In this video, we dive into the latest AI and machine learning updates from Databricks for September 2024! Here’s a breakdown of the key topics covered:
00:12 - Llama 3.2 Models available on Databricks
We kick off with an overview of the newly released Llama 3.2 models, including the 1B and 3B instruct models, perfect for low-cost use cases like entity extraction and translation. We also discuss the powerful multimodal 11B and 90B vision instruct models and learn how they handle text, images, audio, and video. Plus, find out how the 128K context length improves long-document processing.
02:46 - Hybrid Search in GA
Learn about the general availability of hybrid search in Mosaic AI, which combines keyword and vector search for more accurate retrieval of business-specific data.
04:57 - AI Gateway New Features
We explore the AI Gateway’s new features like usage tracking, payload logging, and guardrails to manage AI traffic across models like OpenAI and Anthropic.
07:15 - Improvements to Agent Evaluation LLM Judges
The built-in LLM judges can reason about the semantic correctness of a generated answer with respect to a reference answer. Databricks is continuously improving the effectiveness of these judges so that Agent Evaluation customers can handle more advanced use cases and be more productive in improving the quality of their applications. Databricks announced the launch of an improved answer-correctness judge in Agent Evaluation.
08:38 - Platform Release Notes
OpenAI compatible function calling is now available on Foundation Model APIs provisioned throughput. This launch also includes function calling support for the Llama 3.1 8B Instruct model on provisioned throughput workloads.
Use the AI Playground to prototype, build, and then export tool-calling GenAI agents. You can now give your AI agent tools in the form of Unity Catalog functions and interact with the agent directly in AI Playground. You can export the AI agent to notebooks to iterate further, evaluate quality, and deploy it.
Foundation Model APIs pay-per-token is now available in eu-central-1 and eu-west-1.
Foundation Model APIs provisioned throughput now supports GTE v1.5 (English) models: gte-base-en-v1.5 and gte-large-en-v1.5.
Meta Llama 3.1 70B and Meta Llama 3.1 8B models are now supported in Mosaic AI Model Training.
Meta Llama 3.1 405B Instruct now supported on Databricks Function Calling
The Public Preview for vector_search() is now available to all customers in regions where Mosaic AI Vector Search is supported. Customers no longer need to request access.
External models now supports instance profiles for accessing AWS resources when using models from Amazon Bedrock. See Configure the provider for an endpoint.
Don’t forget to check out the demos and deep dives mentioned throughout the video for hands-on guidance!
Негізгі бет September ML & GenAI Announcements
Пікірлер