Lamini LLM Finetuning on AMD ROCm: A Technical Recipe

lamini.ai

6 pointsdhruvdh4 comments

"we set up a cost-effective server room by adapting a janitor's closet with ventilation and sound insulation"

That's the spirit! Way to go ~

Happy to answer any questions about this.

> When a new request comes in, a few pointers in HBM are updated in approximately 2.93us, making model switching nearly instant. This enables one server to support more than 10,000 different models.

I am a PhD student wanting to work with systems made of multiple LLM powered agents. This makes me think work like "Generative Agents: Interactive Simulacra of Human Behavior" (https://arxiv.org/abs/2304.03442) can potentially be run on a single GPU?

I am interested in multi-agent systems that are hand designed to automate specific tasks, like qualitative data analysis. Would people at Lamini be interested in collaborating on such research?

Researchers are welcome to try our platform, and give us feedback

Send me a note on linkedin, and I can grant expanded functionality, eg access to more models.

We generally believe in the value of AI research and will be supportive where possible.

We will be able to do more to support research as we grow our compute resources.

Reply on news.ycombinator.com