A few Baseten engineers teamed up and released ChatLLaMA, an open-source ChatGPT alternative, as a weekend project. ChatLLaMA uses Alpaca-7B, a fine-tuned variant of LLaMA designed for a chat interface.
Fine-tune FLAN-T5 on Blueprint
Want to fine-tune your own LLM? You can on Blueprint, which now supports fine-tuning FLAN-T5, an instruction-tuned text-to-text-transfer-transformer model developed by Google.
Why fine-tune an LLM?
- Have a top-notch LLM customized with your data — train it on your customer support tickets, user-generated content, or documentation
- Build a special-purpose generative app like a recipe generator or a chatbot that’s an expert on healthcare acronyms
- Save money versus using generic LLM APIs and own your model
We have a step-by-step tutorial for fine-tuning FLAN-T5 as well as docs on creating an effective dataset and kicking off your fine-tuning run to get your project rolling.
We’re here to help or just cheer you along on Twitter and Discord. Get started with four hours of free credits for new accounts here.
AI projects for the hacker spirit
After releasing Blueprint last month, Baseten’s engineers have been building side projects using LLMs and generative AI. Here are a few projects from the team:
- Suren built GPTheus, which gives a voice and visual interface to ChatGPT using Whisper and Stable Diffusion
- Sid’s hackathon group created an AI-powered dev infrastructure tool that writes dockerfiles and uses them to build codespaces
- Philip built a quick poetry generator
- Aaron, Abu, and Suren’s ChatLLaMA deserves a second shout-out for hitting #1 on Hacker News
Here’s a demo and walkthrough of Suren’s project:
Compliance: everyone’s favorite adventure
In the last year, ML has gone from an experiment to a core need for many companies who require elevated levels of security and compliance. Baseten is now SOC 2 Type II certified and HIPAA compliant. We are honored to be entrusted with sensitive data and take our platform’s security and stability seriously.
Before we go, here is a photo from our AI meetup at our San Francisco office. We’re hosting another in April, follow us on Twitter to find out more!
See you next month!
— The team at Baseten
New in May
Open-source models continue to bridge the gap on results quality versus their closed-source counterparts. Discover new models for text generation and text-to-speech, learn more about the GPUs they run on, and plug in to the community forming around open-source models in this newsletter.
Understanding NVIDIA’s Datacenter GPU line
NVIDIA has dozens of GPUs that can serve ML models of different sizes. But understanding the performance and cost of these different cards, not to mention just keeping the names straight, is a challenge. This guide helps you navigate NVIDIA’s datacenter GPU lineup.