Maxime Labonne

Head of Design

London, England, United Kingdom6 yrs 4 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Created popular LLMs on Hugging Face.
Developed award-winning LLM course on GitHub.
Authored technical books on LLMs and Graph Neural Networks.

Stackforce AI infers this person is a Machine Learning Expert in Fintech and Cybersecurity.

Contact

Skills

Core Skills

Large Language Models (llm)Machine LearningDeep LearningTransformerTime Series Analysis

Other Skills

PythonData SciencePyTorchAlgorithmsData StructuresLinear ProgrammingMachine Learning AlgorithmsArtificial Intelligence (AI)Geometric Deep LearningSystems DesignDistributed SystemsDatabasesKubernetesComputer NetworkingNetwork Security

About

𝕏 𝗫: @maximelabonne • 🤗 𝗛𝘂𝗴𝗴𝗶𝗻𝗴 𝗙𝗮𝗰𝗲: huggingface.co/mlabonne/ • 📝 𝗕𝗹𝗼𝗴: mlabonne.github.io/blog/ • 💻 𝗚𝗶𝘁𝗛𝘂𝗯: github.com/mlabonne/ I'm a Machine Learning Scientist with a PhD from the Polytechnic Institute of Paris. I started working with Large Language Models and Graph Neural Networks in 2019 and applied them in diverse contexts, including R&D, industry, finance, and academia. I'm also an AI/ML Google Developer Expert. I created numerous popular LLMs on Hugging Face, such as AlpahMonarch-7B, Beyonder-4x7B, Phixtral, and NeuralBeagle14. I also released LLM tools, like LLM AutoEval, LazyMergekit, LazyAxolotl, and AutoGGUF. I made the popular LLM course on GitHub (>39k stars) and I write technical articles on my blog and Towards Data Science. I'm the author of the technical books "LLM Engineer's Handbook" and "Hands-On Graph Neural Networks using Python". Opinions are my own and not the views of my employer.

Experience

Liquid ai

Head of Post-Training

Mar 2024 – Present · 2 yrs · London Area, United Kingdom · Hybrid

Jpmorgan chase & co.

Lead Machine Learning Scientist

Nov 2022 – Mar 2024 · 1 yr 4 mos · London, England, United Kingdom · Hybrid

Developed a Copilot-like model to autocomplete code based on the firm's internal codebase, outperforming vendor solutions.
IndexGPT: fine-tuned and developed LLM applications for domain- and task-specific use cases.
Spam-T5: developed an LLM-based solution to detect spam emails (published at IJCAI 2023). First open-sourced AI project at JPMorgan.

Large Language Models (LLM)Machine LearningPythonData ScienceDeep Learning

Airbus

Machine Learning Scientist

Feb 2020 – Nov 2022 · 2 yrs 9 mos · Issy-les-Moulineaux, Île-de-France, France

CyBERT: Designed domain-specific LLMs (based on RoBERTa and GPT-2) for network protocol understanding with various applications such as intrusion detection, protocol identification, and packet generation.
Developed a new end-to-end deep generative model for network data in PyTorch, surpassing the state-of-the-art solutions in ten domain knowledge tests.
Developed a heterogeneous time series forecasting (DT, RF, XGBoost) solution with a new tabular embedding, improving the MSE by 19.73%.
Developed a mathematical optimization architecture to solve NP-hard graph problems using Integer Linear Programming (Gurobi) and a custom evolutionary algorithm in C++ (95% optimal with faster inference time).

Deep LearningTransformerTime Series Analysis

Commissariat a l'energie atomique et aux energies alternatives

PHD Student

Jan 2017 – Jan 2020 · 3 yrs · Palaiseau, Île-de-France, France

Designed a patented framework for performing end-to-end protocol-based unsupervised anomaly detection.
Developed a software solution, SIGMO-IDS, based on this framework in Python and TensorFlow that has been utilized in various industrial (STM32WB, Legrand) and European projects (H2020 SCENE, Critical-Chains, etc.).
Created a hierarchical Mixture of Experts architecture for intrusion detection with automated data augmentation and Bayesian optimization.
Designed an RNN-based architecture to detect network congestion before it occurs, trained on millions of synthetic samples generated from a testbed network.

Deep LearningTime Series Analysis

Dga - direction générale de l'armement

Engineering Internship

Feb 2016 – Aug 2016 · 6 mos · Arcueil, France

Intern at the French MoD Battle Lab (LTO) in modeling and simulation.
Designed and implemented a framework for verification and validation of technical-operational simulation models based on a trusted model (instead of real systems or expert results).
Developed plugins in C++ for telecommunications in a military simulation engine in a production environment.