October 2023 Summaries

This week, we overhauled the model management experience on Baseten, improving several core workflows to clarify the model lifecycle. We also shipped a new text embedding model that matches OpenAI's ada-002 in both context window size and benchmark performance. The new model management experience features improved deployability, scalability, and cost-effectiveness, including workspace API keys with more granular permissions and separate measurements for end-to-end response time and inference time. Additionally, the team at Baseten has refreshed their documentation with guides on various topics such as model inference, autoscaling, monitoring, and instance types. The company is also hosting a fireside panel on the state of open source ML in San Francisco and will be attending AWS re:Invent in NVIDIA's generative AI pavilion.

Jina AI has released `jina-embeddings-v2-base-en`, an open-source text embedding model that matches the performance of OpenAI's ada-002 model, in terms of both benchmark performance and context window length. This model can be used for various applications such as search, recommendation, clustering, and retrieval-augmented generation. Unlike other models, `jina-embeddings-v2` has a larger context window, allowing it to process longer texts into a single embedding. The model is currently only available for English text, but plans are in place to expand to other languages soon. To deploy the model, users can access it from Jina AI's model library or use Truss, with options to scale up to more powerful instances as needed.