[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7B5XD9M965H2TAV4ZJMPF6N":3,"skill-same:01M7B5XD9M965H2TAV4ZJMPF6N":93},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":30,"ownerHtmlUrl":31,"ownerBlogUrl":32,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M7B5XD9M965H2TAV4ZJMPF6N","modelopt-speculative-decoding",null,"Train, debug, and validate a speculative decoding draft model (EAGLE3, DFlash, DSpark, Domino) through the ModelOpt launcher pipeline. Use when the user wants to add a new model to a draft-training pipeline, asks why a pipeline run failed, wants experiment logs reviewed, or wants to check whether a run's acceptance rate meets threshold. Triggers on \"EAGLE3\", \"DFlash\", \"DSpark\", \"Domino\", \"draft model\", \"acceptance rate\", \"speculative decoding pipeline\". Do NOT use for quantizing a model (use modelopt-ptq) or serving a checkpoint (use modelopt-deployment).","NVIDIA\u002FModel-Optimizer","NVIDIA","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F1728152?v=4","01M6BXXZYTRNBXSTRXNNBB356H","NVIDIA Corporation","01M6BXY02X3PRVHNNATDHA2D0W",5604,"Apache-2.0","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: modelopt-speculative-decoding\ndescription: >-\n  Train, debug, and validate a speculative decoding draft model (EAGLE3, DFlash,\n  DSpark, Domino) through the ModelOpt launcher pipeline. Use when the user wants\n  to add a new model to a draft-training pipeline, asks why a pipeline run failed,\n  wants experiment logs reviewed, or wants to check whether a run's acceptance rate\n  meets threshold. Triggers on \"EAGLE3\", \"DFlash\", \"DSpark\", \"Domino\", \"draft model\",\n  \"acceptance rate\", \"speculative decoding pipeline\". Do NOT use for quantizing a\n  model (use modelopt-ptq) or serving a checkpoint (use modelopt-deployment).\nuser_invocable: true\n---\n\n# Speculative Decoding Draft-Model Training\n\nEverything needed to take a target model from \"no draft head\" to \"validated\nacceptance rate\" lives in this directory. Work through the stages below in order\nfor a new model; jump straight to a stage when you already know which one you need.\n\n## Two axes: stage and algorithm\n\nThe pipeline is the same shape for every draft-model algorithm — synthesize data,\ndump base-model hidden states, train the draft, benchmark acceptance rate. What\nchanges between algorithms is which training script and recipe run, which knobs\nmatter, and which failures are typical.\n\nSo this skill is split along those two axes, and **you almost always read one file\nfrom each**:\n\n| Axis | Directory | What it holds |\n| --- | --- | --- |\n| Stage | `references\u002Fstages\u002F` | The procedure — algorithm-independent |\n| Algorithm | `references\u002Falgorithms\u002F` | The data sheet — scripts, recipe, knobs, thresholds, known failures |\n\nRead the stage file for *what to do*, and the algorithm sheet for *the values to\nplug in*. When a stage file says \"see the algorithm sheet\", it means the section of\n`references\u002Falgorithms\u002F\u003Calgorithm>.md` with the matching heading.\n\n## Where the code lives\n\nThe launcher scripts in `tools\u002Flauncher\u002Fcommon\u002F{eagle3,specdec}\u002F` are thin wrappers around\n`examples\u002Fspeculative_decoding\u002F` — `main.py` (training), `collect_hidden_states\u002F` (dumps),\n`scripts\u002F{export_hf_checkpoint,ar_validate,quantize_drafter}.py`. Read a wrapper for the\nflags it passes, that directory for what they do. Its `README.md` is also the non-launcher\npath: training invoked directly, without Slurm.