anchen1011 / chatgpt-finetune-ui

Simple python WebUI for fine-tuning ChatGPT (gpt-3.5-turbo)

chatgpt-finetune-ui

Simple python WebUI for fine-tuning gpt-3.5-turbo

Installation

pip install openai
pip install streamlit

Usage

streamlit run main.py --server.port 8001 --server.address 0.0.0.0

Experimental Demo

http://system2.space:8001

chatgpt-finetune-ui's Issues

Tables not displaying on Safari: "BigInt64Array is not available in this environment"

Hello,

I have been using the "chatgpt-finetune-ui" tool for gpt-3.5-turbo fine-tuning and it's been a wonderful experience. However, I stumbled upon a particular issue while browsing the results on the Safari browser. The tables, which should ideally display the training results, are not being shown at all.

In place of the table, I'm confronted with an error message stating, "BigInt64Array is not available in this environment". I believe this issue stems from the use of the 'BigInt64Array' JavaScript object, which, to my understanding, isn't supported by Safari.

The application works fine with other browsers such as Chrome and Firefox, so this issue seems to be Safari-specific.

Could you possibly provide a workaround or use a different mechanism for displaying the data that is compatible with all browsers? Even a polyfill solution would be much appreciated, should that be possible.

Here are further details about my environments:

• Browser: Safari Version 14.0.3 (16610.4.3.1.7)
• OS: MacOS Big Sur 11.2.3

Thank you very much for your attention to this issue. If you need more information, I will be glad to provide. Looking forward to your response.

How to define hyper parameters for the fine-tuning task?

Currently the UI allows fine-tuning ChatGPT model but there is no way to define hyper parameters like number of epochs, batch size etc. It would be useful to have some configurable options to define these hyper parameters for the fine-tuning task running in the backend.

This will give more control and flexibility to the users to fine-tune the model as per their need by tuning these important parameters.

Some suggested parameters that can be configurable:

• Number of epochs
• Batch size
• Learning rate
• Optimizer (Adam, SGD etc)

Please let me know if adding these configurable options via the UI makes sense or if you have any other suggestions.

Issue with openai import: "cannot import name 'where' from 'certifi'"

Hello,

I'm experiencing an error when trying to run the project. Upon launching, I get an error stating, "cannot import name 'where' from 'certifi'" that traces back to the openai import. For clarity, I'm using Python 3.9.6 in a virtual environment.

Here is the error message in full detail:

ImportError Traceback (most recent call last)
<ipython-input-1-0a5ea5a78279> in <module>
----> 1 import openai
      2 
      3 ImportError: cannot import name 'where' from 'certifi'

I've tried reinstalling 'certifi' and 'openai' Python packages using pip, but the error persists.

If anyone has any insights or solutions, I would appreciate any guidance to resolve this.