New in May 2023
The latest foundation models in two clicks
Our refreshed and restocked model library features familiar favorites like Stable Diffusion 2.1, Whisper, and Alpaca, all available to deploy directly to your Baseten account. Deployed library models run on their own instance, giving you control over resource configuration and autoscaling and visibility into logs and metrics.
Exciting new open-source ML models are released every week. Here are three that you should know about:
WizardLM: an open-source ChatGPT
WizardLM is a LLM that has been fine-tuned to behave like ChatGPT. It’s a great option for projects looking for similar behavior, quality, and results but from an open-source model.
Bark: generative AI for audio
Bark is a speech-generation model that creates audio based on text prompts. It’s a remarkably accurate and natural-sounding text-to-speech model.
StableLM: an LLM by the creators of Stable Diffusion
StableLM is a project to train and fine-tune large language models for a variety of powerful and configurable behaviors. StableLM uses some of the largest and most advanced open-source training datasets available.
Everything you need to know about GPUs
NVIDIA’s A10 GPU is a workhorse for model serving.
But, pop quiz: what does “A10” even mean?
Well, the “A” refers to the GPU’s microarchitecture, Ampere, while the 10 identifies the card’s tier. The name helps us understand the card relative to other cards: it’s smaller than an A100 but uses the same architecture, and it’s newer than the T4, another popular and capable graphics card for model inference.
If you want to learn more about the fascinating hardware powering the generative AI boom, we’ve been writing about GPUs on our blog:
ML momentum continues from coast to coast
We had another great meetup in our San Francisco office, this time featuring demos from talented builders across AI. Please join us in NYC for a June 15 meetup, details will be shared soon on Twitter!
The Baseten team got together this month in person for a hackathon. Shoutout to Samiksha for her winning project: “I stream for ice cream,” which prototyped streaming responses for language models with Truss! Many of the hackathon projects are—after some refactoring—making their way into the Baseten platform as exciting new features that we look forward to sharing soon.
See you next month!
— The team at Baseten
