Sharon Zhou and Greg Diamos on Lamini’s pioneering work with AMD in Generative AI.
Subscribe: Apple • Spotify • Overcast • Google • AntennaPod • Podcast Addict • Amazon • RSS.
Sharon Zhou and Greg Diamos are co-founders of Lamini1, a startup at the forefront of enabling enterprise adoption of large language models (LLMs). We discussed Lamini’s work with AMD, which focused on closing the gap between AMD hardware capabilities and software integration in LLM applications.
Despite the perception that AMD’s software stack for machine learning was only 10% complete, Lamini recognized it was actually around 90% complete and set out to fill in the remaining 10%. This involved overcoming challenges in various components of the software stack, such as matrix operations and framework compatibility. Their efforts culminated in successfully running a full LLM application on AMD hardware, including tasks like pre-training, fine-tuning, and inference. This achievement not only demonstrated the viability of AMD’s software and hardware for machine learning applications but also contributed to reducing the dominance of other players in the field, ultimately making powerful computing more accessible and affordable for a broader range of users. Lamini’s approach can be likened to creating a CUDA-like layer for AMD, significantly enhancing AMD’s usability in the AI and machine learning domains.
Interview highlights – key sections from the video version:
- Lamini & AMD for enterprise LLMs
- Significance of Lamini+AMD collaboration
- Other hardware options: CPUs and LLMs
- Enterprise Adoption Blockers for LLMs
- Retrieval augmented generation in the the enterprise
- Enterprise use cases of LLMs
- Function Calling
- The state of open source LLMs
- Data engineering tools for LLMs
Related content:
- A video version of this conversation is available on our YouTube channel.
- Beyond Nvidia: Exploring New Horizons in LLM Inference
- Nir Shavit: LLMs on CPUs, Period
- Philipp Moritz and Goku Mohandas: Navigating the Nuances of Retrieval Augmented Generation
- Apple’s AI Leap: Bridging the Gap in On-Device Intelligence
- Daniel Lenton: Ivy – The One-Stop Interface for AI Model Deployment and Development
- Waleed Kadous: Best Practices for Building LLM-Backed Applications
If you enjoyed this episode, please support our work by encouraging your friends and colleagues to subscribe to our newsletter:
[1] Ben Lorica is an investor/advisor in Lamini and other startups.