\n\n## Stages\n\n| Stage | Reference | Use when |\n| --- | --- | --- |\n| 1. Configure | `references\u002Fstages\u002Fconfigure.md` | Adding a model that has no pipeline YAML yet |\n| 2. Review logs | `references\u002Fstages\u002Freview-logs.md` | A run finished (or died) and you want a pass\u002Ffail summary |\n| 3. Triage | `references\u002Fstages\u002Ftriage.md` | A task failed and you need root cause plus a fix |\n| 4. Validate | `references\u002Fstages\u002Fvalidate.md` | All tasks passed and you need to confirm the acceptance rate gate |\n\nReview-logs and triage overlap by design: review-logs is the fast sweep across all\ntasks, triage is the deep dive into one failing task. Start with review-logs unless\nthe user already knows which task broke.\n\n## Algorithms\n\nAll recipes live in `modelopt_recipes\u002Fgeneral\u002Fspeculative_decoding\u002F\u003Calgorithm>.yaml`.\n\n| Algorithm | Sheet | Family |\n| --- | --- | --- |\n| EAGLE3 | `references\u002Falgorithms\u002Feagle3.md` | Autoregressive draft head |\n| DFlash | `references\u002Falgorithms\u002Fdflash.md` | Block diffusion |\n| DSpark | `references\u002Falgorithms\u002Fdspark.md` | DFlash backbone + Markov head + optional confidence head |\n| Domino | `references\u002Falgorithms\u002Fdomino.md` | DFlash backbone + GRU causal correction head |\n| DFlash2 | `references\u002Falgorithms\u002Fdflash2.md` | DFlash backbone + sublayer convolutions + candidate selector |\n| LiLiCorr | `references\u002Falgorithms\u002Flilicorr.md` | DFlash backbone + candidate-lattice reranker (covers `lilicorr_conv.yaml`) |\n\nEverything below EAGLE3 is a **DFlash variant**, not a separate pipeline: same\n`recipe_type: speculative_dflash`, same training script, same `dflash.*` config\nnamespace, selected by `dflash_architecture_config.projector_type`. Read\n`references\u002Falgorithms\u002Fdflash.md` first, then the variant's sheet for the delta.\n\nLiLiCorr ships two recipes — `lilicorr.yaml` and `lilicorr_conv.yaml`, the latter\nadding DFlash2's sublayer convolutions — but one sheet covers both.\n\nIf the user's algorithm has no sheet yet, the stage procedures still apply — derive\nthe missing values from an existing launcher example for that algorithm\n(`tools\u002Flauncher\u002Fexamples\u002F*\u002F*\u002Fhf_*_\u003Calgorithm>.yaml`) and its recipe, then write the\nsheet as you go. `references\u002Falgorithms\u002FREADME.md` defines what a sheet must contain.\n\n## End-to-end: a new model\n\n1. Confirm the algorithm and find the closest existing launcher example.\n2. **Configure** — write the pipeline YAML (`references\u002Fstages\u002Fconfigure.md`).\n3. Preview with `--dryrun`, then submit:\n\n   ```bash\n   cd tools\u002Flauncher\n   uv run launch.py --yaml examples\u002F\u003COrg>\u002F\u003CModel>\u002F\u003Cconfig>.yaml --yes\n   ```\n\n4. Register the job and set up monitoring per the **modelopt-monitor skill**.\n5. **Review logs** when it finishes (`references\u002Fstages\u002Freview-logs.md`).\n6. **Triage** anything that failed (`references\u002Fstages\u002Ftriage.md`), fix, re-run only\n   the failed tasks onward via `pipeline.task_N.skip=true`.\n7. **Validate** once all tasks pass (`references\u002Fstages\u002Fvalidate.md`).\n\nModel-support gaps that need code changes land in `modelopt\u002Ftorch\u002Fspeculative\u002F` and\nrequire a separate ModelOpt PR — the pipeline YAML alone cannot fix an unrecognized\narchitecture.\n","# Speculative Decoding Draft-Model Training\n\nEverything needed to take a target model from \"no draft head\" to \"validated\nacceptance rate\" lives in this directory. Work through the stages below in order\nfor a new model; jump straight to a stage when you already know which one you need.\n\n## Two axes: stage and algorithm\n\nThe pipeline is the same shape for every draft-model algorithm — synthesize data,\ndump base-model hidden states, train the draft, benchmark acceptance rate. What\nchanges between algorithms is which training script and recipe run, which knobs\nmatter, and which failures are typical.