Phd Position F/m Llm4code and Soprasteria: Software

Le descriptif de l’offre ci-dessous est en Anglais_

**Type de contrat **:CDD

**Niveau de diplôme exigé **:Bac + 5 ou équivalent

**Fonction **:Doctorant

**A propos du centre ou de la direction fonctionnelle**:
The Inria Centre at Rennes University is one of Inria's eight centres and has more than thirty research teams. The Inria centre is a major and recognized player in the field of digital sciences. It is at the heart of a rich R&D and innovation ecosystem: highly innovative PMEs, large industrial groups competitiveness clusters, research and higher education players, laboratories of excellence, technological research institute, etc

**Contexte et atouts du poste**:
Generative AI, in particular the recent Large Language Models (LLMs), show great promise for software developments. Specialized models are now able to perform an impressive variety of programming tasks: solving programming exercises, assisting software developers, or even generating mechanized proofs. Yet, many challenges still need to be addressed to build reliable and productive LLM-based coding assistants: improving the quality of the generated code, increasing the developers’ confidence in the generated code, enabling interaction with other software development tools (verification, test), and providing new capabilities (automated migration and evolution of software).

The goal of the Défi Inria LLM4Code is to leverage LLM capabilities to build code assistants that can enhance both reliability and productivity. The défi is organized along three work packages: Self-improving code generation, Evolution of existing software (WP2), Interactive tools with AI-in-the-loop.

**Mission confiée**:
**Principales activités**:
A vast portion of the software used nowadays in the critical sectors of the industry is
written in legacy languages (e.g. Fortran, COBOL, Ada, etc.) that are prone to be outdated.
These languages do not profit from modern software engineering tools, do not adhere to the
latest standards of quality or security, and are famous for blocking developers in their
everyday work. However, there is no standard solution to migrate an existing code base to
newer technologies that would be stable, secure, and affordable from the time/value
perspective.
We propose to leverage LLMs’ capabilities for software migration. While LLMs excel at
translation tasks for natural language, programming language migration is still challeng
- ing
[Zhu et al., 2022, Pan et al., 2023, Yan et al., 2023]. Incorporating fine-grained examples into
the training of LLMs is essential to capture the nuances of different programming paradigms
and semantics. These examples provide detailed, context-rich scenarios that help LLMs
understand and adapt to various programming structures and logic. Furthermore, leveraging
compilers or transpilers to generate synthetic data can be effective in creating
a diverse
training dataset. On the other hand, LLMs can enhance existing compilers or migration
tools by broadening their scope to cover more diverse and complex corner cases. This results
in tools that are not only more robust, but also capable of addressing a wider range of
migration scenarios. Besides, LLMs are not solely beneficial for translating programs; they
also play a crucial role in comprehending existing codebases, documenting system
architectures, or synthesizing test cases to validate the migration. Both activities are essential
for software migration, and our strategy includes utilizing LLMs to efficiently address these
specific tasks.
We will first experiment on the open challenge of converting Fortran-77 to C. From
many perspectives, the gap between Fortran-77 (as the most spread version of Fortran)
and C is significant. Furthermore, the lack of a reference dataset matching Fortran-77 to
C code, and the validation of the results generated by the LLM raise multiple challenges.
In addition to the challenging case of converting Fortran-77 to C, we aim to explore the
problem of migrating old codebases written in programming languages such as 4GL or old
Java [Fleurey et al., 2007, Verhaeghe et al., 2019]. Software migration involves resolving
many tasks and related problems: reverse engineering (e.g., understanding the existing
codebase and functionality, documenting the system’s architecture), translating code to
a modern platform or programming language, testing (from unit to user acceptance) to
ensure the new migrated system fits original requirements, etc. For each task, LLMs can
be of interest [Xie et al., 2023, Fan et al., 2023, Hou et al., 2023]

Angela Fan, Beliz Gokkaya, Mark Harman, Mitya Lyubarskiy, Shubho Sengupta, Shin Yoo,
and Jie M Zhang. Large language models for software engineering: Survey and open
problems. arXiv preprint arXiv:2310.03533, 2023.

Franck Fleurey, Erwan Breton, Benoit Baudry, Alain Nicolas, and Jean-Marc Jézéquel.
Model-driven engineering for software migration in a large industrial context. In Model
Driven Enginee