Présentation de la société : CEA
Le CEA est un acteur majeur de la recherche, au service des citoyens, de l'économie et de l'Etat.Il apporte des solutions concrètes à leurs besoins dans quatre domaines principaux : transition énergétique, transition numérique, technologies pour la médecine du futur, défense et sécurité sur un socle de recherche fondamentale. Le CEA s'engage depuis plus de 75 ans au service de la souveraineté scientifique, technologique et industrielle de la France et de l'Europe pour un présent et un avenir mieux maîtrisés et plus sûrs.
Implanté au cœur des territoires équipés de très grandes infrastructures de recherche, le CEA dispose d'un large éventail de partenaires académiques et industriels en France, en Europe et à l'international.
Les 20 000 collaboratrices et collaborateurs du CEA partagent trois valeurs fondamentales :
• La conscience des responsabilités
• La coopération
• La curiosité
Missions
The internship aims to enhance the performance of large language models (LLMs) in generating Instruction Set Simulator (ISS) code by using Reinforcement Learning (RL) to optimize and automate prompt tuning. Additionally, the internship seeks to expand dataset coverage by using simulators such as QEMU (Quick Emulator) in-the-loop to simulate and evaluate a wider range of architectures, enabling access to diverse implementations and increasing dataset diversity. While RL will be the primary focus, alternative methods can also be explored throughout the internship.
The main activities will involve using RL to dynamically adjust the prompts fed to the LLM, guiding it to improve code correctness, compilation success, and functional efficiency. Open questions for investigation include, but are not limited to, how to define rewards that balance compilability and functionality, how feedback from these rewards can be used to refine future prompts, and what strategies can effectively integrate RL rewards into the prompt generation process. The RL agent will iteratively adjust the prompt based on feedback from compilation and functional tests, using QEMU to assess the quality of the generated code. By simulating multiple architectures in the QEMU environment, the internship will aim to broaden the dataset coverage, making the model more adaptable to different hardware implementations. The results of this work have the potential to contribute significant insights into this field and may lead to publication in relevant conferences.
During this internship, the student will gain practical experience with advanced AI techniques, such as RL and automatic prompt tuning, while enhancing their knowledge of LLM-based code generation. This project provides a valuable opportunity to develop key skills in AI-driven hardware design and to contribute to innovative research.
Profil recherché
Level required: Master's degree
Duration: 6 months
Skills: Understanding of AI, especially LLM and RL, and knowledge of HW architecture design would be a plus.
Python, C/C++
English, teamwork, curiosity
~1400 monthly "salary", depending on whether the intern has a scholarship or not.
In line with CEA's commitment to integrating people with disabilities, this job is open to all.