\n\nSo this skill is split along those two axes, and **you almost always read one file\nfrom each**:\n\n| Axis | Directory | What it holds |\n| --- | --- | --- |\n| Stage | `references\u002Fstages\u002F` | The procedure — algorithm-independent |\n| Algorithm | `references\u002Falgorithms\u002F` | The data sheet — scripts, recipe, knobs, thresholds, known failures |\n\nRead the stage file for *what to do*, and the algorithm sheet for *the values to\nplug in*. When a stage file says \"see the algorithm sheet\", it means the section of\n`references\u002Falgorithms\u002F\u003Calgorithm>.md` with the matching heading.\n\n## Where the code lives\n\nThe launcher scripts in `tools\u002Flauncher\u002Fcommon\u002F{eagle3,specdec}\u002F` are thin wrappers around\n`examples\u002Fspeculative_decoding\u002F` — `main.py` (training), `collect_hidden_states\u002F` (dumps),\n`scripts\u002F{export_hf_checkpoint,ar_validate,quantize_drafter}.py`. Read a wrapper for the\nflags it passes, that directory for what they do. Its `README.md` is also the non-launcher\npath: training invoked directly, without Slurm.\n\n## Stages\n\n| Stage | Reference | Use when |\n| --- | --- | --- |\n| 1. Configure | `references\u002Fstages\u002Fconfigure.md` | Adding a model that has no pipeline YAML yet |\n| 2. Review logs | `references\u002Fstages\u002Freview-logs.md` | A run finished (or died) and you want a pass\u002Ffail summary |\n| 3. Triage | `references\u002Fstages\u002Ftriage.md` | A task failed and you need root cause plus a fix |\n| 4. Validate | `references\u002Fstages\u002Fvalidate.md` | All tasks passed and you need to confirm the acceptance rate gate |\n\nReview-logs and triage overlap by design: review-logs is the fast sweep across all\ntasks, triage is the deep dive into one failing task. Start with review-logs unless\nthe user already knows which task broke.\n\n## Algorithms\n\nAll recipes live in `modelopt_recipes\u002Fgeneral\u002Fspeculative_decoding\u002F\u003Calgorithm>.yaml`.\n\n| Algorithm | Sheet | Family |\n| --- | --- | --- |\n| EAGLE3 | `references\u002Falgorithms\u002Feagle3.md` | Autoregressive draft head |\n| DFlash | `references\u002Falgorithms\u002Fdflash.md` | Block diffusion |\n| DSpark | `references\u002Falgorithms\u002Fdspark.md` | DFlash backbone + Markov head + optional confidence head |\n| Domino | `references\u002Falgorithms\u002Fdomino.md` | DFlash backbone + GRU causal correction head |\n| DFlash2 | `references\u002Falgorithms\u002Fdflash2.md` | DFlash backbone + sublayer convolutions + candidate selector |\n| LiLiCorr | `references\u002Falgorithms\u002Flilicorr.md` | DFlash backbone + candidate-lattice reranker (covers `lilicorr_conv.yaml`) |\n\nEverything below EAGLE3 is a **DFlash variant**, not a separate pipeline: same\n`recipe_type: speculative_dflash`, same training script, same `dflash.*` config\nnamespace, selected by `dflash_architecture_config.projector_type`. Read\n`references\u002Falgorithms\u002Fdflash.md` first, then the variant's sheet for the delta.\n\nLiLiCorr ships two recipes — `lilicorr.yaml` and `lilicorr_conv.yaml`, the latter\nadding DFlash2's sublayer convolutions — but one sheet covers both.\n\nIf the user's algorithm has no sheet yet, the stage procedures still apply — derive\nthe missing values from an existing launcher example for that algorithm\n(`tools\u002Flauncher\u002Fexamples\u002F*\u002F*\u002Fhf_*_\u003Calgorithm>.yaml`) and its recipe, then write the\nsheet as you go. `references\u002Falgorithms\u002FREADME.md` defines what a sheet must contain.\n\n## End-to-end: a new model\n\n1. Confirm the algorithm and find the closest existing launcher example.\n2. **Configure** — write the pipeline YAML (`references\u002Fstages\u002Fconfigure.md`).\n3. Preview with `--dryrun`, then submit:\n\n   ```bash\n   cd tools\u002Flauncher\n   uv run launch.py --yaml examples\u002F\u003COrg>\u002F\u003CModel>\u002F\u003Cconfig>.yaml --yes\n   ```\n\n4. Register the job and set up monitoring per the **modelopt-monitor skill**.\n5. **Review logs** when it finishes (`references\u002Fstages\u002Freview-logs.md`).\n6. **Triage** anything that failed (`references\u002Fstages\u002Ftriage.md`), fix, re-run only\n   the failed tasks onward via `pipeline.task_N.skip=true`.\n7. **Validate** once all tasks pass (`references\u002Fstages\u002Fvalidate.md`).\n\nModel-support gaps that need code changes land in `modelopt\u002Ftorch\u002Fspeculative\u002F` and\nrequire a separate ModelOpt PR — the pipeline YAML alone cannot fix an unrecognized\narchitecture.\n",{"name":5,"description":7,"user_invocable":24},true,"main","e169b4c2bc62a32c45aaef793cde7599dee2c1b2","skills\u002Fmodelopt-speculative-decoding",806,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer","https:\u002F\u002Fnvidia.github.io\u002FModel-Optimizer\u002F","https:\u002F\u002Fgithub.com\u002FNVIDIA","https:\u002F\u002Fnvidia.com","2026-11-15 02:45:59",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":24},[36,39,42],{"tool":37,"command":38},"npx","npx skills add NVIDIA\u002FModel-Optimizer --skill modelopt-speculative-decoding",{"tool":40,"command":41},"bunx","bunx skills add NVIDIA\u002FModel-Optimizer --skill modelopt-speculative-decoding",{"tool":43,"command":44},"pnpm","pnpm dlx skills add NVIDIA\u002FModel-Optimizer --skill modelopt-speculative-decoding","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002FSKILL.md",[47,53,57,61,65,69,73,77,81,85,89],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Falgorithms\u002Fdflash.md","REFERENCE","参考文档",11665,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002Fdflash.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Falgorithms\u002Fdflash2.md",9859,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002Fdflash2.md",{"path":58,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":59,"sourceHtmlUrl":60},"references\u002Falgorithms\u002Fdomino.md",7147,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002Fdomino.md",{"path":62,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":63,"sourceHtmlUrl":64},"references\u002Falgorithms\u002Fdspark.md",10932,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002Fdspark.md",{"path":66,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":67,"sourceHtmlUrl":68},"references\u002Falgorithms\u002Feagle3.md",6333,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002Feagle3.md",{"path":70,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":71,"sourceHtmlUrl":72},"references\u002Falgorithms\u002Flilicorr.md",13262,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002Flilicorr.md",{"path":74,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":75,"sourceHtmlUrl":76},"references\u002Falgorithms\u002FREADME.md",2588,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Falgorithms\u002FREADME.md",{"path":78,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":79,"sourceHtmlUrl":80},"references\u002Fstages\u002Fconfigure.md",2214,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Fstages\u002Fconfigure.md",{"path":82,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":83,"sourceHtmlUrl":84},"references\u002Fstages\u002Freview-logs.md",2968,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Fstages\u002Freview-logs.md",{"path":86,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":87,"sourceHtmlUrl":88},"references\u002Fstages\u002Ftriage.md",6374,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Fstages\u002Ftriage.md",{"path":90,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":91,"sourceHtmlUrl":92},"references\u002Fstages\u002Fvalidate.md",5026,"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-speculative-decoding\u002Freferences\u002Fstages\u002Fvalidate.md",[94,101,108,115,122,129],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":19},"01M7B5X9SG5FHW174VFK979DQ9","modelopt-run-agentx","Run the AgentX agentic serving benchmark with the SemiAnalysis harness. Use for AgentX