Timothy R.’s Post

Moderated a session this morning on applied GenAI/LLMs for a broad group of folks, thanks to Hydrolix.io for sponsoring the event outside RSA. As many folks think about building out "chain of thought" RAG pipelines, I wanted to mention there was a cool effort by a group to expand the context window of LLAMA 3 from 8k to 1M tokens. https://lnkd.in/eK8U3Pms I started to test with a typical needle in a haystack test, but discovered we need a better way to test efficacy- that more accurately aligns with the way we retrieve data from files and content blocks provided to the model. I will share more as a solution is fully identified. BUT, this Gradient model is a great potential addition to those using GenAI Assistant and Agents with local enterprise data, such as large log datasets housed in systems such as Hydrolix.io.

llama3-gradient

llama3-gradient

ollama.com

To view or add a comment, sign in

Explore topics