We are excited to unveil the latest update to our platform. This new release introduces fresh features, various improvements, and backend advancements. The following details outline the specific updates included in this release.
CoAuthor
CoAuthor has made its debut, offering a seamless and intuitive collaboration platform that utilizes AI and automation to expedite document production.
The user menu items have been updated, including a revised "About" dialog for the following AI applications:
Axon
Cortex
Curate
Layar
Signal
Backend Changes
Layar API
The GPU information for the Layar API has been updated.
New API endpoints have been introduced to enable the creation of embeddings directly, bypassing the need for a call to the Generate API.
A new pipeline has been introduced for pre-configured prompts.
A perplexity score has been added to the response generation.
Score contributions from chunks used in Generate response have been added
New endpoints have been added for asynchronous calls to Generative AI.
An update has been made to the Data Provider counts to accurately reflect the numbers from both remote and local data sources.
There has been an update to the email branding
Reference Data Fabric
The method for ingesting PubMed files has been updated to improve stability.
LLM
A fix was implemented for special Unicode characters.
Performance enhancements have been made to RAG querying, particularly in the area of table row searches.
Implemented a dedicated embedding endpoint for document sources.
The perplexity score, a metric for evaluating language models, has been incorporated
Support for Llama 3.1 70B models has been added.
Additional scores have been added to the response for both vector and keyword searches.
The default threshold for relevance has been lowered to 0.2.
The maximum prompt length has been updated to 350 for shorter context models and 500 for longer context models.