55,"value":1548},"kpi",{"type":55,"value":1550}," block populated",{"type":49,"tag":393,"props":1552,"children":1553},{},[1554,1556,1561,1563,1569,1571,1577,1579,1584,1586,1591],{"type":55,"value":1555},"For ",{"type":49,"tag":77,"props":1557,"children":1559},{"className":1558},[],[1560],{"type":55,"value":167},{"type":55,"value":1562},": inspect per-step ",{"type":49,"tag":77,"props":1564,"children":1566},{"className":1565},[],[1567],{"type":55,"value":1568},"train_loss",{"type":55,"value":1570}," directly — the entrypoint reports ",{"type":49,"tag":77,"props":1572,"children":1574},{"className":1573},[],[1575],{"type":55,"value":1576},"Execution status: PASS",{"type":55,"value":1578}," even when ",{"type":49,"tag":77,"props":1580,"children":1582},{"className":1581},[],[1583],{"type":55,"value":1178},{"type":55,"value":1585}," (see the Metric Variant Finetuning Recipe → Sanity-run PASS criteria in ",{"type":49,"tag":77,"props":1587,"children":1589},{"className":1588},[],[1590],{"type":55,"value":1303},{"type":55,"value":1592},")",{"type":49,"tag":393,"props":1594,"children":1595},{},[1596,1597,1602,1603,1608,1610],{"type":55,"value":1555},{"type":49,"tag":77,"props":1598,"children":1600},{"className":1599},[],[1601],{"type":55,"value":131},{"type":55,"value":262},{"type":49,"tag":77,"props":1604,"children":1606},{"className":1605},[],[1607],{"type":55,"value":139},{"type":55,"value":1609},": artifacts under ",{"type":49,"tag":77,"props":1611,"children":1613},{"className":1612},[],[1614],{"type":55,"value":627},{"type":49,"tag":58,"props":1616,"children":1617},{},[1618,1619,1624,1625,1630,1631,1636,1637,1642,1644,1650,1652,1658],{"type":55,"value":117},{"type":49,"tag":77,"props":1620,"children":1622},{"className":1621},[],[1623],{"type":55,"value":123},{"type":55,"value":125},{"type":49,"tag":77,"props":1626,"children":1628},{"className":1627},[],[1629],{"type":55,"value":131},{"type":55,"value":133},{"type":49,"tag":77,"props":1632,"children":1634},{"className":1633},[],[1635],{"type":55,"value":139},{"type":55,"value":141},{"type":49,"tag":77,"props":1638,"children":1640},{"className":1639},[],[1641],{"type":55,"value":147},{"type":55,"value":1643}," first. Deploy spec templates live in this skill's ",{"type":49,"tag":77,"props":1645,"children":1647},{"className":1646},[],[1648],{"type":55,"value":1649},"references\u002F",{"type":55,"value":1651}," folder with the ",{"type":49,"tag":77,"props":1653,"children":1655},{"className":1654},[],[1656],{"type":55,"value":1657},"spec_template_deploy_*.yaml",{"type":55,"value":1659}," prefix.",{"type":49,"tag":220,"props":1661,"children":1663},{"id":1662},"training-requirements",[1664],{"type":55,"value":1665},"Training Requirements",{"type":49,"tag":389,"props":1667,"children":1668},{},[1669,1786,1796],{"type":49,"tag":393,"props":1670,"children":1671},{},[1672,1690,1692,1698,1699,1705,1706,1712,1713,1719,1720,1726,1727,1733,1734,1739,1740,1745,1746,1751,1752,1757,1759,1764,1766,1771,1773,1778,1780,1785],{"type":49,"tag":69,"props":1673,"children":1674},{},[1675,1677,1682,1684],{"type":55,"value":1676},"Valid ",{"type":49,"tag":77,"props":1678,"children":1680},{"className":1679},[],[1681],{"type":55,"value":930},{"type":55,"value":1683}," values for mono ",{"type":49,"tag":77,"props":1685,"children":1687},{"className":1686},[],[1688],{"type":55,"value":1689},"data_sources",{"type":55,"value":1691}," (case-insensitive): ",{"type":49,"tag":77,"props":1693,"children":1695},{"className":1694},[],[1696],{"type":55,"value":1697},"ThreeDVLM",{"type":55,"value":169},{"type":49,"tag":77,"props":1700,"children":1702},{"className":1701},[],[1703],{"type":55,"value":1704},"FSD",{"type":55,"value":169},{"type":49,"tag":77,"props":1707,"children":1709},{"className":1708},[],[1710],{"type":55,"value":1711},"NvCLIP",{"type":55,"value":169},{"type":49,"tag":77,"props":1714,"children":1716},{"className":1715},[],[1717],{"type":55,"value":1718},"IssacStereo",{"type":55,"value":169},{"type":49,"tag":77,"props":1721,"children":1723},{"className":1722},[],[1724],{"type":55,"value":1725},"Crestereo",{"type":55,"value":169},{"type":49,"tag":77,"props":1728,"children":1730},{"className":1729},[],[1731],{"type":55,"value":1732},"Middlebury",{"type":55,"value":169