[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-nvidia-tao-generate-image-grounding":3,"mdc-gwrx36-key":34,"related-repo-nvidia-tao-generate-image-grounding":1447,"related-org-nvidia-tao-generate-image-grounding":1553},{"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-generate-image-grounding","ground image captions to pixel-space bounding boxes","Two-step image grounding pipeline: extracts referring expressions from (image, caption) pairs and grounds them to pixel-space bounding boxes via a VLM. Use when the user wants to ground captions to bboxes, generate phrase-grounded annotations, auto-label images for grounding, or run the image_grounding pipeline. Triggers include 'image grounding', 'phrase grounding', 'ground captions', 'auto-label image grounding', 'image_grounding'.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":9},"nvidia","NVIDIA","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fnvidia.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"LLM","llm","tag",{"name":17,"slug":18,"type":15},"Images","images",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Computer Vision","computer-vision",2473,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002Fskills","2026-07-14T05:30:34.160505","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-generate-image-grounding","---\nname: tao-generate-image-grounding\ndescription: \"Two-step image grounding pipeline: extracts referring expressions from (image, caption) pairs and grounds them\n  to pixel-space bounding boxes via a VLM. Use when the user wants to ground captions to bboxes, generate phrase-grounded\n  annotations, auto-label images for grounding, or run the image_grounding pipeline. Triggers include 'image grounding',\n  'phrase grounding', 'ground captions', 'auto-label image grounding', 'image_grounding'.\"\nlicense: Apache-2.0\ncompatibility: Requires docker + nvidia-container-toolkit + at least one VLM endpoint (Gemini API key or OpenAI-compatible).\nmetadata:\n  author: NVIDIA Corporation\n  version: \"0.1.0\"\nallowed-tools: Read Bash Write\ntags:\n  - image\n  - grounding\n  - bounding-boxes\n  - auto-label\n  - vlm\n  - 2d-grounding\n---\n\n# Image Grounding Pipeline\n\nTurn `(image, caption)` pairs into per-image grounded annotations: cleaned captions, referring expressions with character spans, and pixel-space bounding boxes for each expression. A single VLM (Gemini or any OpenAI-compatible endpoint) handles both steps.\n\n## Purpose\n\nGenerate phrase-grounded training data for referring-expression and grounding models. The VLM acts as a \"teacher\" annotator: Step 0 extracts referring expressions from the caption while looking at the image; Step 1 returns one bbox set per expression for each image.\n\n## Pipeline Architecture\n\n```\nStep 0: Expression extraction  → VLM cleans caption, extracts referring expressions + char spans\nStep 1: Phrase grounding       → VLM returns pixel bboxes + scores per expression\n```\n\nSteps are individually selectable via `workflow.steps`. Each step writes a per-sample checkpoint to `step_\u003CN>_*\u002F.ckpt\u002F\u003Csample_id>.json` and skips already-processed records on re-run. Set `workflow.force_reprocess: true` to ignore checkpoints and reprocess from scratch.\n\n## Instructions\n\n### Initial setup\n\nWhen a user wants to run this pipeline, walk through these steps:\n\n1. **Input JSONL**: Ask for the JSONL path. Each line must be one object like `{\"image_path\": \"...\", \"caption\": \"...\"}`. `image_path` can be absolute or relative.\n2. **Image root**: If any `image_path` values are relative, set `data.image_root` to the directory they should resolve from.\n3. **API access**: Ask the user which VLM endpoint they want to use. Present these five options and act on the choice:\n   1. **Gemini** — set `vlm.backend: \"gemini\"`; require `GOOGLE_API_KEY` (env var or `vlm.gemini.api_key`).\n   2. **NIM** (e.g. `https:\u002F\u002Finference-api.nvidia.com\u002Fv1`) — set `vlm.backend: \"openai\"`; collect `base_url`, `model_name`, and `api_key`.\n   3. **TAO inference microservice** (self-hosted, OpenAI-compatible). Confirm whether the server is already running:\n      - **Running** — collect `base_url`, `model_name`, and (optionally) `api_key`; set `vlm.backend: \"openai\"`.\n      - **Not running** — guide the user through the `skills\u002Fapplications\u002Ftao-run-inference-service` skill, which stands up a local TAO inference microservice with an OpenAI-compatible API. Before promising a specific model, check `skills\u002Fapplications\u002Ftao-run-inference-service\u002Freferences\u002Fservice.yaml` for `valid_network_arch_config_basenames`. Once the server is up, collect `base_url`, `model_name`, and (optionally) `api_key`; set `vlm.backend: \"openai\"`.\n   4. **vLLM** (self-hosted, OpenAI-compatible). Confirm whether the server is already running:\n      - **Running** — collect `base_url`, `model_name`, and (optionally) `api_key`; set `vlm.backend: \"openai\"`.\n      - **Not running** — follow [references\u002Fvllm_server.md](references\u002Fvllm_server.md) to install and launch a vLLM server, then collect `base_url`, `model_name`, and (optionally) `api_key`; set `vlm.backend: \"openai\"`.\n   5. **Custom** (any other OpenAI-compatible endpoint) — set `vlm.backend: \"openai\"`; collect `base_url`, `model_name`, and (optionally) `api_key`.\n\n   If the user has no endpoint and does not want to set one up, stop and help resolve API access first.\n4. **Workflow steps**: Choose one of:\n   - Full pipeline: `[\"0\", \"1\"]`\n   - Expression extraction only: `[\"0\"]`\n   - Grounding only: `[\"1\"]`, which requires existing step-0 output at `results_dir\u002Fstep_0_expression_extraction\u002Fannotations.jsonl`\n5. **Resume vs fresh run**: By default, the workflow reuses checkpoints and skips completed records. To reprocess everything, set `image_grounding.workflow.force_reprocess=true`.\n\n### Running the pipeline\n\nThe pipeline runs inside the TAO Toolkit container via the `auto_label` CLI:\n\n```bash\nauto_label generate -e \u002Fpath\u002Fto\u002Fspec.yaml \\\n    results_dir=\u002Fresults \\\n    image_grounding.data.input_jsonl=\u002Fdata\u002Fcaptions.jsonl \\\n    image_grounding.data.image_root=\u002Fdata\u002Fimages \\\n    image_grounding.vlm.gemini.api_key=$GOOGLE_API_KEY\n```\n\nGenerate a default spec: `auto_label default_specs results_dir=\u002Fresults module_name=auto_label`, then set `autolabel_type: \"image_grounding\"`. All fields support Hydra dot-notation overrides on the command line.\n\nSee [references\u002Fconfiguration.md](references\u002Fconfiguration.md) for the full YAML structure, all parameters, model\u002Fendpoint setup, and error patterns.\n\n### Recommended pilot workflow\n\n1. Run on 5-10 images with both steps\n2. Inspect `step_0_expression_extraction\u002Fannotations.jsonl` — are `cleaned_caption` and `expressions[]` accurate? Are the right noun phrases captured?\n3. Inspect `step_1_grounding\u002Fannotations.jsonl` — do the bboxes in `expressions[].instances[]` look right? Are confidence scores reasonable?\n4. If quality is insufficient, switch the VLM to a stronger model (e.g. `gemini-2.5-pro`) or raise `media_resolution`\u002F`max_output_tokens`, then re-run with `force_reprocess=true`.\n5. Scale to the full dataset once satisfied.\n\n## Configuration\n\nKey configuration fields (full reference in [references\u002Fconfiguration.md](references\u002Fconfiguration.md)):\n\n| Field | Default | Description |\n|-------|---------|-------------|\n| `workflow.steps` | `[\"0\",\"1\"]` | Which pipeline steps to execute (`\"0\"` = expressions, `\"1\"` = grounding) |\n| `workflow.max_workers` | `4` | Parallel threads per step (watch API rate limits) |\n| `workflow.force_reprocess` | `false` | Ignore per-sample checkpoints and reprocess from scratch |\n| `vlm.backend` | `\"gemini\"` | `\"gemini\"` or `\"openai\"` (OpenAI-compatible endpoint) |\n| `data.input_jsonl` | required | Path to input JSONL with `image_path` + `caption` per line |\n| `data.image_root` | `\"\"` | Optional prefix for resolving relative `image_path` entries |\n\n## Inputs\n\nA single JSONL file at `data.input_jsonl`. One JSON object per line:\n\n| Field | Required | Description |\n|-------|----------|-------------|\n| `image_path` | yes | Absolute path, or relative path resolved against `data.image_root` |\n| `caption` | yes | Free-text caption for the image |\n| `image_id` | no | Stable identifier; auto-derived from the filename if missing |\n| `width`, `height` | no | Image dimensions in pixels; default to `1920×1080` for bbox clamping if missing |\n\n## Outputs\n\nAll outputs go to `results_dir\u002F`:\n\n- `step_0_expression_extraction\u002Fannotations.jsonl` — per-record output enriched with `cleaned_caption` and `expressions[]` (each with `text`, `expression_id`, `char_span`, `noun_chunk`, empty `instances[]`).\n- `step_1_grounding\u002Fannotations.jsonl` — same records with `expressions[].instances[]` filled in (each instance has `bbox: [x1,y1,x2,y2]` in pixel space, `score` in `[0.0, 1.0]`, and `bbox_id`).\n- `results_dir\u002Fannotations.jsonl` — copy of the last step's output for convenience.\n- `step_\u003CN>_*\u002F.ckpt\u002F\u003Csample_id>.json` — per-sample checkpoints used for resume.\n\n## Prerequisites\n\n- **Container**: `nvcr.io\u002Fnvidia\u002Ftao\u002Ftao-toolkit:6.26.3-pyt`\n- **API access**: At least one VLM endpoint (Gemini API key or OpenAI-compatible endpoint capable of image input)\n",{"data":35,"body":48},{"name":4,"description":6,"license":26,"compatibility":36,"metadata":37,"allowed-tools":40,"tags":41},"Requires docker + nvidia-container-toolkit + at least one VLM endpoint (Gemini API key or OpenAI-compatible).",{"author":38,"version":39},"NVIDIA Corporation","0.1.0","Read Bash Write",[42,43,44,45,46,47],"image","grounding","bounding-boxes","auto-label","vlm","2d-grounding",{"type":49,"children":50},"root",[51,60,75,82,87,93,105,134,140,147,152,610,616,629,724,745,757,763,861,867,878,1112,1118,1130,1259,1265,1278,1407,1413,1441],{"type":52,"tag":53,"props":54,"children":56},"element","h1",{"id":55},"image-grounding-pipeline",[57],{"type":58,"value":59},"text","Image Grounding Pipeline",{"type":52,"tag":61,"props":62,"children":63},"p",{},[64,66,73],{"type":58,"value":65},"Turn ",{"type":52,"tag":67,"props":68,"children":70},"code",{"className":69},[],[71],{"type":58,"value":72},"(image, caption)",{"type":58,"value":74}," pairs into per-image grounded annotations: cleaned captions, referring expressions with character spans, and pixel-space bounding boxes for each expression. A single VLM (Gemini or any OpenAI-compatible endpoint) handles both steps.",{"type":52,"tag":76,"props":77,"children":79},"h2",{"id":78},"purpose",[80],{"type":58,"value":81},"Purpose",{"type":52,"tag":61,"props":83,"children":84},{},[85],{"type":58,"value":86},"Generate phrase-grounded training data for referring-expression and grounding models. The VLM acts as a \"teacher\" annotator: Step 0 extracts referring expressions from the caption while looking at the image; Step 1 returns one bbox set per expression for each image.",{"type":52,"tag":76,"props":88,"children":90},{"id":89},"pipeline-architecture",[91],{"type":58,"value":92},"Pipeline Architecture",{"type":52,"tag":94,"props":95,"children":99},"pre",{"className":96,"code":98,"language":58},[97],"language-text","Step 0: Expression extraction  → VLM cleans caption, extracts referring expressions + char spans\nStep 1: Phrase grounding       → VLM returns pixel bboxes + scores per expression\n",[100],{"type":52,"tag":67,"props":101,"children":103},{"__ignoreMap":102},"",[104],{"type":58,"value":98},{"type":52,"tag":61,"props":106,"children":107},{},[108,110,116,118,124,126,132],{"type":58,"value":109},"Steps are individually selectable via ",{"type":52,"tag":67,"props":111,"children":113},{"className":112},[],[114],{"type":58,"value":115},"workflow.steps",{"type":58,"value":117},". Each step writes a per-sample checkpoint to ",{"type":52,"tag":67,"props":119,"children":121},{"className":120},[],[122],{"type":58,"value":123},"step_\u003CN>_*\u002F.ckpt\u002F\u003Csample_id>.json",{"type":58,"value":125}," and skips already-processed records on re-run. Set ",{"type":52,"tag":67,"props":127,"children":129},{"className":128},[],[130],{"type":58,"value":131},"workflow.force_reprocess: