DeepSpeed-Chat: Affordable RLHF Training for AI
Plus: OpenAI's new ChatGPT updates, Latest Vicuna models based on LLaMA-2.
Hello Engineering Leaders and AI Enthusiasts!
Welcome to the 78th edition of The AI Edge newsletter. This edition brings you DeepSpeed-Chat for quick & affordable RLHF training for LLMs at all scales.
And a huge shoutout to our amazing readers. Your support is invaluable to us!😊
In today’s edition:
🤖 DeepSpeed-Chat: Affordable RLHF training for AI
🔥 OpenAI is rolling out new updates to improve ChatGPT
🚀 Latest versions of Vicuna, based on the open LLaMA-2
🧠 Knowledge Nugget: Top Challenges LLMs Need to Address, along with Possible Solutions by
Let’s go!
DeepSpeed-Chat: Affordable RLHF training for AI
New Microsoft research has introduced DeepSpeed-Chat, a novel system that makes complex RLHF (Reinforcement Learning with Human Feedback) training fast, affordable, and easily accessible to the AI community (open-sourced). It has three key capabilities:
Easy-to-use Training and Inference Experience for ChatGPT Like Models
A DeepSpeed-RLHF pipeline that replicates the training pipeline from InstructGPT
A robust DeepSpeed-RLHF system that combines various optimizations for training and inference in a unified way
The system delivers unparalleled efficiency and scalability, enabling training of models with hundreds of billions of parameters in record time and at a fraction of the cost. Here’s how it compares to two other frameworks (Colossal-AI and HuggingFace DDP) for accelerating RLHF training on a single NVIDIA A100-40G commodity GPU.
Why does it matter?
The current landscape lacks an accessible, efficient, and cost-effective end-to-end RLHF training pipeline for powerful models like ChatGPT, particularly when training at the scale of billions of parameters. DeepSpeed-Chat paves the way for broader access to advanced RLHF training, thereby fostering innovation and further development in the field of AI.
(Source)
OpenAI is rolling out new updates to improve ChatGPT
OpenAI is shipping out a bunch of small updates over the next week to improve the ChatGPT experience. Here’s a tl;dr
1. Prompt examples: At the beginning of a new chat, you will now see examples to help you get started.
2. Suggested replies: ChatGPT will suggest relevant ways to continue your conversation.
3. GPT-4 by default: When starting a new chat as a Plus user, ChatGPT will remember your previously selected model – no more defaulting back to GPT-3.5.
4. Upload multiple files: Now, ChatGPT can analyze data and generate insights across multiple files.
5. Stay logged in: You’ll no longer be logged out every 2 weeks!
6. Keyboard shortcuts: Work faster with shortcuts, like ⌘ (Ctrl) + Shift + ; to copy last code block. Try ⌘ (Ctrl) + / to see the complete list.
Why does it matter?
These improvements make ChatGPT more user-friendly and streamline human-AI interactions, making it a more user-friendly and powerful tool overall. It will set the stage for improved and advanced AI applications as ChatGPT is today's leading LLM.
(Source)
Latest versions of Vicuna, based on the open LLaMA-2
The latest Vicuna v1.5 series based on Llama 2 features 4K and 16K context lengths (has extended context length via positional interpolation by Meta), and have improved performance on almost all benchmarks. Vicuna 1.5 tl;dr
7B & 13B parameter versions
4096 and 16384 token context window
trained on 125k ShareGPT conversations
Commercial use
Evaluated with standard benchmarks, human preference, and LLM-as-a-judge
Why does this matter?
Since its release, Vicuna has been one of the most popular chat LLMs. It has enabled pioneering research on multi-modality, AI safety, and evaluation. Since the latest versions are based on the open-source Llama-2, they can be an open LLM alternative to ChatGPT/GPT-4.
Knowledge Nugget: Top Challenges Large Language Models Need to Address, along with Possible Solutions.
While LLMs have shown great promise for applications solving real-world problems, they also face several challenges that must be addressed to realize their full potential beyond the prototype stage.
In this thought-provoking article,
delves into 8 major challenges large language models face, and discusses possible solutions for each. It addresses challenges such as over-reliance on prompts and fine-tuning, limited multilingual and cross-cultural capabilities, data privacy and security concerns, economic impact and the digital divide, bias and ethics issues, and more.Why does this matter?
The article provides practical solutions to important challenges on the path toward broader LLM adoption. It can also pave a path toward more effective and inclusive AI applications.
What Else Is Happening❗
🔥‘Every single’ Amazon team is working on generative AI, says CEO (Link)
🚀Twilio’s new integration will bring OpenAI’s GPT-4 model to its Engage platform (Link)
🤖Datadog launches generative AI assistant Bits and new model monitoring solution (Link)
💡Pinterest is now using next-gen AI for more relevant and personalized content and ads (Link)
🔗AI.com now redirects to Elon Musk's X.ai instead of taking you to ChatGPT (Link)
🌟📝Friday Featured Prompt
This Week's Prompt: Act as a Startup Idea Generator
Generate digital startup ideas based on the wish of the people. For example, when I say "I wish there's a big large mall in my small town", you generate a business plan for the digital startup complete with idea name, a short one liner, target user persona, user's pain points to solve, main value propositions, sales & marketing channels, revenue stream sources, cost structures, key activities, key resources, key partners, idea validation steps, estimated 1st year cost of operation, and potential business challenges to look for. Write the result in a markdown table.Is there a problem you've always wished to solve while finding a way to turn it into a profitable venture? Explore the possibilities with ChatGPT or simply have some productive fun.
Let your ideas take flight with this prompt!
🛠️ Trending Tools
AIO: Design your own clothing using our AI, connect with manufacturers, and resell your designs.
Alpaca: Personalized AI toolkit for artists in Photoshop. Render sketches into stunning images with AI seamlessly.
QR Fiddle: Generate QR codes using just a prompt or seed image.
Bau AI Interior Designer: Upload a photo, select a style, and watch AI transform your space. Shop recommended home products.
SceneProv: Co-create a screenplay scene with our AI-driven improve game. Fun, engaging, and educational!
Toma: AI-powered Diet tool to track symptoms and pinpoint trigger foods. Trusted by patients and nutritionists.
FitMate AI: Achieve your fitness goals with personalized workouts, customization, and privacy.
Writer.md: Craft SEO-optimized blog post drafts effortlessly with our AI-powered tool.
That's all for now!
If you are new to ‘The AI Edge’ newsletter. Subscribe to receive the ‘Ultimate AI tools and ChatGPT Prompt guide’ specifically designed for Engineering Leaders and AI Enthusiasts.
Thanks for reading, and see you Monday. 😊