},{"type":49,"tag":77,"props":1735,"children":1737},{"className":1736},[],[1738],{"type":55,"value":1162},{"type":55,"value":169},{"type":49,"tag":77,"props":1741,"children":1743},{"className":1742},[],[1744],{"type":55,"value":1126},{"type":55,"value":169},{"type":49,"tag":77,"props":1747,"children":1749},{"className":1748},[],[1750],{"type":55,"value":996},{"type":55,"value":169},{"type":49,"tag":77,"props":1753,"children":1755},{"className":1754},[],[1756],{"type":55,"value":1022},{"type":55,"value":1758},". ",{"type":49,"tag":77,"props":1760,"children":1762},{"className":1761},[],[1763],{"type":55,"value":1162},{"type":55,"value":1765}," carries metric depth GT (meters) — pair with ",{"type":49,"tag":77,"props":1767,"children":1769},{"className":1768},[],[1770],{"type":55,"value":1013},{"type":55,"value":1772},"; ",{"type":49,"tag":77,"props":1774,"children":1776},{"className":1775},[],[1777],{"type":55,"value":1126},{"type":55,"value":1779}," is the same data with relative-depth conventions — pair with ",{"type":49,"tag":77,"props":1781,"children":1783},{"className":1782},[],[1784],{"type":55,"value":987},{"type":55,"value":84},{"type":49,"tag":393,"props":1787,"children":1788},{},[1789,1794],{"type":49,"tag":69,"props":1790,"children":1791},{},[1792],{"type":55,"value":1793},"Monitoring metric:",{"type":55,"value":1795}," val\u002Fd1, val\u002Floss",{"type":49,"tag":393,"props":1797,"children":1798},{},[1799,1801,1807,1809,1815,1817,1823],{"type":55,"value":1800},"For AutoML sanity runs on the packaged relative-depth smoke data, use ",{"type":49,"tag":77,"props":1802,"children":1804},{"className":1803},[],[1805],{"type":55,"value":1806},"val\u002Fd1",{"type":55,"value":1808}," as the primary monitor. ",{"type":49,"tag":77,"props":1810,"children":1812},{"className":1811},[],[1813],{"type":55,"value":1814},"val\u002Floss",{"type":55,"value":1816}," can be emitted as ",{"type":49,"tag":77,"props":1818,"children":1820},{"className":1819},[],[1821],{"type":55,"value":1822},"NaN",{"type":55,"value":1824}," even when the trainer exits successfully and writes a usable checkpoint, so it is not a reliable AutoML objective unless the run's status metrics show a finite value.",{"type":49,"tag":377,"props":1826,"children":1828},{"id":1827},"per-action-dataset-requirements",[1829],{"type":55,"value":1830},"Per-Action Dataset Requirements",{"type":49,"tag":486,"props":1832,"children":1833},{},[1834,1865],{"type":49,"tag":490,"props":1835,"children":1836},{},[1837],{"type":49,"tag":494,"props":1838,"children":1839},{},[1840,1845,1850,1855,1860],{"type":49,"tag":498,"props":1841,"children":1842},{},[1843],{"type":55,"value":1844},"Action",{"type":49,"tag":498,"props":1846,"children":1847},{},[1848],{"type":55,"value":1849},"Spec Key",{"type":49,"tag":498,"props":1851,"children":1852},{},[1853],{"type":55,"value":1854},"Source",{"type":49,"tag":498,"props":1856,"children":1857},{},[1858],{"type":55,"value":1859},"Files",{"type":49,"tag":498,"props":1861,"children":1862},{},[1863],{"type":55,"value":1864},"List?",{"type":49,"tag":514,"props":1866,"children":1867},{},[1868,1895,1920,1945,1969,1995,2018],{"type":49,"tag":494,"props":1869,"children":1870},{},[1871,1875,1880,1885,1890],{"type":49,"tag":521,"props":1872,"children":1873},{},[1874],{"type":55,"value":131},{"type":49,"tag":521,"props":1876,"children":1877},{},[1878],{"type":55,"value":1879},"dataset.test_dataset.data_sources",{"type":49,"tag":521,"props":1881,"children":1882},{},[1883],{"type":55,"value":1884},"eval_dataset",{"type":49,"tag":521,"props":1886,"children":1887},{},[1888],{"type":55,"value":1889},"data_file: annotations.txt + dataset_name",{"type":49,"tag":521,"props":1891,"children":1892},{},[1893],{"type":55,"value":1894},"Yes",{"type":49,"tag":494,"props":1896,"children":1897},{},[1898,1902,1907,1912,1916],{"type":49,"tag":521,"props":1899,"children":1900},{},[1901],{"type":55,"value":139},{"type":49,"tag":521,"props":1903,"children":1904},{},[1905],{"type":55,"value":1906},"dataset.infer_dataset.data_sources",{"type":49,"tag":521,"props":1908,"children":1909},{},[1910],{"type":55,"value":1911},"inference_dataset",{"type":49,"tag":521,"props":1913,"children":1914},{},[1915],{"type":55,"value":1889},{"type":49,"tag":521,"props":1917,"children":1918},{},[1919],{"type":55,"value":1894},{"type":49,"tag":494,"props":1921,"children":1922},{},[1923,1927,1932,1937,1941],{"type":49,"tag":521,"props":1924,"children":1925},{},[1926],{"type":55,"value":195},{"type":49,"tag":521,"props":1928,"children":1929},{},[1930],{"type":55,"value":1931},"dataset.train_dataset.data_sources",{"type":49,"tag":521,"props":1933,"children":1934},{},[1935],{"type":55,"value":1936},"train_datasets",{"type":49,"tag":521,"props":1938,"children":1939},{},[1940],{"type":55,"value":1889},{"type":49,"tag":521,"props":1942,"children":1943},{},[1944],{"type":55,"value":1894},{"type":49,"tag":494,"props":1946,"children":1947},{},[1948,1952,1957,1961,1965],{"type":49,"tag":521,"props":1949,"children":1950},{},[1951],{"type":55,"value":195},{"type":49,"tag":521,"props":1953,"children":1954},{},[1955],{"type":55,"value":1956},"dataset.val_dataset.data_sources",{"type":49,"tag":521,"props":1958,"children":1959},{},[1960],{"type":55,"value":1884},{"type":49,"tag":521,"props":1962,"children":1963},{},[1964],{"type":55,"value":1889},{"type":49,"tag":521,"props":1966,"children":1967},{},[1968],{"type":55,"value":1894},{"type":49,"tag":494,"props":1970,"children":1971},{},[1972,1976,1981,1985,1990],{"type":49,"tag":521,"props":1973,"children":1974},{},[1975],{"type":55,"value":195},{"type":49,"tag":521,"props":1977,"children":1978},{},[1979],{"type":55,"value":1980},"dataset.quant_calibration_dataset.images_dir",{"type":49,"tag":521,"props":1982,"children":1983},{},[1984],{"type":55,"value":1936},{"type":49,"tag":521,"props":1986,"children":1987},{},[1988],{"type":55,"value":1989},"images.tar.gz",{"type":49,"tag":521,"props":1991,"children":1992},{},[1993],{"type":55,"value":1994},"No",{"type":49,"tag":494,"props":1996,"children":1997},{},[1998,2002,2006,2010,2014],{"type":49,"tag":521,"props":1999,"children":2000},{},[2001],{"type":55,"value":167},{"type":49,"tag":521,"props":2003,"children":2004},{},[2005],{"type":55,"value":1931},{"type":49,"tag":521,"props":2007,"children":2008},{},[2009],{"type":55,"value":1936},{"type":49,"tag":521,"props":2011,"children":2012},{},[2013],{"type":55,"value":1889},{"type":49,"tag":521,"props":2015,"children":2016},{},[2017],{"type":55,"value":1894},{"type":49,"tag":494,"props":2019,"children":2020},{},[2021,2025,2029,2033,2037],{"type":49,"tag":521,"props":2022,"children":2023},{},[2024],{"type":55,"value":167},{"type":49,"tag":521,"props":2026,"children":2027},{},[2028],{"type":55,"value":1956},{"type":49,"tag":521,"props":2030,"children":2031},{},[2032],{"type":55,"value":1884},{"type":49,"tag":521,"props":2034,"children":2035},{},[2036],{"type":55,"value":1889},{"type":49,"tag":521,"props":2038,"children":2039},{},[2040],{"type":55,"value":1894},{"type":49,"tag":377,"props":2042,"children":2044},{"id":2043},"typical-spec-overrides",[2045],{"type":55,"value":422},{"type":49,"tag":58,"props":2047,"children":2048},{},[2049,2051,2056,2058,2064,2066,2071,2073,2078,2080,2086,2087,2092,2094,2099,2101,2106,2107,2112,2113,2118,2119,2124,2125,2130,2132,2137,2138,2143,2145,2150],{"type":55,"value":2050},"Data source overrides are ",{"type":49,"tag":69,"props":2052,"children":2053},{},[2054],{"type":55,"value":2055},"mandatory for every action",{"type":55,"value":2057}," — construct data source paths from the Per-Action Dataset Requirements table above and include them in ",{"type":49,"tag":77,"props":2059,"children":2061},{"className":2060},[],[2062],{"type":55,"value":2063},"spec_overrides",{"type":55,"value":2065},". Each ",{"type":49,"tag":77,"props":2067,"children":2069},{"className":2068},[],[2070],{"type":55,"value":1689},{"type":55,"value":2072}," entry is a dict with ",{"type":49,"tag":69,"props":2074,"children":2075},{},[2076],{"type":55,"value":2077},"two mandatory fields",{"type":55,"value":2079},": ",{"type":49,"tag":77,"props":2081,"children":2083},{"className":2082},[],[2084],{"type":55,"value":2085},"data_file",{"type":55,"value":300},{"type":49,"tag":77,"props":2088,"children":2090},{"className":2089},[],[2091],{"type":55,"value":930},{"type":55,"value":2093},". See ",{"type":49,"tag":77,"props":2095,"children":2097},{"className":2096},[],[2098],{"type":55,"value":1251},{"type":55,"value":2100}," for