true",{"type":58,"value":133}," to ignore checkpoints and reprocess from scratch.",{"type":52,"tag":76,"props":135,"children":137},{"id":136},"instructions",[138],{"type":58,"value":139},"Instructions",{"type":52,"tag":141,"props":142,"children":144},"h3",{"id":143},"initial-setup",[145],{"type":58,"value":146},"Initial setup",{"type":52,"tag":61,"props":148,"children":149},{},[150],{"type":58,"value":151},"When a user wants to run this pipeline, walk through these steps:",{"type":52,"tag":153,"props":154,"children":155},"ol",{},[156,184,209,539,593],{"type":52,"tag":157,"props":158,"children":159},"li",{},[160,166,168,174,176,182],{"type":52,"tag":161,"props":162,"children":163},"strong",{},[164],{"type":58,"value":165},"Input JSONL",{"type":58,"value":167},": Ask for the JSONL path. Each line must be one object like ",{"type":52,"tag":67,"props":169,"children":171},{"className":170},[],[172],{"type":58,"value":173},"{\"image_path\": \"...\", \"caption\": \"...\"}",{"type":58,"value":175},". ",{"type":52,"tag":67,"props":177,"children":179},{"className":178},[],[180],{"type":58,"value":181},"image_path",{"type":58,"value":183}," can be absolute or relative.",{"type":52,"tag":157,"props":185,"children":186},{},[187,192,194,199,201,207],{"type":52,"tag":161,"props":188,"children":189},{},[190],{"type":58,"value":191},"Image root",{"type":58,"value":193},": If any ",{"type":52,"tag":67,"props":195,"children":197},{"className":196},[],[198],{"type":58,"value":181},{"type":58,"value":200}," values are relative, set ",{"type":52,"tag":67,"props":202,"children":204},{"className":203},[],[205],{"type":58,"value":206},"data.image_root",{"type":58,"value":208}," to the directory they should resolve from.",{"type":52,"tag":157,"props":210,"children":211},{},[212,217,219,533,537],{"type":52,"tag":161,"props":213,"children":214},{},[215],{"type":58,"value":216},"API access",{"type":58,"value":218},": Ask the user which VLM endpoint they want to use. Present these five options and act on the choice:",{"type":52,"tag":153,"props":220,"children":221},{},[222,256,306,414,499],{"type":52,"tag":157,"props":223,"children":224},{},[225,230,232,238,240,246,248,254],{"type":52,"tag":161,"props":226,"children":227},{},[228],{"type":58,"value":229},"Gemini",{"type":58,"value":231}," — set ",{"type":52,"tag":67,"props":233,"children":235},{"className":234},[],[236],{"type":58,"value":237},"vlm.backend: \"gemini\"",{"type":58,"value":239},"; require ",{"type":52,"tag":67,"props":241,"children":243},{"className":242},[],[244],{"type":58,"value":245},"GOOGLE_API_KEY",{"type":58,"value":247}," (env var or ",{"type":52,"tag":67,"props":249,"children":251},{"className":250},[],[252],{"type":58,"value":253},"vlm.gemini.api_key",{"type":58,"value":255},").",{"type":52,"tag":157,"props":257,"children":258},{},[259,264,266,272,274,280,282,288,290,296,298,304],{"type":52,"tag":161,"props":260,"children":261},{},[262],{"type":58,"value":263},"NIM",{"type":58,"value":265}," (e.g. ",{"type":52,"tag":67,"props":267,"children":269},{"className":268},[],[270],{"type":58,"value":271},"https:\u002F\u002Finference-api.nvidia.com\u002Fv1",{"type":58,"value":273},") — set ",{"type":52,"tag":67,"props":275,"children":277},{"className":276},[],[278],{"type":58,"value":279},"vlm.backend: \"openai\"",{"type":58,"value":281},"; collect ",{"type":52,"tag":67,"props":283,"children":285},{"className":284},[],[286],{"type":58,"value":287},"base_url",{"type":58,"value":289},", ",{"type":52,"tag":67,"props":291,"children":293},{"className":292},[],[294],{"type":58,"value":295},"model_name",{"type":58,"value":297},", and ",{"type":52,"tag":67,"props":299,"children":301},{"className":300},[],[302],{"type":58,"value":303},"api_key",{"type":58,"value":305},".",{"type":52,"tag":157,"props":307,"children":308},{},[309,314,316],{"type":52,"tag":161,"props":310,"children":311},{},[312],{"type":58,"value":313},"TAO inference microservice",{"type":58,"value":315}," (self-hosted, OpenAI-compatible). Confirm whether the server is already running:\n",{"type":52,"tag":317,"props":318,"children":319},"ul",{},[320,356],{"type":52,"tag":157,"props":321,"children":322},{},[323,328,330,335,336,341,343,348,350,355],{"type":52,"tag":161,"props":324,"children":325},{},[326],{"type":58,"value":327},"Running",{"type":58,"value":329}," — collect ",{"type":52,"tag":67,"props":331,"children":333},{"className":332},[],[334],{"type":58,"value":287},{"type":58,"value":289},{"type":52,"tag":67,"props":337,"children":339},{"className":338},[],[340],{"type":58,"value":295},{"type":58,"value":342},", and (optionally) ",{"type":52,"tag":67,"props":344,"children":346},{"className":345},[],[347],{"type":58,"value":303},{"type":58,"value":349},"; set ",{"type":52,"tag":67,"props":351,"children":353},{"className":352},[],[354],{"type":58,"value":279},{"type":58,"value":305},{"type":52,"tag":157,"props":357,"children":358},{},[359,364,366,372,374,380,382,388,390,395,396,401,402,407,408,413],{"type":52,"tag":161,"props":360,"children":361},{},[362],{"type":58,"value":363},"Not running",{"type":58,"value":365}," — guide the user through the ",{"type":52,"tag":67,"props":367,"children":369},{"className":368},[],[370],{"type":58,"value":371},"skills\u002Fapplications\u002Ftao-run-inference-service",{"type":58,"value":373}," skill, which stands up a local TAO inference microservice with an OpenAI-compatible API. Before promising a specific model, check ",{"type":52,"tag":67,"props":375,"children":377},{"className":376},[],[378],{"type":58,"value":379},"skills\u002Fapplications\u002Ftao-run-inference-service\u002Freferences\u002Fservice.yaml",{"type":58,"value":381}," for ",{"type":52,"tag":67,"props":383,"children":385},{"className":384},[],[386],{"type":58,"value":387},"valid_network_arch_config_basenames",{"type":58,"value":389},". Once the server is up, collect ",{"type":52,"tag":67,"props":391,"children":393},{"className":392},[],[394],{"type":58,"value":287},{"type":58,"value":289},{"type":52,"tag":67,"props":397,"children":399},{"className":398},[],[400],{"type":58,"value":295},{"type":58,"value":342},{"type":52,"tag":67,"props":403,"children":405},{"className":404},[],[406],{"type":58,"value":303},{"type":58,"value":349},{"type":52,"tag":67,"props":409,"children":411},{"className":410},[],[412],{"type":58,"value":279},{"type":58,"value":305},{"type":52,"tag":157,"props":415,"children":416},{},[417,422,423],{"type":52,"tag":161,"props":418,"children":419},{},[420],{"type":58,"value":421},"vLLM",{"type":58,"value":315},{"type":52,"tag":317,"props":424,"children":425},{},[426,458],{"type":52,"tag":157,"props":427,"children":428},{},[429,433,434,439,440,445,446,451,452,457],{"type":52,"tag":161,"props":430,"children":431},{},[432],{"type":58,"value":327},{"type":58,"value":329},{"type":52,"tag":67,"props":435,"children":437},{"className":436},[],[438],{"type":58,"value":287},{"type":58,"value":289},{"type":52,"tag":67,"props":441,"children":443},{"className":442},[],[444],{"type":58,"value":295},{"type":58,"value":342},{"type":52,"tag":67,"props":447,"children":449},{"className":448},[],[450],{"type":58,"value":303},{"type":58,"value":349},{"type":52,"tag":67,"props":453,"children":455},{"className":454},[],[456],{"type":58,"value":279},{"type":58,"value":305},{"type":52,"tag":157,"props":459,"children":460},{},[461,465,467,473,475,480,481,486,487,492,493,498],{"type":52,"tag":161,"props":462,"children":463},{},[464],{"type":58,"value":363},{"type":58,"value":466}," — follow ",{"type":52,"tag":468,"props":469,"children":471},"a",{"href":470},"references\u002Fvllm_server.md",[472],{"type":58,"value":470},{"type":58,"value":474}," to install and launch a vLLM server, then collect ",{"type":52,"tag":67,"props":476,"children":478},{"className":477},[],[479],{"type":58,"value":287},{"type":58,"value":289},{"type":52,"tag":67,"props":482,"children":484},{"className":483},[],[485],{"type":58,"value":295},{"type":58,"value":342},{"type":52,"tag":67,"props":488,"children":490},{"className":489},[],[491],{"type":58,"value":303},{"type":58,"value":349},{"type":52,"tag":67,"props":494,"children":496},{"className":495},[],[497],{"type":58,"value":279},{"type":58,"value":305},{"type":52,"tag":157,"props":500,"children":501},{},[502,507,509,514,515,520,521,526,527,532],{"type":52,"tag":161,"props":503,"children":504},{},[505],{"type":58,"value":506},"Custom",{"type":58,"value":508}," (any other OpenAI-compatible endpoint) — set ",{"type":52,"tag":67,"props":510,"children":512},{"className":511},[],[513],{"type":58,"value":279},{"type":58,"value":281},{"type":52,"tag":67,"props":516,"children":518},{"className":517},[],[519],{"type":58,"value":287},{"type":58,"value":289},{"type":52,"tag":67,"props":522,"children":524},{"className":523},[],[525],{"type":58,"value":295},{"type":58,"value":342},{"type":52,"tag":67,"props":528,"children":530},{"className":529},[],[531],{"type":58,"value":303},{"type":58,"value":305},{"type":52,"tag":534,"props":535,"children":536},"br",{},[],{"type":58,"value":538},"If the user has no endpoint and does not want to set one up, stop and help resolve API access first.",{"type":52,"tag":157,"props":540,"children":541},{},[542,547,549],{"type":52,"tag":161,"props":543,"children":544},{},[545],{"type":58,"value":546},"Workflow steps",{"type":58,"value":548},": Choose one of:",{"type":52,"tag":317,"props":550,"children":551},{},[552,563,574],{"type":52,"tag":157,"props":553,"children":554},{},[555,557],{"type":58,"value":556},"Full pipeline: ",{"type":52,"tag":67,"props":558,"children":560},{"className":559},[],[561],{"type":58,"value":562},"[\"0\", \"1\"]",{"type":52,"tag":157,"props":564,"children":565},{},[566,568],{"type":58,"value":567},"Expression extraction only: ",{"type":52,"tag":67,"props":569,"children":571},{"className":570},[],[572],{"type":58,"value":573},"[\"0\"]",{"type":52,"tag":157,"props":575,"children":576},{},[577,579,585,587],{"type":58,"value":578},"Grounding only: ",{"type":52,"tag":67,"props":580,"children":582},{"className":581},[],[583],{"type":58,"value":584},"[\"1\"]",{"type":58,"value":586},", which requires existing step-0 output at ",{"type":52,"tag":67,"props":588,"children":590},{"className":589},[],[591],{"type":58,"value":592},"results_dir\u002Fstep_0_expression_extraction\u002Fannotations.jsonl",{"type":52,"tag":157,"props":594,"children":595},{},[596,601,603,609],{"type":52,"tag":161,"props":597,"children":598},{},[599],{"type":58,"value":600},"Resume vs fresh run",{"type":58,"value":602},": By default, the workflow reuses checkpoints and skips completed records. To reprocess everything, set ",{"type":52,"tag":67,"props":604,"children":606},{"className":605},[],[607],{"type":58,"value":608},"image_grounding.workflow.force_reprocess=true",{"type":58,"value":305},{"type":52,"tag":141,"props":611,"children":613},{"id":612},"running-the-pipeline",[614],{"type":58,"value":615},"Running the pipeline",{"type":52,"tag":61,"props":617,"children":618},{},[619,621,627],{"type":58,"value":620},"The pipeline runs inside the TAO Toolkit container via the ",{"type":52,"tag":67,"props":622,"children":624},{"className":623},[],[625],{"type":58,"value":626},"auto_label",{"type":58,"value":628}," CLI:",{"type":52,"tag":94,"props":630,"children":634},{"className":631,"code":632,"language":633,"meta":102,"style":102},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","auto_label generate -e \u002Fpath\u002Fto\u002Fspec.yaml \\\n    results_dir=\u002Fresults \\\n    image_grounding.data.input_jsonl=\u002Fdata\u002Fcaptions.jsonl \\\n    image_grounding.data.image_root=\u002Fdata\u002Fimages \\\n    image_grounding.vlm.gemini.api_key=$GOOGLE_API_KEY\n","bash",[635],{"type":52,"tag":67,"props":636,"children":637},{"__ignoreMap":102},[638,671,684,697,710],{"type":52,"tag":639,"props":640,"children":643},"span",{"class":641,"line":642},"line",1,[644,649,655,660,665],{"type":52,"tag":639,"props":645,"children":647},{"style":646},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[648],{"type":58,"value":626},{"type":52,"tag":639,"props":650,"children":652},{"style":651},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[653],{"type":58,"value":654}," generate",{"type":52,"tag":639,"props":656,"children":657},{"style":651},[658],{"type":58,"value":659}," -e",{"type":52,"tag":639,"props":661,"children":662},{"style":651},[663],{"type":58,"value":664}," \u002Fpath\u002Fto\u002Fspec.yaml",{"type":52,"tag":639,"props":666,"children":668},{"style":667},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[669],{"type":58,"value":670}," \\\n",{"type":52,"tag":639,"props":672,"children":674},{"class":641,"line":673},2,[675,680],{"type":52,"tag":639,"props":676,"children":677},{"style":651},[678],{"type":58,"value":679},"    