Today's Open Source (2024-08-29): AI Chat with LLM Routing; Anthropic Prompt Tutorials
Explore the latest AI open-source projects like AI Router Chat, Anthropic Courses, UniBench, AutoGluon-RAG, Sapiens, and AnyGraph. Enhance your AI toolkit today!
Here are some interesting AI open-source models and frameworks I wanted to share today:
Project: AI Router Chat
AI Router Chat is a chat app with an LLM routing module. It dynamically improves model routing to give users better answers.
Users can run the app by integrating API keys from multiple LLM providers like OpenAI, Anthropic, Groq, Perplexity, and Google.
https://github.com/mckaywrigley/ai-router-chat
Project: Anthropic Courses
Anthropic's educational courses are official lessons on using the Claude SDK for prompt engineering and tools.
The courses cover basics like getting API keys, setting model parameters, writing multimodal prompts, and streaming responses. They include interactive tutorials and real-world examples to help users apply prompt techniques.
https://github.com/anthropics/courses
Project: UniBeach
UniBench is a Python library for evaluating the robustness of Visual Language Models (VLMs).
It offers tools and scripts to simplify the evaluation process for VLMs and benchmarks.
The project includes 60 VLMs, such as the latest large-scale model EVA-CLIP, with up to 4.3 billion parameters and 12.8 billion training samples. It also provides 40 benchmark implementations.
https://github.com/facebookresearch/unibench
https://arxiv.org/abs/2408.04810
Project: AutoGluon-RAG
AutoGluon-RAG is a framework that simplifies RAG development.
Traditionally, building RAG pipelines required handling complex modules. AutoGluon-RAG allows users to create custom RAG pipelines with just a few lines of code.
The framework has a user-friendly interface, abstracting the underlying modules, so users can focus on their specific needs without getting into technical details.
https://github.com/autogluon/autogluon-rag
Project: Sapiens
Sapiens offers a comprehensive set of human vision task models (e.g., 2D pose, part segmentation, depth, normals).
The models are pre-trained on 300 million images of people in the wild, showing strong generalization in unconstrained conditions.
These models are designed to extract high-resolution features, trained natively at 1024 x 1024 resolution with 16-pixel patch sizes.
https://github.com/facebookresearch/sapiens
https://arxiv.org/abs/2408.12569
Project: AnyGraph
AnyGraph is a graph-based model designed for zero-shot prediction across domains.
The model handles distribution shifts in graph structures and diverse feature spaces, adapting efficiently to new graph domains.
AnyGraph demonstrates improved performance with increased data and model parameters, showing wide adaptability and strong generalization.
