IN TODAY'S SIGNAL

📰 Top News

• NVIDIA's releases Nemotron 340B, an open LLM matching GPT-4 performance for chat applications and synthetic data generation.

📌 AI4 Conference

• Join the world's top AI conference: Free passes, 350+ speakers, 150+ exhibitors.

⚡️ Top 5 Signals

• Runway releases a powerful video generation model: Gen-3 Alpha
  
  

• Abacus and Yann Lecun launched an ungameable LLM benchmark.
  
  

• Apple drops 20 CoreML models and 4 datasets on HuggingFace.
  
  

• Mistral's releases a series of new tutorials, builds RAG pipeline.
  
  

• Cohere CEO: AI will generate money by itself sooner than expected.

🛠️ Top of Github

• Google Mesop: Rapidly build Python web apps with strong typing.
  
  

• Cognita: helps quickly build and deploy modular RAG systems
  
  

• Vanna: Generate accurate SQL queries from natural language using RAG.

🥬 Salad AI

• Discover how Civitai generates 10M images/day, trains 15K LoRAs/month using Salad.

🧠 Tutorial

• How to implement LLM instruction fine-tuning from scratch.

Enjoying this newsletter?
Please forward it to a friend or colleague. It helps us keep this content free.

TOP NEWS

Open Source

⇧ 2,104 Likes

What's New

NVIDIA recently launched the Nemotron 340B, a comprehensive suite designed for synthetic data generation that enhances the development of large language models (LLMs).

This release includes three specialized models: the Nemotron 340B Base, Instruct, and Reward, each tailored to optimize different stages of data generation and model training.

Key Features and Capabilities

• Advanced Data Generation: The Instruct model generates synthetic text replicating real-world data features, while the Reward model evaluates and refines this data across multiple quality attributes such as helpfulness and coherence. This model ranks first on the Hugging Face RewardBench leaderboard.
  
  
• Integration and Optimization: Fully compatible with NVIDIA NeMo and TensorRT-LLM, the models leverage tensor parallelism to efficiently distribute computations across multiple GPUs.

Training and Customization Options

• Customization Through NeMo: Developers can customize the base model, which has been pretrained on 9 trillion tokens, using various fine-tuning techniques like low-rank adaptation (LoRA) and supervised fine-tuning.
  
  
• Model Alignment: The NeMo Aligner allows developers to align model outputs with specific standards and goals through reinforcement learning from human feedback (RLHF), ensuring safety and accuracy.

Accessibility and Licensing

• Wide Accessibility: The models are available for download on Hugging Face and will soon be offered as an NVIDIA NIM microservice.
  
  
• Open Model License: NVIDIA provides these models under an open license, facilitating their broad distribution, modification, and use, helping overcome significant challenges in accessing quality training data.

Ai4, the world’s largest gathering of artificial intelligence leaders in business, is coming to Las Vegas - August 12-14, 2024.

Join 4500+ attendees, 350+ speakers, and 150+ AI exhibitors from 75+ countries at the epicenter of AI innovation.

Don’t wait - passes are going fast. Apply today for a complimentary pass or register now for 35% off final prices.

TRENDING SIGNALS

Video Generation

⇧ 2012  Likes

Benchmarking

⇧ 927 Likes

On-Device ML

⇧ 375 Likes

Open Source

⇧ 402 Likes

Opinion

⇧ 1248 Likes

The world's unused compute meets innovative AI companies. Civitai, one of the most visited AI sites, is serving inference on SaladCloud (powered by latent compute in everyday PCs).

By switching to Salad, Civitai generates 10 Million AI images per day & trains over 15,000 LoRAs per month.

Read the case study ↗️

TOP OF GITHUB

Web Apps

☆ 3048

Mesop helps you rapidly build Python-based web apps. It offers an intuitive UI framework, allowing you to write UI in idiomatic Python. Mesop supports hot reload, strong type safety, and component-based architecture. Used at Google, it enables frictionless development for demos and internal apps. Get started in less than 10 lines of code.

RAG

☆ 2809

Cognita helps quickly build and deploy modular RAG systems. It integrates parsers, embedders, retrievers, and LLMs for production-ready applications. Cognita supports incremental indexing, multi-modal parsing, and custom query controllers. 

Database

☆ 7991

Vanna helps you generate SQL queries from natural language using Retrieval-Augmented Generation (RAG). Train a model on your database schema and ask questions to get accurate SQL code. Vanna supports any SQL database, maintains high accuracy with complex datasets, and runs queries locally for security.

TUTORIAL

Fine-tuning