results_dir=\u002Fresults",{"type":52,"tag":639,"props":681,"children":682},{"style":667},[683],{"type":58,"value":670},{"type":52,"tag":639,"props":685,"children":687},{"class":641,"line":686},3,[688,693],{"type":52,"tag":639,"props":689,"children":690},{"style":651},[691],{"type":58,"value":692},"    image_grounding.data.input_jsonl=\u002Fdata\u002Fcaptions.jsonl",{"type":52,"tag":639,"props":694,"children":695},{"style":667},[696],{"type":58,"value":670},{"type":52,"tag":639,"props":698,"children":700},{"class":641,"line":699},4,[701,706],{"type":52,"tag":639,"props":702,"children":703},{"style":651},[704],{"type":58,"value":705},"    image_grounding.data.image_root=\u002Fdata\u002Fimages",{"type":52,"tag":639,"props":707,"children":708},{"style":667},[709],{"type":58,"value":670},{"type":52,"tag":639,"props":711,"children":713},{"class":641,"line":712},5,[714,719],{"type":52,"tag":639,"props":715,"children":716},{"style":651},[717],{"type":58,"value":718},"    image_grounding.vlm.gemini.api_key=",{"type":52,"tag":639,"props":720,"children":721},{"style":667},[722],{"type":58,"value":723},"$GOOGLE_API_KEY\n",{"type":52,"tag":61,"props":725,"children":726},{},[727,729,735,737,743],{"type":58,"value":728},"Generate a default spec: ",{"type":52,"tag":67,"props":730,"children":732},{"className":731},[],[733],{"type":58,"value":734},"auto_label default_specs results_dir=\u002Fresults module_name=auto_label",{"type":58,"value":736},", then set ",{"type":52,"tag":67,"props":738,"children":740},{"className":739},[],[741],{"type":58,"value":742},"autolabel_type: \"image_grounding\"",{"type":58,"value":744},". All fields support Hydra dot-notation overrides on the command line.",{"type":52,"tag":61,"props":746,"children":747},{},[748,750,755],{"type":58,"value":749},"See ",{"type":52,"tag":468,"props":751,"children":753},{"href":752},"references\u002Fconfiguration.md",[754],{"type":58,"value":752},{"type":58,"value":756}," for the full YAML structure, all parameters, model\u002Fendpoint setup, and error patterns.",{"type":52,"tag":141,"props":758,"children":760},{"id":759},"recommended-pilot-workflow",[761],{"type":58,"value":762},"Recommended pilot workflow",{"type":52,"tag":153,"props":764,"children":765},{},[766,771,800,820,856],{"type":52,"tag":157,"props":767,"children":768},{},[769],{"type":58,"value":770},"Run on 5-10 images with both steps",{"type":52,"tag":157,"props":772,"children":773},{},[774,776,782,784,790,792,798],{"type":58,"value":775},"Inspect ",{"type":52,"tag":67,"props":777,"children":779},{"className":778},[],[780],{"type":58,"value":781},"step_0_expression_extraction\u002Fannotations.jsonl",{"type":58,"value":783}," — are ",{"type":52,"tag":67,"props":785,"children":787},{"className":786},[],[788],{"type":58,"value":789},"cleaned_caption",{"type":58,"value":791}," and ",{"type":52,"tag":67,"props":793,"children":795},{"className":794},[],[796],{"type":58,"value":797},"expressions[]",{"type":58,"value":799}," accurate? Are the right noun phrases captured?",{"type":52,"tag":157,"props":801,"children":802},{},[803,804,810,812,818],{"type":58,"value":775},{"type":52,"tag":67,"props":805,"children":807},{"className":806},[],[808],{"type":58,"value":809},"step_1_grounding\u002Fannotations.jsonl",{"type":58,"value":811}," — do the bboxes in ",{"type":52,"tag":67,"props":813,"children":815},{"className":814},[],[816],{"type":58,"value":817},"expressions[].instances[]",{"type":58,"value":819}," look right? Are confidence scores reasonable?",{"type":52,"tag":157,"props":821,"children":822},{},[823,825,831,833,839,841,847,849,855],{"type":58,"value":824},"If quality is insufficient, switch the VLM to a stronger model (e.g. ",{"type":52,"tag":67,"props":826,"children":828},{"className":827},[],[829],{"type":58,"value":830},"gemini-2.5-pro",{"type":58,"value":832},") or raise ",{"type":52,"tag":67,"props":834,"children":836},{"className":835},[],[837],{"type":58,"value":838},"media_resolution",{"type":58,"value":840},"\u002F",{"type":52,"tag":67,"props":842,"children":844},{"className":843},[],[845],{"type":58,"value":846},"max_output_tokens",{"type":58,"value":848},", then re-run with ",{"type":52,"tag":67,"props":850,"children":852},{"className":851},[],[853],{"type":58,"value":854},"force_reprocess=true",{"type":58,"value":305},{"type":52,"tag":157,"props":857,"children":858},{},[859],{"type":58,"value":860},"Scale to the full dataset once satisfied.",{"type":52,"tag":76,"props":862,"children":864},{"id":863},"configuration",[865],{"type":58,"value":866},"Configuration",{"type":52,"tag":61,"props":868,"children":869},{},[870,872,876],{"type":58,"value":871},"Key configuration fields (full reference in ",{"type":52,"tag":468,"props":873,"children":874},{"href":752},[875],{"type":58,"value":752},{"type":58,"value":877},"):",{"type":52,"tag":879,"props":880,"children":881},"table",{},[882,906],{"type":52,"tag":883,"props":884,"children":885},"thead",{},[886],{"type":52,"tag":887,"props":888,"children":889},"tr",{},[890,896,901],{"type":52,"tag":891,"props":892,"children":893},"th",{},[894],{"type":58,"value":895},"Field",{"type":52,"tag":891,"props":897,"children":898},{},[899],{"type":58,"value":900},"Default",{"type":52,"tag":891,"props":902,"children":903},{},[904],{"type":58,"value":905},"Description",{"type":52,"tag":907,"props":908,"children":909},"tbody",{},[910,952,978,1004,1043,1080],{"type":52,"tag":887,"props":911,"children":912},{},[913,922,931],{"type":52,"tag":914,"props":915,"children":916},"td",{},[917],{"type":52,"tag":67,"props":918,"children":920},{"className":919},[],[921],{"type":58,"value":115},{"type":52,"tag":914,"props":923,"children":924},{},[925],{"type":52,"tag":67,"props":926,"children":928},{"className":927},[],[929],{"type":58,"value":930},"[\"0\",\"1\"]",{"type":52,"tag":914,"props":932,"children":933},{},[934,936,942,944,950],{"type":58,"value":935},"Which pipeline steps to execute (",{"type":52,"tag":67,"props":937,"children":939},{"className":938},[],[940],{"type":58,"value":941},"\"0\"",{"type":58,"value":943}," = expressions, ",{"type":52,"tag":67,"props":945,"children":947},{"className":946},[],[948],{"type":58,"value":949},"\"1\"",{"type":58,"value":951}," = grounding)",{"type":52,"tag":887,"props":953,"children":954},{},[955,964,973],{"type":52,"tag":914,"props":956,"children":957},{},[958],{"type":52,"tag":67,"props":959,"children":961},{"className":960},[],[962],{"type":58,"value":963},"workflow.max_workers",{"type":52,"tag":914,"props":965,"children":966},{},[967],{"type":52,"tag":67,"props":968,"children":970},{"className":969},[],[971],{"type":58,"value":972},"4",{"type":52,"tag":914,"props":974,"children":975},{},[976],{"type":58,"value":977},"Parallel threads per step (watch API rate limits)",{"type":52,"tag":887,"props":979,"children":980},{},[981,990,999],{"type":52,"tag":914,"props":982,"children":983},{},[984],{"type":52,"tag":67,"props":985,"children":987},{"className":986},[],[988],{"type":58,"value":989},"workflow.force_reprocess",{"type":52,"tag":914,"props":991,"children":992},{},[993],{"type":52,"tag":67,"props":994,"children":996},{"className":995},[],[997],{"type":58,"value":998},"false",{"type":52,"tag":914,"props":1000,"children":1001},{},[1002],{"type":58,"value":1003},"Ignore per-sample checkpoints and reprocess from scratch",{"type":52,"tag":887,"props":1005,"children":1006},{},[1007,1016,1025],{"type":52,"tag":914,"props":1008,"children":1009},{},[1010],{"type":52,"tag":67,"props":1011,"children":1013},{"className":1012},[],[1014],{"type":58,"value":1015},"vlm.backend",{"type":52,"tag":914,"props":1017,"children":1018},{},[1019],{"type":52,"tag":67,"props":1020,"children":1022},{"className":1021},[],[1023],{"type":58,"value":1024},"\"gemini\"",{"type":52,"tag":914,"props":1026,"children":1027},{},[1028,1033,1035,1041],{"type":52,"tag":67,"props":1029,"children":1031},{"className":1030},[],[1032],{"type":58,"value":1024},{"type":58,"value":1034}," or ",{"type":52,"tag":67,"props":1036,"children":1038},{"className":1037},[],[1039],{"type":58,"value":1040},"\"openai\"",{"type":58,"value":1042}," (OpenAI-compatible endpoint)",{"type":52,"tag":887,"props":1044,"children":1045},{},[1046,1055,1060],{"type":52,"tag":914,"props":1047,"children":1048},{},[1049],{"type":52,"tag":67,"props":1050,"children":1052},{"className":1051},[],[1053],{"type":58,"value":1054},"data.input_jsonl",{"type":52,"tag":914,"props":1056,"children":1057},{},[1058],{"type":58,"value":1059},"required",{"type":52,"tag":914,"props":1061,"children":1062},{},[1063,1065,1070,1072,1078],{"type":58,"value":1064},"Path to input JSONL with ",{"type":52,"tag":67,"props":1066,"children":1068},{"className":1067},[],[1069],{"type":58,"value":181},{"type":58,"value":1071}," + ",{"type":52,"tag":67,"props":1073,"children":1075},{"className":1074},[],[1076],{"type":58,"value":1077},"caption",{"type":58,"value":1079}," per line",{"type":52,"tag":887,"props":1081,"children":1082},{},[1083,1091,1100],{"type":52,"tag":914,"props":1084,"children":1085},{},[1086],{"type":52,"tag":67,"props":1087,"children":1089},{"className":1088},[],[1090],{"type":58,"value":206},{"type":52,"tag":914,"props":1092,"children":1093},{},[1094],{"type":52,"tag":67,"props":1095,"children":1097},{"className":1096},[],[1098],{"type":58,"value":1099},"\"\"",{"type":52,"tag":914,"props":1101,"children":1102},{},[1103,1105,1110],{"type":58,"value":1104},"Optional prefix for resolving relative ",{"type":52,"tag":67,"props":1106,"children":1108},{"className":1107},[],[1109],{"type":58,"value":181},{"type":58,"value":1111}," entries",{"type":52,"tag":76,"props":1113,"children":1115},{"id":1114},"inputs",[1116],{"type":58,"value":1117},"Inputs",{"type":52,"tag":61,"props":1119,"children":1120},{},[1121,1123,1128],{"type":58,"value":1122},"A single JSONL file at ",{"type":52,"tag":67,"props":1124,"children":1126},{"className":1125},[],[1127],{"type":58,"value":1054},{"type":58,"value":1129},". One JSON object per line:",{"type":52,"tag":879,"props":1131,"children":1132},{},[1133,1152],{"type":52,"tag":883,"props":1134,"children":1135},{},[1136],{"type":52,"tag":887,"props":1137,"children":1138},{},[1139,1143,1148],{"type":52,"tag":891,"props":1140,"children":1141},{},[1142],{"type":58,"value":895},{"type":52,"tag":891,"props":1144,"children":1145},{},[1146],{"type":58,"value":1147},"Required",{"type":52,"tag":891,"props":1149,"children":1150},{},[1151],{"type":58,"value":905},{"type":52,"tag":907,"props":1153,"children":1154},{},[1155,1181,1201,1223],{"type":52,"tag":887,"props":1156,"children":1157},{},[1158,1166,1171],{"type":52,"tag":914,"props":1159,"children":1160},{},[1161],{"type":52,"tag":67,"props":1162,"children":1164},{"className":1163},[],[1165],{"type":58,"value":181},{"type":52,"tag":914,"props":1167,"children":1168},{},[1169],{"type":58,"value":1170},"yes",{"type":52,"tag":914,"props":1172,"children":1173},{},[1174,1176],{"type":58,"value":1175},"Absolute path, or relative path resolved against ",{"type":52,"tag":67,"props":1177,"children":1179},{"className":1178},[],[1180],{"type":58,"value":206},{"type":52,"tag":887,"props":1182,"children":1183},{},[1184,1192,1196],{"type":52,"tag":914,"props":1185,"children":1186},{},[1187],{"type":52,"tag":67,"props":1188,"children":1190},{"className":1189},[],[1191],{"type":58,"value":1077},{"type":52,"tag":914,"props":1193,"children":1194},{},[1195],{"type":58,"value":1170},{"type":52,"tag":914,"props":1197,"children":1198},{},[1199],{"type":58,"value":1200},"Free-text caption for the image",{"type":52,"tag":887,"props":1202,"children":1203},{},[1204,1213,1218],{"type":52,"tag":914,"props":1205,"children":1206},{},[1207],{"type":52,"tag":67,"props":1208,"children":1210},{"className":1209},[],[1211],{"type":58,"value":1212},"image_id",{"type":52,"tag":914,"props":1214,"children":1215},{},[1216],{"type":58,"value":1217},"no",{"type":52,"tag":914,"props":1219,"children":1220},{},[1221],{"type":58,"value":1222},"Stable