Generative AI for properties programing language.

Ompratap Singh 0

I am currently working on a Generative AI project.

The goal is to leverage pre-trained models like GPT-4/GPT-3, Codex, etc., to generate responses in my proprietary language syntax.

For instance, if I ask for Java code to print numbers from 1 to 10, I would expect to receive the corresponding Java code. I aim to achieve the same functionality with my proprietary code.

In the initial steps:

I have uploaded numerous sample data and grammar documents for my language to the storage account and created an index. Additionally, I have added system prompts to my chat playground app and deployed the GPT-4 model.

However, the responses I'm receiving are not as accurate as expected. Could you guide me on how to fine-tune the model or improve the accuracy of the responses?

YutongTie-MSFT 53,126 Reputation points

2024-06-12T00:10:13.9866667+00:00

Hello Ompratap Singh

Thanks for reaching out to us. May I know when you talk about not expected, what's kind of error or issue happened for your model?

Have you tried Assistant API with GPT-4 for code generator? Here is the document -

https://learn.microsoft.com/en-us/azure/ai-services/openai/assistants-quickstart?tabs=command-line%2Ctypescript&pivots=programming-language-studio
Ompratap Singh 0 Reputation points

2024-06-12T09:03:57.4866667+00:00

My ultimate objective is to ground my proprietary language into a generative AI model. The aim is for the model to treat this as a programming language and return code, refactor code in my proprietary language, and adhere to best practices.

What have I done so far? I've uploaded my programming guides and several samples for each code syntax to a storage account. I've added data to the chat playground with an index and vector search to interact with my data. Additionally, I've included system messages for further instructions, such as where to look for specific questions, etc.

Is this the correct approach to achieve my goal, or should I adopt a different strategy?

I haven't used the Assistant API with GPT-4 yet. Would it assist me in achieving this, and if so, is there an SDK available for it?
YutongTie-MSFT 53,126 Reputation points

2024-07-05T02:52:59.3333333+00:00
Hello Ompratap, I would suggest you going through the Assistant(preview) as well to see if that is a good way for you, I have posted the both API and SDK below for your reference.

Assistant API (GPT-4): Using the Assistant API with GPT-4 can significantly enhance your AI model's capabilities in understanding and generating code based on natural language queries.

Document for your reference - https://github.com/Azure-Samples/azureai-samples/tree/main/scenarios/Assistants

SDK you are looking for - https://learn.microsoft.com/en-us/azure/ai-services/openai/assistants-quickstart?tabs=command-line%2Ctypescript&pivots=programming-language-python

Training and Integration: You can train the GPT-4 model on your specific domain by fine-tuning it with your proprietary language data. This approach can lead to more accurate and contextually relevant responses if you have the training data set, and you are good at fine-tuning.

Next Steps I will suggest you consider -

Explore Assistant(preview): Familiarize yourself with the Assistant capabilities and how they align with your requirements for generating and refactoring code.

SDK Integration: Utilize Python SDKs or REST APIs to integrate GPT-4 into your existing system. This will allow you to leverage its capabilities directly within your application.

Iterative Improvement: Continuously refine your approach based on user feedback and performance metrics. Monitor model accuracy and user satisfaction to drive improvements.

Share via

Generative AI for properties programing language.

Your answer