the full per-action override blocks (",{"type":49,"tag":77,"props":2102,"children":2104},{"className":2103},[],[2105],{"type":55,"value":167},{"type":55,"value":169},{"type":49,"tag":77,"props":2108,"children":2110},{"className":2109},[],[2111],{"type":55,"value":131},{"type":55,"value":169},{"type":49,"tag":77,"props":2114,"children":2116},{"className":2115},[],[2117],{"type":55,"value":187},{"type":55,"value":169},{"type":49,"tag":77,"props":2120,"children":2122},{"className":2121},[],[2123],{"type":55,"value":139},{"type":55,"value":169},{"type":49,"tag":77,"props":2126,"children":2128},{"className":2127},[],[2129],{"type":55,"value":195},{"type":55,"value":2131},"), the ",{"type":49,"tag":77,"props":2133,"children":2135},{"className":2134},[],[2136],{"type":55,"value":408},{"type":55,"value":262},{"type":49,"tag":77,"props":2139,"children":2141},{"className":2140},[],[2142],{"type":55,"value":415},{"type":55,"value":2144}," placeholders, the relative-variant precision recommendation, and the ",{"type":49,"tag":77,"props":2146,"children":2148},{"className":2147},[],[2149],{"type":55,"value":195},{"type":55,"value":2151}," known-issue note.",{"type":49,"tag":220,"props":2153,"children":2155},{"id":2154},"eval-dataset",[2156],{"type":55,"value":2157},"Eval Dataset",{"type":49,"tag":58,"props":2159,"children":2160},{},[2161,2163,2168,2170,2175,2176,2181],{"type":55,"value":2162},"Optional. Val dataset configured via ",{"type":49,"tag":77,"props":2164,"children":2166},{"className":2165},[],[2167],{"type":55,"value":1956},{"type":55,"value":2169}," (each entry needs ",{"type":49,"tag":77,"props":2171,"children":2173},{"className":2172},[],[2174],{"type":55,"value":2085},{"type":55,"value":300},{"type":49,"tag":77,"props":2177,"children":2179},{"className":2178},[],[2180],{"type":55,"value":930},{"type":55,"value":112},{"type":49,"tag":220,"props":2183,"children":2185},{"id":2184},"important-parameters",[2186],{"type":55,"value":2187},"Important Parameters",{"type":49,"tag":58,"props":2189,"children":2190},{},[2191,2193,2198,2200,2204],{"type":55,"value":2192},"See ",{"type":49,"tag":77,"props":2194,"children":2196},{"className":2195},[],[2197],{"type":55,"value":82},{"type":55,"value":2199}," for the full parameter glossary (model, train, dataset, export, and inference keys with options, defaults, and sources) and the ",{"type":49,"tag":69,"props":2201,"children":2202},{},[2203],{"type":55,"value":73},{"type":55,"value":84},{"type":49,"tag":220,"props":2206,"children":2208},{"id":2207},"finetuning-recipes",[2209],{"type":55,"value":2210},"Finetuning Recipes",{"type":49,"tag":58,"props":2212,"children":2213},{},[2214,2215,2220],{"type":55,"value":2192},{"type":49,"tag":77,"props":2216,"children":2218},{"className":2217},[],[2219],{"type":55,"value":1303},{"type":55,"value":2221}," for:",{"type":49,"tag":389,"props":2223,"children":2224},{},[2225,2257],{"type":49,"tag":393,"props":2226,"children":2227},{},[2228,2233,2235,2240,2242,2248,2249,2255],{"type":49,"tag":69,"props":2229,"children":2230},{},[2231],{"type":55,"value":2232},"Relative Variant Finetuning Recipe",{"type":55,"value":2234}," — finetune from a TAO-trained ",{"type":49,"tag":77,"props":2236,"children":2238},{"className":2237},[],[2239],{"type":55,"value":987},{"type":55,"value":2241}," checkpoint (lr ",{"type":49,"tag":77,"props":2243,"children":2245},{"className":2244},[],[2246],{"type":55,"value":2247},"5e-6",{"type":55,"value":169},{"type":49,"tag":77,"props":2250,"children":2252},{"className":2251},[],[2253],{"type":55,"value":2254},"LambdaLR",{"type":55,"value":2256},", sanity-vs-convergent guidance, deploy LSQ alignment note).",{"type":49,"tag":393,"props":2258,"children":2259},{},[2260,2264,2266,2272,2274,2280,2281,2287],{"type":49,"tag":69,"props":2261,"children":2262},{},[2263],{"type":55,"value":1296},{"type":55,"value":2265}," — checkpoint compatibility, required overrides, the dataset normalization block (",{"type":49,"tag":77,"props":2267,"children":2269},{"className":2268},[],[2270],{"type":55,"value":2271},"normalize_depth",{"type":55,"value":2273},"\u002F",{"type":49,"tag":77,"props":2275,"children":2277},{"className":2276},[],[2278],{"type":55,"value":2279},"min_depth",{"type":55,"value":2273},{"type":49,"tag":77,"props":2282,"children":2284},{"className":2283},[],[2285],{"type":55,"value":2286},"max_depth",{"type":55,"value":2288},") required