identifier; auto-derived from the filename if missing",{"type":52,"tag":887,"props":1224,"children":1225},{},[1226,1242,1246],{"type":52,"tag":914,"props":1227,"children":1228},{},[1229,1235,1236],{"type":52,"tag":67,"props":1230,"children":1232},{"className":1231},[],[1233],{"type":58,"value":1234},"width",{"type":58,"value":289},{"type":52,"tag":67,"props":1237,"children":1239},{"className":1238},[],[1240],{"type":58,"value":1241},"height",{"type":52,"tag":914,"props":1243,"children":1244},{},[1245],{"type":58,"value":1217},{"type":52,"tag":914,"props":1247,"children":1248},{},[1249,1251,1257],{"type":58,"value":1250},"Image dimensions in pixels; default to ",{"type":52,"tag":67,"props":1252,"children":1254},{"className":1253},[],[1255],{"type":58,"value":1256},"1920×1080",{"type":58,"value":1258}," for bbox clamping if missing",{"type":52,"tag":76,"props":1260,"children":1262},{"id":1261},"outputs",[1263],{"type":58,"value":1264},"Outputs",{"type":52,"tag":61,"props":1266,"children":1267},{},[1268,1270,1276],{"type":58,"value":1269},"All outputs go to ",{"type":52,"tag":67,"props":1271,"children":1273},{"className":1272},[],[1274],{"type":58,"value":1275},"results_dir\u002F",{"type":58,"value":1277},":",{"type":52,"tag":317,"props":1279,"children":1280},{},[1281,1339,1386,1397],{"type":52,"tag":157,"props":1282,"children":1283},{},[1284,1289,1291,1296,1297,1302,1304,1309,1310,1316,1317,1323,1324,1330,1332,1338],{"type":52,"tag":67,"props":1285,"children":1287},{"className":1286},[],[1288],{"type":58,"value":781},{"type":58,"value":1290}," — per-record output enriched with ",{"type":52,"tag":67,"props":1292,"children":1294},{"className":1293},[],[1295],{"type":58,"value":789},{"type":58,"value":791},{"type":52,"tag":67,"props":1298,"children":1300},{"className":1299},[],[1301],{"type":58,"value":797},{"type":58,"value":1303}," (each with ",{"type":52,"tag":67,"props":1305,"children":1307},{"className":1306},[],[1308],{"type":58,"value":58},{"type":58,"value":289},{"type":52,"tag":67,"props":1311,"children":1313},{"className":1312},[],[1314],{"type":58,"value":1315},"expression_id",{"type":58,"value":289},{"type":52,"tag":67,"props":1318,"children":1320},{"className":1319},[],[1321],{"type":58,"value":1322},"char_span",{"type":58,"value":289},{"type":52,"tag":67,"props":1325,"children":1327},{"className":1326},[],[1328],{"type":58,"value":1329},"noun_chunk",{"type":58,"value":1331},", empty ",{"type":52,"tag":67,"props":1333,"children":1335},{"className":1334},[],[1336],{"type":58,"value":1337},"instances[]",{"type":58,"value":255},{"type":52,"tag":157,"props":1340,"children":1341},{},[1342,1347,1349,1354,1356,1362,1364,1370,1372,1378,1379,1385],{"type":52,"tag":67,"props":1343,"children":1345},{"className":1344},[],[1346],{"type":58,"value":809},{"type":58,"value":1348}," — same records with ",{"type":52,"tag":67,"props":1350,"children":1352},{"className":1351},[],[1353],{"type":58,"value":817},{"type":58,"value":1355}," filled in (each instance has ",{"type":52,"tag":67,"props":1357,"children":1359},{"className":1358},[],[1360],{"type":58,"value":1361},"bbox: [x1,y1,x2,y2]",{"type":58,"value":1363}," in pixel space, ",{"type":52,"tag":67,"props":1365,"children":1367},{"className":1366},[],[1368],{"type":58,"value":1369},"score",{"type":58,"value":1371}," in ",{"type":52,"tag":67,"props":1373,"children":1375},{"className":1374},[],[1376],{"type":58,"value":1377},"[0.0, 1.0]",{"type":58,"value":297},{"type":52,"tag":67,"props":1380,"children":1382},{"className":1381},[],[1383],{"type":58,"value":1384},"bbox_id",{"type":58,"value":255},{"type":52,"tag":157,"props":1387,"children":1388},{},[1389,1395],{"type":52,"tag":67,"props":1390,"children":1392},{"className":1391},[],[1393],{"type":58,"value":1394},"results_dir\u002Fannotations.jsonl",{"type":58,"value":1396}," — copy of the last step's output for convenience.",{"type":52,"tag":157,"props":1398,"children":1399},{},[1400,1405],{"type":52,"tag":67,"props":1401,"children":1403},{"className":1402},[],[1404],{"type":58,"value":123},{"type":58,"value":1406}," — per-sample checkpoints used for resume.",{"type":52,"tag":76,"props":1408,"children":1410},{"id":1409},"prerequisites",[1411],{"type":58,"value":1412},"Prerequisites",{"type":52,"tag":317,"props":1414,"children":1415},{},[1416,1432],{"type":52,"tag":157,"props":1417,"children":1418},{},[1419,1424,1426],{"type":52,"tag":161,"props":1420,"children":1421},{},[1422],{"type":58,"value":1423},"Container",{"type":58,"value":1425},": ",{"type":52,"tag":67,"props":1427,"children":1429},{"className":1428},[],[1430],{"type":58,"value":1431},"nvcr.io\u002Fnvidia\u002Ftao\u002Ftao-toolkit:6.26.3-pyt",{"type":52,"tag":157,"props":1433,"children":1434},{},[1435,1439],{"type":52,"tag":161,"props":1436,"children":1437},{},[1438],{"type":58,"value":216},{"type":58,"value":1440},": At least one VLM endpoint (Gemini API key or OpenAI-compatible endpoint capable of image input)",{"type":52,"tag":1442,"props":1443,"children":1444},"style",{},[1445],{"type":58,"value":1446},"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":1448,"total":1552},[1449,1466,1480,1494,1506,1523,1538],{"slug":1450,"name":1450,"fn":1451,"description":1452,"org":1453,"tags":1454,"stars":23,"repoUrl":24,"updatedAt":1465},"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},[1455,1458,1461,1462],{"name":1456,"slug":1457,"type":15},"Data Analysis","data-analysis",{"name":1459,"slug":1460,"type":15},"Data