Job Description
Analyse and evaluate strategies for fine tuning available AI generative models (eg LLM’s) to use within morphis-tech's K.Explorer tool, in order to incorporate domain specific knowledge.
DESCRIPTION
K.Explorer is a code assistant tool developed by Morphis to assist in code completion and generation using some recent generative models (eg GPT like models) and ontology/knowledge graphs (https://k-explorer.com/)
Recent research proved that efficient models can be built by fine-tuning existing models with specific domain data, without requiring a full training process.
The goal of this work is to explore the usage and existing strategies to fine-tune current open-source LLM’s (or similar, in the context of the K.Explorer generative goal). As an example, the process used to create the Alpaca model from the Llama LLM (https://github.com/tatsu-lab/stanford_alpaca)
One of the focuses is the construction of training datasets, either from existing data or from generated data, to be applied in that process, regarding some specific domains like
· Test case generation
· Quality/Security and vulnerabilities
· Architectural constructions
Other domains may be meanwhile identified to be relevant.
The work will also include the evaluation of each model with the relevant metrics, if possible.
REQUIREMENTS
Java
Python + TensorFlow or PyTorch
NLP
AI models knowledge
COORDINATION
K.Explorer Product Owner
Expected Duration: 8 weeks
Location
Portugal
Conditions / Salary Information
Internship grant including transport allowance
IMPORTANT NOTICE:
Hybrid – mostly remote
There may be a need for some interaction to start, especially in Onboarding, but the rest of the time will be remote.
Qualifications
Course frequency
Desired Degree
Electrical and Computer Engineering, Information and Enterprise Systems, Information Systems and Computer Engineering, Mathematics and Applications, Telecommunications and Informatics Engineering
Desired Skills
Java
Python
NLP
Neural networks
AI model training