setup, dataset downloads, concurrency sweeps, or serving latency and throughput comparisons.","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-run-agentx\u002FSKILL.md",[],[],{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":19},"01M7B5X4C0XGVA8N0BXF9N4HCX","modelopt-release-cherry-pick","Audit merged bug-fix PRs and release NVBugs for missing cherry-pick labels, then cherry-pick labeled PRs into a release branch and open a PR. Use when asked to \"cherry-pick PRs for release\u002FX.Y.Z\", \"pick PRs to release branch\", \"verify cherry-pick labels\", or \"cherry-pick labeled PRs\".","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-release-cherry-pick\u002FSKILL.md",[],[],{"skillNo":109,"name":110,"nameZh":6,"description":111,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":112,"categories":113,"keywords":114,"publishedAt":19},"01M7B5WXMDYXCK43YA5ZEF7ECY","modelopt-quant-recipe-search","Use when the user asks to find, search for, or optimize the best quantization recipe for a model, including direct requests like \"find the best quantization recipe and generate a PTQ checkpoint.\" Guides the multi-candidate loop: choose compute-vs-memory success metrics, select ModelOpt recipe baselines, design AutoQuant\u002Fmanual recipe deltas, interpret sensitivity, and decide next candidates. Do NOT use for a single known PTQ recipe run (use modelopt-ptq), serving (use modelopt-deployment), creating\u002Frunning evals (use modelopt-evaluation or modelopt-launching-evals), monitoring jobs (use modelopt-monitor), MLflow browsing (use modelopt-accessing-mlflow), or comparing completed baseline-vs-candidate scores only (use modelopt-compare-results).","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-quant-recipe-search\u002FSKILL.md",[],[],{"skillNo":116,"name":117,"nameZh":6,"description":118,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":119,"categories":120,"keywords":121,"publishedAt":19},"01M7B5WRMPZE23YXCFMZBM8KJA","modelopt-qad","Run explicitly requested ModelOpt Quantization-Aware Distillation (QAD) through Megatron Bridge (recommended) or Hugging Face Transformers to recover a measured BF16-to-PTQ accuracy gap. Use only when the user explicitly asks for QAD, including its topology, data preparation, launch, resume, checkpoint export, or recovery decisions.","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-qad\u002FSKILL.md",[],[],{"skillNo":123,"name":124,"nameZh":6,"description":125,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":126,"categories":127,"keywords":128,"publishedAt":19},"01M7B5WJHB4D787EZCY4RNN44T","modelopt-ptq","Use when the user asks to \"quantize a model\", \"run PTQ\", \"post-training quantization\", \"NVFP4 quantization\", \"FP8 quantization\", \"INT8 quantization\", \"INT4 AWQ\", \"quantize LLM\", \"quantize MoE\", \"quantize VLM\", or needs to produce a quantized HuggingFace checkpoint from a pretrained model using ModelOpt. Do NOT use for multi-candidate recipe exploration or optimization (use modelopt-quant-recipe-search).","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-ptq\u002FSKILL.md",[],[],{"skillNo":130,"name":131,"nameZh":6,"description":132,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":133,"categories":134,"keywords":135,"publishedAt":19},"01M7B5WDB7KYNXD84PQRA69RPT","modelopt-monitor","Monitor submitted jobs (PTQ, evaluation, deployment) on SLURM clusters. Use when the user asks \"check job status\", \"is my job done\", \"monitor my evaluation\", \"what's the status of the PTQ\", \"check on job\" followed by a SLURM job ID, or after any skill submits a long-running job. Also triggers on \"nel status\", \"squeue\", or any request to check progress of a previously submitted job.","https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FModel-Optimizer\u002Fblob\u002Fmain\u002Fskills\u002Fmodelopt-monitor\u002FSKILL.md",[],[]]