STAGE - Benchmark Small Language Models (SLMs) in semiconductor tool calling context F/H

Beschreibung:

At STMicroelectronics, we believe in the power of technology to drive innovation and make a positive impact on people, businesses, and society. As a global semiconductor company, our advanced technologies and chips form the hidden foundation of the world we live in today.
When you join ST, you will be part of a global business with more than 115 nationalities, present in 40 countries, and comprising over 50,000 diverse and dedicated creators and makers of technology around the world.
Developing technologies takes more than talent: it takes amazing people who understand collaboration and respect. People with passion and the desire to disrupt the status quo, drive innovation, and unlock their own potential.
Embark on a journey with us, where you can innovate for a future that we want to make smarter and greener, in a responsible and sustainable way. Our technology starts with you.
A propos de vos missions
Vous aiderez à sélectionner et tester des Small Language Models (SLMs) performants sur CPU et à les connecter avec des outils locaux via le Model Context Protocol (MCP). Vous mettrez également en place des workflows simples pour gérer les données, suivre les expériences et évaluer ces PML.
* Sélectionner et évaluer des Petits Modèles de Langage adaptés aux CPU afin de tester leurs performances prêtes à l'emploi.
* Intégrer les SLMs avec les serveurs MCP fournis ; créer éventuellement des serveurs MCP simples pour les outils manquants.
* Collecter et organiser un jeu de données standardisé d'instructions associées à des exemples d'utilisation d'outils/API issus de la documentation existante.
* Créer un dispositif de test pour comparer les modèles de base avec leurs versions affinées en utilisant le même jeu de données organisé.
* Mettre en place un workflow LLMOps léger incluant :
* La préparation d'un jeu de données unique et standardisé d'instructions et d'utilisation d'outils/API
* Le suivi des expériences
* La gestion des versions des modèles et des prompts
* La génération automatique de rapports d'évaluation montrant les différences entre modèles de base et modèles affinés
* Réaliser des expériences de fine-tuning sur Azure ML en utilisant des frameworks comme Unsloth, Axolotl ou similaires.
* Créer un tableau comparatif illustrant la qualité des modèles, leur vitesse, leur consommation mémoire et les améliorations apportées par le fine-tuning, et fournir des recommandations sur les modèles à utiliser.
A propos de vous
* Étudiant(e) en dernière année de Master ou d'ingénierie en Informatique, IA, Data Science ou Systèmes Embarqués
* Solides compétences en programmation Python
* Connaissance des environnements d'exécution et formats de modèles tels que GGUF, llama.cpp, Ollama, etc.
* Expérience dans la création et la gestion de datasets pour le fine-tuning de modèles de langage
* Notions de base ou expérience avec Azure ML ou autres environnements cloud ML
* Compréhension du MCP (Model Context Protocol) et des appels d'outils/fonctions.
* Connaissances de base en quantification et notions de concepts comme LoRA/QLoRA
* Familiarité avec Git, CLI, Docker
* Compétences en communication technique en anglais : B2 minimum.

YOUR ROLE
You will help select and test Small Language Models (SLMs) that work well on CPUs and connect them with local tools through the Model Context Protocol (MCP). You will also set up simple workflows to manage data, track experiments, and evaluate these SLMs.
* Select and benchmark CPU friendly Small Language Models (SLMs) to test their out-of-the-box performance.
* Integrate SLMs with provided MCP servers; optionally create simple MCP servers for any missing tools.
* Collect and curate a standardized dataset of instructions paired with tool/API usage examples from existing documentation.
* Create a test setup to compare base models with fine-tuned versions using the same curated dataset.
* Set up a lightweight LLMOps workflow that includes:
* Preparing a single, standardized dataset of instructions and tool/API usage
* Tracking experiments
* Managing versions of models and prompts
* Automatically generating evaluation reports showing differences between base and fine-tuned models
* Run fine-tuning experiments on Azure ML using frameworks like Unsloth, Axolotl, or similar.
* Create a comparison chart that shows model quality, speed, memory use, and fine-tuning improvements, and give advice on which models to use.
YOUR SKILLS & EXPERIENCES
* Master or final-year engineering student in Computer Science, AI, Data Science, or Embedded Systems
* Strong Python programming skills
* Familiar with model runtimes and formats such as GGUF, llama.cpp, Ollama, etc.
* Experience creating and curating datasets for fine-tuning language models
* Basic or notions of Azure ML or other cloud ML environment use
* Understanding of MCP (Model Context Protocol) and tool/function calling
* Basic quantization and notion of concepts like LoRA/QLoRA
* Familiar with Git, CLI, Docker
* English technical communication skills (B2).
ST is proud to be one of the 17 companies certified as a 2025 Global Top Employer and the first and only semiconductor company to achieve this distinction. ST was recognized in this ranking thanks to its continuous improvement approach and stands out particularly in the areas of ethics & integrity, purpose & values, organization & change, business strategy, and performance.
At ST, we endeavor to foster a diverse and inclusive workplace, and we do not tolerate discrimination. We aim to recruit and retain a diverse workforce that reflects the societies around us. We strive for equity in career development, career opportunities, and equal remuneration. We encourage candidates who may not meet every single requirement to apply, as we appreciate diverse perspectives and provide opportunities for growth and learning. Diversity, equity, and inclusion (DEI) is woven into our company culture.

To discover more, visit st.com/careers