Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Route LLM requests to the best model for the task at hand.
| Date | Stars |
|---|---|
| 2026-07-31 | 332 |
| 2026-08-03 | 332 |
| 2026-08-06 | 332 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<h2>Deprecated: NVIDIA AI Blueprint for LLM Router </h2> > **⚠️ Deprecation Notice**: NVIDIA is excited to announce a new routing library, [NVIDIA NeMo Switchyard](https://github.com/NVIDIA-NeMo/Switchyard). The Switchyard library implements the same patterns as LLM Router v1 and LLM Router v2 and adds broader extensibility, improved performance, and deeper integrations with the NVIDIA NeMo Platform. ## Important Notes **LLM Router v2 is currently experimental** and not yet backwards compatible with v1. Key differences: | Feature | v1 (Main Branch) | v2 (Experimental) | |---------|------------------|-------------------| | **Server Implementation** | Rust proxy | NVIDIA NeMo Agent Toolkit (FastAPI) | | **Inference Backend** | BERT model + NVIDIA Triton Inference Server | Qwen 1.7B LLM or CLIP + Neural Network | | **Functionality** | Classification + Proxying to LLM | Classification only (returns model name) | | **Input Support** | Text only | Text + Images (multimodal) | | **Routing Methods** | Task or complexity classification | Intent-based or Auto-routing (neural network) | **Future Plans**: The intent is to make v2 fully backwards compatible with v1's proxying capabilities, then merge to main and retire the experimental label. ## Overview Ever struggled to decide which LLM or Vision-Language Model (VLM) to use for a specific task? In an ideal world the most accurate model would also be the cheapest and fastest, but in practice modern agentic AI systems have to make trade-offs between accuracy, speed, and cost. This blueprint provides an experimental next-generation router that automates these tradeoffs by analyzing user prompts and identifying optimal models. Given a user prompt (text or multimodal), the router: - applies one of two routing strategies: intent-based classification or auto-routing (based on a trained neural network) - analyzes the prompt content, including images if present - returns the name of the most appropriate LLM or VLM for the task For example, using intent-based routing: | User Prompt | Intent Classification | Recommended Model | |---|---|---| | "What's in this image?" (with image) | image_understanding | nvidia/nemotron-nano-12b-v2-vl | | "Solve this complex math problem: ..." | hard_question | gpt-5-chat | | "Hello, how are you?" | chit_chat | nvidia/nvidia-nemotron-nano-9b-v2 | The key features of the experimental LLM Router v2 are: - **Multimodal Support**: Route based on both text and images, optimized for VLMs - **Two Routing Strategies**: Intent-based (using Qwen 1.7B) OR auto-routing (using CLIP embeddings + trained neural network) - **OpenAI API compliant**: Returns model recommendations via chat completions endpoint - **Flexible**: Use pre-configured intent mappings or train custom neural network routers on your own data ### Models This blueprint is pre-configured to route between three complementary models: | Model | Type | Provider | Use Case | |-------|------|----------|----------| | **gpt-5-chat** | Frontier LLM | Azure OpenAI or OpenAI | Complex reasoning, hard questions | | **nvidia/nemotron-nano-12b-v2-vl** | Open VLM | NVIDIA Build API | Multimodal queries, image understanding | | **nvidia/nvidia-nemotron-nano-9b-v2** | Small Open LLM | NVIDIA Build API | Simple text queries, chit chat | - **gpt-5-chat**: Can be sourced from Azure OpenAI (default) or standard OpenAI API - **Nemotron models**: Configured to use NVIDIA Build API endpoints (hosted), but can also use locally deployed NVIDIA NIMs for on-premise deployment #### Using Different Models The three default models are **examples only** - you can route to any models by (1) updating the intent router's configuration or (b) re-training the auto-router. *The main goal of the LLM router is to intelligently route across frontier and open models* to optimize the cost-quality-latency tradeoff. ## Target Audience This experimental blueprint is for: - **AI Engineers and Developers**: Developers interested in explo
Excerpt of 19,750 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:6174027f623c3e81, llm:description: 'Route LLM requests to the best model for the task at hand.'; topic: 'llm-router'
matched fp:6174027f623c3e81, llm:description: 'Route LLM requests to the best model for the task at hand.'; topic: 'llm-router'
matched fp:6174027f623c3e81, llm:description: 'Route LLM requests to the best model for the task at hand.'; topic: 'llm-router'