Model Optimizer plugin for Grok Build

Name: Model Optimizer
Author: NVIDIA

Community Claude Code plugin, installs in GrokPluginDeploymentMCP: manual setup

A unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture search, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed.

How to install Model Optimizer in Grok Build

1. Add this repo as a plugin marketplace in Grok Build (once per computer):

grok plugin marketplace add NVIDIA/Model-Optimizer

2. Install the plugin:

grok plugin install modelopt@NVIDIA/Model-Optimizer --trust

✓ Install-checked in Grok Buildchecked Oct 18, 2026

Review before you trust it. --trust lets this plugin's hooks, MCP servers and scripts run code on your computer. It's a community repo that neither xAI nor GrokStacks has reviewed, so read its code first.

Then start a new Grok session, or press r in /plugins, to load it.

Inside Grok Build: type /plugins, open the Marketplace tab, press a and add NVIDIA/Model-Optimizer, then select modelopt and press i.

Command format: Grok Build's plugin guide

Skills work, MCP part needs manual setup. Grok loads Model Optimizer's skills and commands, but not its MCP server: it needs settings that Claude Code asks for when you install it (userConfig), and Grok doesn't ask for them. Add the server yourself:

grok mcp add modelopt uvx -- --from 'git+https://github.com/NVIDIA/Model-Optimizer.git#subdirectory=tools/mcp' modelopt-mcp

Then start a new session or press r in /mcps. If it asks you to sign in, select it in /mcps and press i.

What we checked: on Oct 18, 2026 it installed with grok 1.0.62 and Grok found 16 skills, 1 subagent. That's an install check only: we haven't run it in a live Grok session.

What it contains

Skills (16)

accessing-mlflow — Query and browse evaluation results stored in MLflow. Use when the user wants to look up runs by invocation ID, compare metrics across models, fetch artifacts (configs, logs, results), or set up the MLflow MCP server. ALWAYS triggers on men
benchmark-model-kernels — Inspect Hugging Face decoder layers on meta tensors and plan or run per-rank BF16, FP8, and NVFP4 GEMM or fused-MoE microbenchmarks with the bundled scripts and a local FlashInfer checkout. Use when choosing a model, GPU, TP, EP, or M/token
common — Shared ModelOpt support files. Use only when another ModelOpt skill directs you here.
compare-results — Establish baseline-vs-candidate evaluation plans, delegate missing evaluations, compare validated results, and decide quantization feasibility. Use when the user asks to compare baseline vs quantized runs, explain an accuracy drop/regressio
day0-release — Deterministic end-to-end driver for day-0 quantized-checkpoint releases — chains PTQ → evaluation → comparison with enforced gates between stages (the evaluation stage deploys the checkpoint itself), and returns a publish decision (ACCEPT /
debug — Run commands inside a remote Docker container via the file-based command relay (tools/debugger). Use when the user says "run in Docker", "run on GPU", "debug remotely", "run test in container", "check nvidia-smi", "run pytest in Docker", or
deployment — Serve a quantized or unquantized LLM checkpoint as an OpenAI-compatible API endpoint using vLLM, SGLang, or TRT-LLM. Use when user says "deploy model", "serve model", "start vLLM server", "launch SGLang", "TRT-LLM deploy", "AutoDeploy", "be
evaluation — Evaluates accuracy of quantized or unquantized LLMs using NeMo Evaluator Launcher (NEL). Triggers on "evaluate model", "benchmark accuracy", "run MMLU", "evaluate quantized model", "run nel". Handles deployment, config generation, and evalu
launching-evals — Run, monitor, analyze, and debug LLM evaluations via nemo-evaluator-launcher. Covers running evaluations, checking status and live progress, debugging failed runs, exporting artifacts and logs, and analyzing results. ALWAYS triggers on ment
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 <slurm_job_id>", or after any skill
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 quantize
qad — Run explicitly requested ModelOpt Quantization-Aware Distillation (QAD) on Slurm through Megatron Bridge to recover a measured BF16-to-PTQ accuracy gap. Use only when the user explicitly asks for QAD, including its topology, data preparatio
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 comp
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/X.Y.Z", "pick PRs to release branch", "verify cherr
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.
speculative-decoding — 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,