Engineering","data-engineering",{"name":9,"slug":8,"type":15},{"name":1463,"slug":1464,"type":15},"Performance","performance","2026-07-14T05:28:43.176466",{"slug":1467,"name":1467,"fn":1468,"description":1469,"org":1470,"tags":1471,"stars":23,"repoUrl":24,"updatedAt":1479},"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},[1472,1475,1478],{"name":1473,"slug":1474,"type":15},"Deployment","deployment",{"name":1476,"slug":1477,"type":15},"Infrastructure","infrastructure",{"name":9,"slug":8,"type":15},"2026-07-14T05:29:06.667109",{"slug":1481,"name":1481,"fn":1482,"description":1483,"org":1484,"tags":1485,"stars":23,"repoUrl":24,"updatedAt":1493},"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},[1486,1489,1490],{"name":1487,"slug":1488,"type":15},"Agents","agents",{"name":9,"slug":8,"type":15},{"name":1491,"slug":1492,"type":15},"Research","research","2026-07-14T05:28:06.816956",{"slug":1495,"name":1495,"fn":1496,"description":1497,"org":1498,"tags":1499,"stars":23,"repoUrl":24,"updatedAt":1505},"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},[1500,1501,1502],{"name":1456,"slug":1457,"type":15},{"name":9,"slug":8,"type":15},{"name":1503,"slug":1504,"type":15},"Testing","testing","2026-07-17T05:29:03.913266",{"slug":1507,"name":1507,"fn":1508,"description":1509,"org":1510,"tags":1511,"stars":23,"repoUrl":24,"updatedAt":1522},"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},[1512,1515,1518,1519],{"name":1513,"slug":1514,"type":15},"Automation","automation",{"name":1516,"slug":1517,"type":15},"Imaging","imaging",{"name":9,"slug":8,"type":15},{"name":1520,"slug":1521,"type":15},"Video","video","2026-07-17T05:28:53.905004",{"slug":1524,"name":1524,"fn":1525,"description":1526,"org":1527,"tags":1528,"stars":23,"repoUrl":24,"updatedAt":1537},"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},[1529,1530,1533,1534],{"name":1473,"slug":1474,"type":15},{"name":1531,"slug":1532,"type":15},"Docker","docker",{"name":9,"slug":8,"type":15},{"name":1535,"slug":1536,"type":15},"Operations","operations","2026-07-17T05:28:56.913999",{"slug":1539,"name":1539,"fn":1540,"description":1541,"org":1542,"tags":1543,"stars":23,"repoUrl":24,"updatedAt":1551},"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},[1544,1545,1548],{"name":9,"slug":8,"type":15},{"name":1546,"slug":1547,"type":15},"Quantum Computing","quantum-computing",{"name":1549,"slug":1550,"type":15},"Simulation","simulation","2026-07-14T05:26:58.898253",305,{"items":1554,"total":1705},[1555,1573,1589,1600,1612,1626,1639,1653,1664,1673,1687,1696],{"slug":1556,"name":1556,"fn":1557,"description":1558,"org":1559,"tags":1560,"stars":1570,"repoUrl":1571,"updatedAt":1572},"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},[1561,1564,1567],{"name":1562,"slug":1563,"type":15},"Documentation","documentation",{"name":1565,"slug":1566,"type":15},"MCP","mcp",{"name":1568,"slug":1569,"type":15},"Search","search",21777,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw","2026-07-20T06:00:01.461044",{"slug":1574,"name":1574,"fn":1575,"description":1576,"org":1577,"tags":1578,"stars":1586,"repoUrl":1587,"updatedAt":1588},"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},[1579,1582,1583],{"name":1580,"slug":1581,"type":15},"Containers","containers",{"name":1473,"slug":1474,"type":15},{"name":1584,"slug":1585,"type":15},"Python","python",17049,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FMegatron-LM","2026-07-27T06:06:11.249662",{"slug":1590,"name":1590,"fn":1591,"description":1592,"org":1593,"tags":1594,"stars":1586,"repoUrl":1587,"updatedAt":1599},"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},[1595,1598],{"name":1596,"slug":1597,"type":15},"CI\u002FCD","ci-cd",{"name":1473,"slug":1474,"type":15},"2026-07-14T05:25:59.97109",{"slug":1601,"name":1601,"fn":1602,"description":1603,"org":1604,"tags":1605,"stars":1586,"repoUrl":1587,"updatedAt":1611},"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},[1606,1607,1608],{"name":1596,"slug":1597,"type":15},{"name":1473,"slug":1474,"type":15},{"name":1609,"slug":1610,"type":15},"GitHub","github","2026-07-27T06:06:12.278222",{"slug":1613,"name":1613,"fn":1614,"description":1615,"org":1616,"tags":1617,"stars":1586,"repoUrl":1587,"updatedAt":1625},"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},[1618,1621,1622],{"name":1619,"slug":1620,"type":15},"Debugging","debugging",{"name":1609,"slug":1610,"type":15},{"name":1623,"slug":1624,"type":15},"Triage","triage","2026-07-14T05:25:57.442089",{"slug":1627,"name":1627,"fn":1628,"description":1629,"org":1630,"tags":1631,"stars":1586,"repoUrl":1587,"updatedAt":1638},"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},[1632,1635],{"name":1633,"slug":1634,"type":15},"Best Practices","best-practices",{"name":1636,"slug":1637,"type":15},"Code Analysis","code-analysis","2026-07-14T05:25:56.18433",{"slug":1640,"name":1640,"fn":1641,"description":1642,"org":1643,"tags":1644,"stars":1586,"repoUrl":1587,"updatedAt":1652},"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},[1645,1648,1651],{"name":1646,"slug":1647,"type":15},"Machine Learning","machine-learning",{"name":1649,"slug":1650,"type":15},"Migration","migration",{"name":9,"slug":8,"type":15},"2026-07-17T06:07:11.777011",{"slug":1654,"name":1654,"fn":1655,"description":1656,"org":1657,"tags":1658,"stars":1586,"repoUrl":1587,"updatedAt":1663},"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},[1659,1662],{"name":1660,"slug":1661,"type":15},"QA","qa",{"name":1503,"slug":1504,"type":15},"2026-07-14T05:25:53.673039",{"slug":1665,"name":1665,"fn":1666,"description":1667,"org":1668,"tags":1669,"stars":1586,"repoUrl":1587,"updatedAt":1672},"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},[1670,1671],{"name":1473,"slug":1474,"type":15},{"name":1476,"slug":1477,"type":15},"2026-07-14T05:25:49.362534",{"slug":1674,"name":1674,"fn":1675,"description":1676,"org":1677,"tags":1678,"stars":1586,"repoUrl":1587,"updatedAt":1686},"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},[1679,1682,1683],{"name":1680,"slug":1681,"type":15},"Code Review","code-review",{"name":1609,"slug":1610,"type":15},{"name":1684,"slug":1685,"type":15},"Pull Requests","pull-requests","2026-07-14T05:26:01.226578",{"slug":1688,"name":1688,"fn":1689,"description":1690,"org":1691,"tags":1692,"stars":1586,"repoUrl":1587,"updatedAt":1695},"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},[1693,1694],{"name":1660,"slug":1661,"type":15},{"name":1503,"slug":1504,"type":15},"2026-07-14T05:25:54.928983",{"slug":1697,"name":1697,"fn":1698,"description":1699,"org":1700,"tags":1701,"stars":1586,"repoUrl":1587,"updatedAt":1704},"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},[1702,1703],{"name":1513,"slug":1514,"type":15},{"name":1596,"slug":1597,"type":15},"2026-07-30T05:29:03.275638",496]