How cobusgreyling/nemotron-3-ultra-showcase works
This is a Gradio feature showcase for NVIDIA Nemotron 3 Ultra. A user selects a reasoning mode and budget, submits a prompt, and sees streamed reasoning and an answer from the NVIDIA endpoint. A separate tool-calling workflow streams tool requests, executes the bundled arithmetic evaluator locally, then requests a final model response. The app also presents model architecture and generation guidance. The single source file implements the runtime; relationships not visible in the excerpts are supported only where the README documents them.
Main entry points: app.py Gradio tabs
Components 8
Showcase interface
- Gradio tabs
app.py - Reasoning playground
app.py - Model card
app.py
Model interaction
- Reasoning stream
app.py - Nemotron endpoint
Tool execution
- Tool-call loop
app.py - Safe math evaluator
app.py
Other components
- User
Connections 10
- User → Gradio tabs: uses
README.md - Gradio tabs → Reasoning playground: offers
README.md - Gradio tabs → Model card: offers
README.md - Reasoning playground → Reasoning stream: submits prompt
README.md - Reasoning stream → Nemotron endpoint: streams request
README.md - Nemotron endpoint → Reasoning stream: streams response
README.md - Gradio tabs → Tool-call loop: offers tool demo
README.md - Tool-call loop → Nemotron endpoint: requests tool calls
README.md - Tool-call loop → Safe math evaluator: executes tool
app.py - Tool-call loop → Nemotron endpoint: requests final answer
app.py