in train AND export specs, trainer-enforced defaults, precision, the 1-epoch sanity-run override, and the Sanity-run PASS criteria with the NaN-mitigation order.",{"type":49,"tag":220,"props":2290,"children":2292},{"id":2291},"multi-gpu-multi-node",[2293],{"type":55,"value":2294},"Multi-GPU \u002F Multi-Node",{"type":49,"tag":58,"props":2296,"children":2297},{},[2298,2303,2305,2311],{"type":49,"tag":69,"props":2299,"children":2300},{},[2301],{"type":55,"value":2302},"Launch method:",{"type":55,"value":2304}," Lightning-managed (single ",{"type":49,"tag":77,"props":2306,"children":2308},{"className":2307},[],[2309],{"type":55,"value":2310},"python",{"type":55,"value":2312}," process, Lightning spawns workers).",{"type":49,"tag":486,"props":2314,"children":2315},{},[2316,2336],{"type":49,"tag":490,"props":2317,"children":2318},{},[2319],{"type":49,"tag":494,"props":2320,"children":2321},{},[2322,2326,2331],{"type":49,"tag":498,"props":2323,"children":2324},{},[2325],{"type":55,"value":1849},{"type":49,"tag":498,"props":2327,"children":2328},{},[2329],{"type":55,"value":2330},"Description",{"type":49,"tag":498,"props":2332,"children":2333},{},[2334],{"type":55,"value":2335},"Default",{"type":49,"tag":514,"props":2337,"children":2338},{},[2339,2360,2385,2406],{"type":49,"tag":494,"props":2340,"children":2341},{},[2342,2351,2356],{"type":49,"tag":521,"props":2343,"children":2344},{},[2345],{"type":49,"tag":77,"props":2346,"children":2348},{"className":2347},[],[2349],{"type":55,"value":2350},"train.num_gpus",{"type":49,"tag":521,"props":2352,"children":2353},{},[2354],{"type":55,"value":2355},"Number of GPUs",{"type":49,"tag":521,"props":2357,"children":2358},{},[2359],{"type":55,"value":525},{"type":49,"tag":494,"props":2361,"children":2362},{},[2363,2372,2377],{"type":49,"tag":521,"props":2364,"children":2365},{},[2366],{"type":49,"tag":77,"props":2367,"children":2369},{"className":2368},[],[2370],{"type":55,"value":2371},"train.gpu_ids",{"type":49,"tag":521,"props":2373,"children":2374},{},[2375],{"type":55,"value":2376},"GPU device indices",{"type":49,"tag":521,"props":2378,"children":2379},{},[2380],{"type":49,"tag":617,"props":2381,"children":2382},{},[2383],{"type":55,"value":2384},"0",{"type":49,"tag":494,"props":2386,"children":2387},{},[2388,2397,2402],{"type":49,"tag":521,"props":2389,"children":2390},{},[2391],{"type":49,"tag":77,"props":2392,"children":2394},{"className":2393},[],[2395],{"type":55,"value":2396},"train.num_nodes",{"type":49,"tag":521,"props":2398,"children":2399},{},[2400],{"type":55,"value":2401},"Number of nodes",{"type":49,"tag":521,"props":2403,"children":2404},{},[2405],{"type":55,"value":525},{"type":49,"tag":494,"props":2407,"children":2408},{},[2409,2418,2434],{"type":49,"tag":521,"props":2410,"children":2411},{},[2412],{"type":49,"tag":77,"props":2413,"children":2415},{"className":2414},[],[2416],{"type":55,"value":2417},"train.distributed_strategy",{"type":49,"tag":521,"props":2419,"children":2420},{},[2421,2427,2428],{"type":49,"tag":77,"props":2422,"children":2424},{"className":2423},[],[2425],{"type":55,"value":2426},"ddp",{"type":55,"value":1045},{"type":49,"tag":77,"props":2429,"children":2431},{"className":2430},[],[2432],{"type":55,"value":2433},"fsdp",{"type":49,"tag":521,"props":2435,"children":2436},{},[2437],{"type":49,"tag":77,"props":2438,"children":2440},{"className":2439},[],[2441],{"type":55,"value":2426},{"type":49,"tag":389,"props":2443,"children":2444},{},[2445,2461,2477],{"type":49,"tag":393,"props":2446,"children":2447},{},[2448,2453,2455],{"type":49,"tag":77,"props":2449,"children":2451},{"className":2450},[],[2452],{"type":55,"value":2426},{"type":55,"value":2454}," with activation checkpointing: ",{"type":49,"tag":77,"props":2456,"children":2458},{"className":2457},[],[2459],{"type":55,"value":2460},"find_unused_parameters=False",{"type":49,"tag":393,"props":2462,"children":2463},{},[2464,2469,2471],{"type":49,"tag":77,"props":2465,"children":2467},{"className":2466},[],[2468],{"type":55,"value":2426},{"type":55,"value":2470}," without: ",{"type":49,"tag":77,"props":2472,"children":2474},{"className":2473},[],[2475],{"type":55,"value":2476},"find_unused_parameters=True",{"type":49,"tag":393,"props":2478,"children":2479},{},[2480,2485],{"type":49,"tag":77,"props":2481,"children":2483},{"className":2482},[],[2484],{"type":55,"value":2433},{"type":55,"value":2486}," forces precision to FP16",{"type":49,"tag":58,"props":2488,"children":2489},{},[2490,2495,2497,2503,2504,2510,2511,2517,2518,2524,2525,2531],{"type":49,"tag":69,"props":2491,"children":2492},{},[2493],{"type":55,"value":2494},"Multi-node env vars",{"type":55,"value":2496}," (set by orchestrator): ",{"type":49,"tag":77,"props":2498,"children":2500},{"className":2499},[],[2501],{"type":55,"value":2502},"WORLD_SIZE",{"type":55,"value":169},{"type":49,"tag":77,"props":2505,"children":2507},{"className":2506},[],[2508],{"type":55,"value":2509},"NODE_RANK",{"type":55,"value":169},{"type":49,"tag":77,"props":2512,"children":2514},{"className":2513},[],[2515],{"type":55,"value":2516},"MASTER_ADDR",{"type":55,"value":169},{"type":49,"tag":77,"props":2519,"children":2521},{"className":2520},[],[2522],{"type":55,"value":2523},"MASTER_PORT",{"type":55,"value":169},{"type":49,"tag":77,"props":2526,"children":2528},{"className":2527},[],[2529],{"type":55,"value":2530},"NUM_GPU_PER_NODE",{"type":55,"value":84},{"type":49,"tag":220,"props":2533,"children":2535},{"id":2534},"export-trt-defaults",[2536],{"type":55,"value":2537},"Export \u002F TRT Defaults",{"type":49,"tag":389,"props":2539,"children":2540},{},[2541,2546],{"type":49,"tag":393,"props":2542,"children":2543},{},[2544],{"type":55,"value":2545},"TRT data types: FP32, BF16 (Ampere SM80+). FP16 is not supported for the ViT-L mono backbone.",{"type":49,"tag":393,"props":2547,"children":2548},{},[2549,2551,2557],{"type":55,"value":2550},"Fresh-install TRT precision: ",{"type":49,"tag":77,"props":2552,"children":2554},{"className":2553},[],[2555],{"type":55,"value":2556},"fp32",{"type":55,"value":2558},". BF16 is supported on Ampere SM80+ hardware, but keep smoke tests on FP32 unless the user explicitly requests BF16.",{"type":49,"tag":220,"props":2560,"children":2562},{"id":2561},"hardware",[2563],{"type":55,"value":2564},"Hardware",{"type":49,"tag":58,"props":2566,"children":2567},{},[2568,2570,2575,2576,2581],{"type":55,"value":2569},"Minimum 1 GPU(s), recommended 2 GPU(s). 24GB+ VRAM per GPU. ViT-Large encoder is memory intensive. Use ",{"type":49,"tag":77,"props":2571,"children":2573},{"className":2572},[],[2574],{"type":55,"value":2556},{"type":55,"value":1317},{"type":49,"tag":77,"props":2577,"children":2579},{"className":2578},[],[2580],{"type":55,"value":1323},{"type":55,"value":2582}," (Ampere SM80+, alternative) for training. Activation checkpointing is available for larger inputs.",{"type":49,"tag":220,"props":2584,"children":2586},{"id":2585},"error-patterns",[2587],{"type":55,"value":2588},"Error Patterns",{"type":49,"tag":58,"props":2590,"children":2591},{},[2592,2593,2599,2601,2606,2608,2614,2616,2621,2623,2629],{"type":55,"value":2192},{"type":49,"tag":77,"props":2594,"children":2596},{"className":2595},[],[2597],{"type":55,"value":2598},"references\u002Ftroubleshooting.md",{"type":55,"value":2600}," for the full error-pattern catalog (depth range mismatch, relative dataset rejecting ",{"type":49,"tag":77,"props":2602,"children":2604},{"className":2603},[],[2605],{"type":55,"value":2279},{"type":55,"value":2607},", missing pretrained weights, ",{"type":49,"tag":77,"props":2609,"children":2611},{"className":2610},[],[2612],{"type":55,"value":2613},"encoder",{"type":55,"value":2615}," key location, ",{"type":49,"tag":77,"props":2617,"children":2619},{"className":2618},[],[2620],{"type":55,"value":930},{"type":55,"value":2622}," not in struct, ",{"type":49,"tag":77,"props":2624,"children":2626},{"className":2625},[],[2627],{"type":55,"value":2628},"depth_net_mono",{"type":55,"value":2630}," not found, metric variant hyperparameter sourcing, and export ONNX overwrite).",{"type":49,"tag":220,"props":2632,"children":2634},{"id":2633},"spec-param-parent-model-inference",[2635],{"type":55,"value":2636},"Spec Param \u002F Parent Model Inference",{"type":49,"tag":58,"props":2638,"children":2639},{},[2640,2641,2647,2649,2655,2657,2663,2665,2671,2673,2678,2680,2686,2687,2693],{"type":55,"value":2192},{"type":49,"tag":77,"props":2642,"children":2644},{"className":2643},[],[2645],{"type":55,"value":2646},"references\u002Fspec-param-inference.md",{"type":55,"value":2648}," for the model-specific inference mappings (the TAO Core ",{"type":49,"tag":77,"props":2650,"children":2652},{"className":2651},[],[2653],{"type":55,"value":2654},"depth_net_mono.config.json",{"type":55,"value":2656}," action table), checkpoint-file naming under ",{"type":49,"tag":77,"props":2658,"children":2660},{"className":2659},[],[2661],{"type":55,"value":2662},"\u003Cresults_dir>\u002Ftrain\u002F",{"type":55,"value":2664},", the ",{"type":49,"tag":77,"props":2666,"children":2668},{"className":2667},[],[2669],{"type":55,"value":2670},"dn_model_latest.pth",{"type":55,"value":2672}," policy, the parent-",{"type":49,"tag":77,"props":2674,"children":2676},{"className":2675},[],[2677],{"type":55,"value":123},{"type":55,"value":2679}," rationale, and the ",{"type":49,"tag":77,"props":2681,"children":2683},{"className":2682},[],[2684],{"type":55,"value":2685},"parent_model",{"type":55,"value":262},{"type":49,"tag":77,"props":2688,"children":2690},{"className":2689},[],[2691],{"type":55,"value":2692},"parent_job_id",{"type":55,"value":2694}," resolution rules.",{"type":49,"tag":220,"props":2696,"children":2698},{"id":2697},"deployment",[2699],{"type":55,"value":2700},"Deployment",{"type":49,"tag":389,"props":2702,"children":2703},{},[2704],{"type":49,"tag":393,"props":2705,"children":2706},{},[2707],{"type":49,"tag":2708,"props":2709,"children":2710},"a",{"href":147},[2711],{"type":55,"value":2712},"tao-deploy-depth-anything-v2",{"type":49,"tag":2714,"props":2715,"children":2716},"style",{},[2717],{"type":55,"value":2718},"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":2720,"total":2820},[2721,2738,2750,2764,2776,2791,2806],{"slug":2722,"name":2722,"fn":2723,"description":2724,"org":2725,"tags":2726,"stars":23,"repoUrl":24,"updatedAt":2737},"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},[2727,2730,2733,2734],{"name":2728,"slug":2729,"type":15},"Data Analysis","data-analysis",{"name":2731,"slug":2732,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":2735,"slug":2736,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":2739,"name":2739,"fn":2740,"description":2741,"org":2742,"tags":2743,"stars":23,"repoUrl":24,"updatedAt":2749},"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},[2744,2745,2748],{"name":2700,"slug":2697,"type":15},{"name":2746,"slug":2747,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":2751,"name":2751,"fn":2752,"description":2753,"org":2754,"tags":2755,"stars":23,"repoUrl":24,"updatedAt":2763},"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},[2756,2759,2760],{"name":2757,"slug":2758,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":2761,"slug":2762,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":2765,"name":2765,"fn":2766,"description":2767,"org":2768,"tags":2769,"stars":23,"repoUrl":24,"updatedAt":2775},"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},[2770,2771,2772],{"name":2728,"slug":2729,"type":15},{"name":9,"slug":8,"type":15},{"name":2773,"slug":2774,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":2777,"name":2777,"fn":2778,"description":2779,"org":2780,"tags":2781,"stars":23,"repoUrl":24,"updatedAt":2790},"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},[2782,2785,2786,2787],{"name":2783,"slug":2784,"type":15},"Automation","automation",{"name":18,"slug":19,"type":15},{"name":9,"slug":8,"type":15},{"name":2788,"slug":2789,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":2792,"name":2792,"fn":2793,"description":2794,"org":2795,"tags":2796,"stars":23,"repoUrl":24,"updatedAt":2805},"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},[2797,2798,2801,2802],{"name":2700,"slug":2697,"type":15},{"name":2799,"slug":2800,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":2803,"slug":2804,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":2807,"name":2807,"fn":2808,"description":2809,"org":2810,"tags":2811,"stars":23,"repoUrl":24,"updatedAt":2819},"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},[2812,2813,2816],{"name":9,"slug":8,"type":15},{"name":2814,"slug":2815,"type":15},"Quantum