How can we effectively fine-tune an OCI AI LLM for Code Generation?

AldoMontesDeOca-Oracle

We wish to use a model (LLM) capable of generating code in JSON format. To achieve this, we plan to train the model with a collection of scripts of the types of code we want to generate, which will serve as examples to fit specific scenarios defined by the user. The main issue lies in the training dataset for fine-tuning, as being limited to words restricts the model's ability to learn code. The ultimate goal of using OCI AI services is to automate 5G use cases or testing. Another idea we have is to obtain information from 3GPP, where all use cases or tests are documented by Network Function and Interface, to train an AI model that can generate Ansible code or code in another programming language to model these scenarios as automated tests. The problem we see is that OCI does not offer an AI service capable of generating code from a traditional language prompt or input. If there is an expert in this field, we would greatly appreciate their insights and vision on this topic.

Benjamin Arnulf-Oracle

This forum is for Oracle Analytics. Please post in the Oracle AI forum or check internally for Oracle AI best practices.