Open Source Today (2024-07-24): Meta Officially Releases Llama 3.1
Explore cutting-edge AI models like Meta’s Llama 3.1 and Alibaba’s AudioLCM. Discover efficient tools like LlamaParse, Mandark, StockBot, and more.
Here are some interesting AI open-source models and frameworks I wanted to share today:
Project: Llama 3.1
Meta has open-sourced Llama 3.1, which includes 8B, 70B, and 405B parameter versions. Each comes in both base and instruction-tuned variants.
Llama 3.1 supports 128k context and is proficient in eight languages, including English, German, and French.
It was trained on over 15 trillion tokens on a custom GPU cluster, taking a total of 39.3 million GPU hours.
Notably, Llama 3.1 uses high-quality synthetic data for SFT tasks such as code generation, math, reasoning, and tool usage.
https://huggingface.co/collections/meta-llama/llama-31-669fc079a0c406a149a5738f
Project: AudioLCM
AudioLCM, developed by Alibaba and Zhejiang University, is a high-quality text-to-audio model.
It synthesizes high-fidelity audio in just two iterations, matching the quality of state-of-the-art models that require hundreds of steps.
AudioLCM achieves a sampling speed 333 times faster than real-time on a single NVIDIA 4090Ti GPU.
The project supports training and inference with custom datasets.
https://arxiv.org/abs/2406.00356v1
https://github.com/Text-to-Audio/AudioLCM
Project: LlamaParse
LlamaParse is an API by LlamaIndex designed for efficient file parsing and representation, enhancing retrieval and context using the LlamaIndex framework.
It integrates directly with LlamaIndex, offering free and paid plans. The free plan supports parsing up to 1,000 pages per day, and the paid plan supports 7,000 pages per week.
https://github.com/run-llama/llama_parse
Project: Mandark
Mandark is a simple AI coding assistant that can handle most coding tasks, including self-improvement.
It supports multiple AI models and can edit and create files, verify command-line differences, and estimate tokens and costs before execution.
Mandark works with any codebase and runs without installation.
https://github.com/hrishioa/mandark
Project: StockBot
StockBot is an AI chatbot powered by Llama3-70b, Groq, Vercel AI SDK, and TradingView widgets.
It provides real-time, interactive stock charts, financial data, news, and stock screeners through natural language conversations.
Groq's high performance enables instant tool calls and responses, offering an excellent user experience.
https://github.com/bklieger-groq/stockbot-on-groq
Project: Awesome-Role-Play-Papers
This project compiles excellent papers, datasets, and codebases related to language model role-playing.
While it covers most research in the role-playing field, it may not include all notable works.
The project aims to provide a comprehensive resource collection to help researchers and developers better understand and apply role-playing techniques.