Computing","quantum-computing",{"name":2817,"slug":2818,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":2822,"total":2972},[2823,2841,2856,2867,2879,2893,2906,2920,2931,2940,2954,2963],{"slug":2824,"name":2824,"fn":2825,"description":2826,"org":2827,"tags":2828,"stars":2838,"repoUrl":2839,"updatedAt":2840},"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},[2829,2832,2835],{"name":2830,"slug":2831,"type":15},"Documentation","documentation",{"name":2833,"slug":2834,"type":15},"MCP","mcp",{"name":2836,"slug":2837,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":2842,"name":2842,"fn":2843,"description":2844,"org":2845,"tags":2846,"stars":2853,"repoUrl":2854,"updatedAt":2855},"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},[2847,2850,2851],{"name":2848,"slug":2849,"type":15},"Containers","containers",{"name":2700,"slug":2697,"type":15},{"name":2852,"slug":2310,"type":15},"Python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":2857,"name":2857,"fn":2858,"description":2859,"org":2860,"tags":2861,"stars":2853,"repoUrl":2854,"updatedAt":2866},"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},[2862,2865],{"name":2863,"slug":2864,"type":15},"CI\u002FCD","ci-cd",{"name":2700,"slug":2697,"type":15},"2026-07-14T05:25:59.97109",{"slug":2868,"name":2868,"fn":2869,"description":2870,"org":2871,"tags":2872,"stars":2853,"repoUrl":2854,"updatedAt":2878},"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},[2873,2874,2875],{"name":2863,"slug":2864,"type":15},{"name":2700,"slug":2697,"type":15},{"name":2876,"slug":2877,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":2880,"name":2880,"fn":2881,"description":2882,"org":2883,"tags":2884,"stars":2853,"repoUrl":2854,"updatedAt":2892},"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},[2885,2888,2889],{"name":2886,"slug":2887,"type":15},"Debugging","debugging",{"name":2876,"slug":2877,"type":15},{"name":2890,"slug":2891,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":2894,"name":2894,"fn":2895,"description":2896,"org":2897,"tags":2898,"stars":2853,"repoUrl":2854,"updatedAt":2905},"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},[2899,2902],{"name":2900,"slug":2901,"type":15},"Best Practices","best-practices",{"name":2903,"slug":2904,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":2907,"name":2907,"fn":2908,"description":2909,"org":2910,"tags":2911,"stars":2853,"repoUrl":2854,"updatedAt":2919},"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},[2912,2915,2918],{"name":2913,"slug":2914,"type":15},"Machine Learning","machine-learning",{"name":2916,"slug":2917,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":2921,"name":2921,"fn":2922,"description":2923,"org":2924,"tags":2925,"stars":2853,"repoUrl":2854,"updatedAt":2930},"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},[2926,2929],{"name":2927,"slug":2928,"type":15},"QA","qa",{"name":2773,"slug":2774,"type":15},"2026-07-14T05:25:53.673039",{"slug":2932,"name":2932,"fn":2933,"description":2934,"org":2935,"tags":2936,"stars":2853,"repoUrl":2854,"updatedAt":2939},"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},[2937,2938],{"name":2700,"slug":2697,"type":15},{"name":2746,"slug":2747,"type":15},"2026-07-14T05:25:49.362534",{"slug":2941,"name":2941,"fn":2942,"description":2943,"org":2944,"tags":2945,"stars":2853,"repoUrl":2854,"updatedAt":2953},"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},[2946,2949,2950],{"name":2947,"slug":2948,"type":15},"Code Review","code-review",{"name":2876,"slug":2877,"type":15},{"name":2951,"slug":2952,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":2955,"name":2955,"fn":2956,"description":2957,"org":2958,"tags":2959,"stars":2853,"repoUrl":2854,"updatedAt":2962},"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},[2960,2961],{"name":2927,"slug":2928,"type":15},{"name":2773,"slug":2774,"type":15},"2026-07-14T05:25:54.928983",{"slug":2964,"name":2964,"fn":2965,"description":2966,"org":2967,"tags":2968,"stars":2853,"repoUrl":2854,"updatedAt":2971},"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},[2969,2970],{"name":2783,"slug":2784,"type":15},{"name":2863,"slug":2864,"type":15},"2026-07-30T05:29:03.275638",496]