Skip to content

adityonugrohoid/nim-explorer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

NIM Explorer

Python 3.12+ License: MIT Models

Machine-readable catalog of 107 Nvidia NIM LLM models with capability detection and an HTML browser.

Getting Started | Model Catalog | API Reference | Scripts


Features

  • Model Catalog -- JSON catalog of all 107 available NIM models with metadata (family, params, capabilities, quirks)
  • HTML Browser -- Static page to filter, search, and explore models by capability
  • Capability Probing -- Automated detection of tool calling, JSON mode, thinking support per model
  • API Reference -- Complete Nvidia NIM endpoint documentation based on live testing

Getting Started

Prerequisites

Installation

git clone https://github.com/adityonugrohoid/nim-explorer.git
cd nim-explorer

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Configuration

cp .env.example .env

Edit .env with your Nvidia API key:

NVIDIA_API_KEY=nvapi-...

Model Catalog

The catalog (models/catalog.json) contains metadata for every callable model on Nvidia NIM's free tier.

By Organization

Organization Models Highlights
meta 11 Llama 3.x/4, vision, guard
nvidia 16 Nemotron (thinking + tools), safety, translation
mistralai 13 Mistral/Mixtral, Codestral, Mathstral
microsoft 11 Phi 3/3.5/4, reasoning, vision
google 10 Gemma 2/3/3n, ShieldGemma
qwen 9 Qwen 2/2.5/3/3.5, QwQ, coding
deepseek-ai 5 R1-distill (thinking), V3.x
20+ others 32 Regional, specialized, emerging

Refreshing the Catalog

# Probe all models and rebuild catalog.json
python scripts/build_catalog.py

API Reference

See docs/api-reference.md for complete Nvidia NIM endpoint documentation including:

  • All request parameters with types, defaults, and ranges
  • Response shape variations by model backend
  • Tool calling format and supported models
  • Thinking/reasoning patterns (3 different approaches)
  • Streaming SSE format
  • Error response inconsistencies

Scripts

# Build the full model catalog
python scripts/build_catalog.py

# Probe a single model for detailed info
python scripts/probe_model.py meta/llama-3.3-70b-instruct

Project Structure

nim-explorer/
├── models/
│   └── catalog.json            # Machine-readable model catalog
├── docs/
│   ├── index.html              # HTML model browser
│   └── api-reference.md        # Complete NIM API documentation
├── scripts/
│   ├── build_catalog.py        # Probe all models, build catalog
│   └── probe_model.py          # Probe single model
├── results/
│   └── raw/                    # Raw API responses (gitignored)
├── CLAUDE.md
├── README.md
├── ROADMAP.md
├── requirements.txt
└── .env.example

Roadmap

  • v0.1: Model catalog + HTML browser + API reference
  • v0.2: Tool calling / JSON mode / thinking probes + benchmarks
  • v0.3: Arena integration (multi-model comparison app)

License

This project is licensed under the MIT License.

Author

Adityo Nugroho (@adityonugrohoid)

About

Nvidia NIM model catalog, benchmarks & comparison toolkit. 107 models probed and cataloged.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages