We’re thrilled to announce our next MLOps Community Bristol Meetup on 6th of March 2025!

This event will be sponsor and hosted by Graphcore!

We have a stellar lineup of speakers for this event:

DeepSeek-v3 from first principles: experience from the frontier by Alexandre Payot, ML Engineer at Graphcore

Last month DeepSeek changed the perception of what it cost to train frontier AI models. In this talk, we will go from the basics of the transformer architecture and their execution on current hardware, to the details in the DeepSeek-v3 technical report to understand how their engineering team managed a final training run for less than $6M. Furthermore, we will compare the design choices of DeepSeek-v3 and Llama 3.1-405B and how that impacts your deployment and fine-tuning options.

(Lightning Talk) Driving LLM Efficiency Using Novel Low-Precision Data Formats by Alex Titterton, ML Engineer at Graphcore

In recent years AI models, in particular LLMs, have scaled up enormously both in terms of capability and hardware requirements. Providing the required computational power, storage capacity and memory bandwidth all come at a cost, leading to increased research activity into low-precision data formats both for storage and compute. In this talk we discuss recent advances in low-precision training and inference, quantisation methods and new microscaling (MX) data formats designed to offer efficient AI compute with minimal loss in accuracy and without requiring changes to model training workflows.

From Structure to Flow: Leveraging Graphs to Guide LLM-Powered Dialogue by Sneha Sen, ML Engineer at Cleo and Adnan Shahzada, Principal ML Engineer at Cleo

The challenge of balancing the structural precision of predefined conversation flows with the flexibility of fluid, natural interactions is central to conversational AI. This talk introduces an approach that merges graph-based chatflow design with the adaptive capabilities of Large Language Models (LLMs). Representing conversational flows as directed graphs creates a dynamic framework that ensures logical progression while allowing LLMs to handle deviations with contextual reasoning. This approach emphasizes the structural clarity of chatflows through nodes encapsulating messages and decision logic, forming a navigable map of the conversation. The graph-based representation serves as both a constraint and an enabler, guiding interactions while leveraging LLMs’ adaptability to user input, creating intelligent, scalable, and user-friendly dialogues.

Agenda:

18:00: Doors Open & Networking
18:30: Alexandre Payot, ML Engineer at Graphcore
19:10: Alex Titterton, ML Engineer at Graphcore
19:25: Sneha Sen, ML Engineer at Cleo and Adnan Shahzada, Principal ML Engineer at Cleo
20:00: More Networking, Drinks & Close

Join us for an evening of insightful talks, great conversations, and networking with fellow MLOps enthusiasts. Whether you're deep in the MLOps trenches or just starting your journey, there's something for everyone.

We look forward to seeing you there!

MLOps Bristol - Meetup 15