Subagents (5)

modelopt:modelopt-model-downloader — Use this agent when a Hugging Face model must be staged in a Day 0 workspace. <example>The user asks to download a model for quantization. Use this agent.</example> <example>A remote workflow needs an
modelopt:modelopt-model-evaluator — Use this agent when a baseline or candidate needs one comparable NEL accuracy evaluation. <example>The user asks for baseline accuracy. Use this agent for the baseline run.</example> <example>A quanti
modelopt:modelopt-model-performance-benchmarker — Use this agent when a verified Model Optimizer endpoint needs AIPerf measurement. <example>The user asks for throughput and latency results. Use this agent.</example> <example>A candidate needs a matc
modelopt:modelopt-model-quantize-recipe-searcher — Use this agent when a Day 0 quantization search needs its next evidence-backed candidate. <example>The user asks which recipe to try next. Use this agent.</example> <example>Evaluation evidence rules
modelopt:modelopt-model-quantizer — Use this agent when one selected Day 0 recipe needs a validated Model Optimizer PTQ checkpoint. <example>The user asks to quantize a model with a chosen recipe. Use this agent.</example> <example>A se

About this listing

Model Optimizer is a community plugin on GitHub by NVIDIA, written as a Claude Code plugin that Grok Build installs. We install-checked it in Grok; we haven't run it in a live session or reviewed its code. When we installed it, Grok found 16 skills, 5 subagents. Its license is Apache-2.0. We checked commit d7cbff3486a3. It's one of 41 listings in the Deployment category on GrokStacks. It also appears in best Grok Build plugins for deployment and hosting.

Model Optimizer FAQ

How do I install Model Optimizer in Grok Build?

Add the repo as a marketplace with grok plugin marketplace add NVIDIA/Model-Optimizer, then run grok plugin install modelopt@NVIDIA/Model-Optimizer --trust. Then start a new Grok session, or press r in /plugins, to load it.

Is Model Optimizer an official Grok Build plugin?

No. Model Optimizer is a community plugin on GitHub by NVIDIA, not in xAI's official Grok Build marketplace. We checked that it installs and what Grok loads from it, but haven't run it in a live Grok session or reviewed its code.

What does Model Optimizer include?

According to what Grok loaded in our install check, Model Optimizer includes 16 skills, 5 subagents: accessing-mlflow, benchmark-model-kernels, common, compare-results, day0-release, debug….

Details

Type: Plugin
Category: Deployment
Source: Community Claude Code plugin, installs in Grok
License: Apache-2.0
Author: NVIDIA
Source repo: https://github.com/NVIDIA/Model-Optimizer
Checked commit: d7cbff3486a3
Homepage: https://nvidia.github.io/Model-Optimizer/
Checked by GrokStacks: Install-checked in Grok Build on Oct 18, 2026; not run in a live session

Something wrong or missing? Submit or improve this listing. Marketplace listings also refresh automatically every night.

For maintainers: badge and embeds

Maintain Model Optimizer? Add the badge; it links to this page.

Markdown

[![Listed on GrokStacks](https://grokstacks.com/badge.svg)](https://grokstacks.com/p/model-optimizer/)

HTML

<a href="https://grokstacks.com/p/model-optimizer/"><img src="https://grokstacks.com/badge.svg" alt="Listed on GrokStacks" width="164" height="28"></a>

Embed card (static HTML and inline CSS, no JavaScript). Preview:

Model OptimizerCommunity plugin for Grok Build · 16 skills, 5 subagentsA unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture…View on GrokStacks →

<a href="https://grokstacks.com/p/model-optimizer/" style="display:block;max-width:380px;box-sizing:border-box;padding:14px 16px;border:1px solid #2a2e3c;border-radius:12px;background:#0a0b0f;color:#e8eaf0;font:14px/1.45 -apple-system,Segoe UI,Roboto,Helvetica,Arial,sans-serif;text-decoration:none"><span style="display:block;font-size:16px;font-weight:700;color:#e8eaf0">Model Optimizer</span><span style="display:block;margin:2px 0 8px;font-size:12px;color:#9aa0b1">Community plugin for Grok Build · 16 skills, 5 subagents</span><span style="display:block;margin:0 0 10px;color:#c9cdd8">A unified library of SOTA model optimization techniques like quantization, distillation, pruning, neural architecture…</span><span style="font-size:12px;font-weight:600;color:#7c9cff">View on GrokStacks →</span></a>

Markdown summary

**[Model Optimizer on GrokStacks](https://grokstacks.com/p/model-optimizer/)**: Community plugin for Grok Build with 16 skills, 5 subagents. Install: `grok plugin install modelopt@NVIDIA/Model-Optimizer --trust` (after `grok plugin marketplace add NVIDIA/Model-Optimizer`)

More on badges, embeds and keeping your listing up to date: GrokStacks for plugin authors.

Also in Deployment

Affiliate Skills

50 AI agent skills for affiliate marketing. Research trending content, write data-backed posts, generate infographics, build landing pages, deploy — full flywheel with social intelligence. Works with Claude Code, Pi, ChatGPT, Gemini, Cursor, Windsurf, any AI.

Community Claude Code plugin, installs in GrokPluginDeployment