PhD Position F/M Experimentation with LLMs for Fortran migration

Download job offer in PDF format

Contract type : Fixed-term contract

Level of qualifications required : Graduate degree or equivalent

Fonction : PhD Position

Level of experience : Up to 3 years

About the research centre or Inria department

The  Inria University of Lille centre, created in 2008, employs 360 people  including 305 scientists in 15 research teams. Recognised for its strong  involvement in the socio-economic development of the Hauts-De-France  region, the Inria University of Lille centre pursues a close  relationship with large companies and SMEs. By promoting synergies  between researchers and industrialists, Inria participates in the  transfer of skills and expertise in digital technologies and provides  access to the best European and international research for the benefit  of innovation and companies, particularly in the region.For more  than 10 years, the Inria University of Lille centre has been located at  the heart of Lille's university and scientific ecosystem, as well as at  the heart of Frenchtech, with a technology showroom based on Avenue de  Bretagne in Lille, on the EuraTechnologies site of economic excellence  dedicated to information and communication technologies (ICT)

Context

This PhD will happen in the context of the Inria LLM4Code défi. LLM4Code is an ambitious project incorporating several INRIA groups and external partners for building reliable and productive solutions based on Large Language Models.

Assignment

During the project the phd student will focus on assessing the possibility of performing a software migration with LLMs in the specific context of a given niche technology for a given organization (specific domain, specific development culture).

Context

We are engaged with an industrial partner on a code transformation project that aims to migrate a Fortran-77 + proprietary extension code base into modern Fortran code. The project uses a model driven approach where the existing code is modeled, this model is "refactored" and then regenerated in modern Fortran.

Challenges

The performance of LLMs is correlated with their training data quality. The majority of the training dataset comes from publicly available software artifacts, and often these data can be of questionable quality, riddled with vulnerabilities, biased and produce varying outputs for identical prompts.

Generic LLMs are trained from millions of “documents”. For software engineering and code generation, specialized LLMs (like HuggingFace or Llama) have been trained, but they are bound to contain less Fortran examples as less Fortran project are available in common open-source repositories (like github).

The project will need to evaluate how such imperfect LLMs can be used for migration, what are the consequences on the quality of the result and what techniques (if any) can be used to improve these results.

Outcome

The project will propose a methodology to realize code migration of a niche technology for a specific organization using LLMs.

More importantly, it will identify the key points required in such a project and the advantages and drawback of such a project as compared for example to a deterministic model based approach?

Bibliography

  • Frank F. Xu, Uri Alon, Graham Neubig, and Vincent Josua Hellendoorn. 2022. “A systematic evaluation of large language models of code”. In Proceedings of the 6th ACM SIGPLAN International Symposium on Machine Programming (MAPS 2022). Association for Computing Machinery, New York, NY, USA, 1–10. https://doi.org/10.1145/3520312.3534862

  • Mahmood, Hina & Jilani, Atif & Rauf, Abdul. (2023). “Code Swarm: A Code Generation Tool Based on the Automatic Derivation of Transformation Rule Set”. International Journal of Software Engineering & Applications. 14. 1-11.

  • Gustavo Pinto, Cleidson de Souza, João Batista Neto, Alberto de Souza, Tarcísio Gotto, and Edward Monteiro, “Lessons from Building CodeBuddy: A Contextualized AI Coding Assistant”, arXiv e-prints, 2023. doi:10.48550/arXiv.2311.18450.

 

Main activities

Responsibilities:

  • Analysis and reverse engineering of existing codebases (leveraged by Software Heritage archive)
  • Applying LLM for analysis of existing code, tests and migration results
  • Contributing to summarization and dissemination of results, writing scientific articles.

Skills

  • Good foundation in Machine Learning and Software Engineering.
  • Proficiency in OOP is required (knowing of Pharo programming language is a plus)
  • Excellent problem-solving abilities and a strong interest in research.
  • Ability to work independently and collaboratively in a dynamic team.
  • Good communication skills (English required, French is a plus)

Benefits package

  • Subsidized meals
  • Partial reimbursement of public transport costs
  • Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
  • Possibility of teleworking and flexible organization of working hours
  • Professional equipment available (videoconferencing, loan of computer equipment, etc.)
  • Social, cultural and sports events and activities
  • Access to vocational training
  • Social security coverage

Remuneration

2100€ gross per month for the 1st and 2nd years

2190€ gross per month for the 3rd year