Overview

We are excited to present Best Practices in AI Afternoon which will be held on the 5th of July, 12-5pm at Workroom 2, 38 Mappin, Sheffield, S1 4DT and online. The afternoon will consist of talks and walkthroughs on best practices for research, design, development, and deployment of AI systems with guest speakers from Nvidia, University of Cambridge and University of Bristol. A focus on practical aspects such as tooling, optimisation, profiling, tips and tricks to supercharge AI in your research!

Buffet lunch and coffee will be provided.

This event is held in collaboration between the Research Software Engineering (RSE) group and the Centre for Machine Intelligence (CMI) in the University of Sheffield.

Agenda

11:00-12:00  1:1 meetings with Nvidia

Book a 1 to 1 meeting with Nvidia experts on the day! For details, see below.

12:00-13:00  Networking Lunch

13:00-13:05  Welcome & Housekeeping

13:05-14:00  Maximizing Efficiency in Large Language Models: Compute, Memory, and Fine-Tuning

Karin Sevegnani, Senior Solutions Architect, Nvidia

In this talk, we will explore the intricate balance between computational resources, memory limitations, and parameter-efficient fine-tuning techniques in large language models (LLMs). We will analyse strategies to optimize the performance of LLMs while managing these constraints effectively. From efficient memory utilization to streamlined parameter fine-tuning methods, we will discuss practical approaches to maximize the efficiency of LLMs without sacrificing performance.

14:00-14:30  Docker for Machine Learning

Ryan Daniels, University of Cambridge

Writing research software in Python presents numerous challenges to reproducibility - what version of Python is being used? What about the versions of PyTorch, Scikit Learn or Numpy? Should we use Conda, or venv, or Poetry to manage dependencies and environments? How can we control randomness? Do I have the right version of Cuda Toolkit? In principle, given the same data, and same algorithms and methodology, we should be able to reproduce the results of any given experiment to within an acceptable degree of error. Dealing with the above questions introduces significant problems to reproducing experiments in machine learning. In this talk, I would like to convince you that Docker can help alleviate almost all of these questions. Furthermore, combining Docker, git and GitHub can be a powerful workflow, helping to minimise your tech stack, and declutter your python development experience.

14:30-15:00  How do you unit test an ML model?

Wahab Kawafi, University of Bristol

Covering methods such as mock testing, simulation, experiment tracking, and dataset curation. With examples in medicine, chemistry, aerospace engineering, and LLMs.

15:00-15:15  Coffee break

15:15-15:20  Lightning talk 1: Research Software Engineering

TBC

15:20-15:25  Lightning talk 2: AMRC Standardised Data-Centric Manufacturing Workflow

Lindsay Lee, AMRC, University of Sheffield

We have created a standardised workflow for data-driven projects at the AMRC including a Github site for consistent documentation, data management and presentation and a wiki with supporting documentation. These are based on the Microsoft Team Data Science Process and CookieCutter as well as our own experiences and existing processes.

15:25-15:30  Lightning talk 3: TBC

TBC

15:30-16:00  How to make your machine learning code faster

Edwin Brown, Research Software Engineering, University of Sheffield

Practical guide to profile machine learning code to find bottlenecks and to remove these bottlenecks.

16:00-16:30  TBC

TBC

16:30-16:55  Q&A Panel

16:55-17:00  Wrap up
 

Speaker Profiles

Karin Sevegnani, Senior Solutions Architect, Nvidia

Karin is a Senior Solutions Architect at NVIDIA, with a specific focus on the Higher Education and Research (HER) industry in the UK. At NVIDIA, she’s leading collaborative efforts with the NVAITC initiative, particularly centred around Isambard-AI. Prior to joining NVIDIA, Karin was a research engineer at Alana AI, where she specialized in Large Language Models (LLMs) and Retrieval Augmented Generation (RAG) systems. Her primary research interests include exploring topic transitions within conversational systems and devising effective strategies for recommendation settings. Karin holds a Ph.D. in Natural Language Processing from Heriot-Watt University, UK, which she completed in July 2023. During her doctoral studies, she focused on advancing conversational AI and recommendation systems. Notably, she contributed to Heriot-Watt’s participation in the Amazon Alexa Prize 2018 competition. Additionally, Karin gained valuable industry experience through an internship at Amazon in 2021, where she worked as an applied scientist on recommendation algorithms.

Ryan Daniels, University of Cambridge

Ryan is a machine learning engineer at the Accelerate Programme for Scientific Discovery and is interested in driving forward scientific research which is grounded in excellent software engineering and machine learning fundamentals. Before working at Accelerate, Ryan’s research interests explored unconventional approaches to computing using complex physical devices from the world of condensed matter physics.

More profiles coming soon…

Book a 1 to 1 with Nvidia

Book a 1 to 1 meeting with Nvidia experts on the day! They are open to discuss your research project at any stage whether you are already familiar and use accelerated computing looking to optimise or scale your project or whether you are relatively new and want to explore how to use AI/HPC in your project. Example discussion topics include:

• Possible approaches and techniques for your research problem/domain
• Available training courses
• Which GPU to use for your workload
• Generative AI and LLM related topics

Nvidia experts on the day:

• Andy Grant, EMEA Director for Supercomputing and AI at Nvidia
• Denis Battistella, Higher Education and Research at Nvidia
• Karin Sevegnani, Senior Solutions Architect at Nvidia

In-person meetings will take place on the day and slots are limited and are first come, first served. Remote meetings (may not happen on the day) will be offered once we’ve run out of meeting slots.

Registration

We hope to see you on the day whether in-person or online! Please use the buttons below to register for the event, present a talk or book a 1 to 1 meeting with Nvidia.

Register for the event

Speak at the event1:1 with Nvidia
Find out more on the event webpage here.