Instructions to fine-tune a new OpenAI GPT 3.5 model

To run the OpenAI fine-tuning utility, follow these steps:

Steps to run in Windows

1. Open the Command Prompt: Press Win + R on your keyboard, type cmd, and enter to open the Command Prompt window.

2. Navigate to the Directory: Use the cd command to change directories to where the OpenAI trainer utility is located.

   cd <:INSTALLED_DIRECTORY>\BoldServices\utilities\openai_trainer

3. Execute the Application: Once in the correct directory, run the executable file by typing the following command, replacing <:API_KEY> with your actual OpenAI API key.

   OpenAITrainer.exe -provider OpenAI -OpenAIKey <:API_KEY>

   Ensure to replace <:API_KEY> with the API key provided to you by OpenAI.

NOTE : You can generate your API key by visiting the OpenAI portal at https://platform.openai.com/api-keys.

Steps to run in Linux

1. Open a Terminal: Open a terminal using the appropriate key combination (e.g., Ctrl + Alt + T).

2. Navigate to the Directory: Use the cd command to change directories to where the OpenAI trainer utility is located.

   cd /var/www/bold-services/application/utilities/openai_trainer

3. Once in the correct directory, run the dll file by typing the following command, replacing <:API_KEY> with your actual OpenAI API key.

   /var/www/bold-services/dotnet/dotnet OpenAITrainer.dll -provider OpenAI -OpenAIKey <:API_KEY>

Steps to run in Docker

1. Run the following command to list all the containers and note the BoldBI container name.

   docker ps

2. Execute the following command for bashing boldbi running container.

   docker exec -it <BoldBI container name> bash 

3. Navigate to the Directory: Use the cd command to change directories to where the OpenAI trainer utility is located.

   cd /application/utilities/openai_trainer

4. Once in the correct directory, run the dll file by typing the following command, replacing <:API_KEY> with your actual OpenAI API key.

   dotnet OpenAITrainer.dll -provider OpenAI -OpenAIKey <:API_KEY>

Once the training and validation JSONL files are uploaded for fine-tuning a machine-learning model, the system will begin processing the data. During this time, you will receive a jobId, which is a unique identifier for your fine-tuning task. This ID allows you to track the progress of your fine-tuning job.

Status of fine-tuning

The system will also provide you with status updates. The status indicates the current stage of the fine-tuning process. Initially, the status will show as validating_files as the system prepares and starts the training with your data.

Subsequently, the status will transition to running and persist until the fine-tuning process succeeds or encounters a failure.

Hitting the esc key triggers the termination process for the ongoing fine-tuning task.

The status of the fine-tuning job will be displayed after every one minute.

Post fine-tuning

After a successful fine-tuning process, the console will display the name of the fine-tuned model. It is important to make a note of this model name for future use-cases and configure it in BoldBI.

NOTE : An error message will be displayed if an error occurs during the fine-tuning process.

Steps to update fine-tuned model name

Bold BI version between 7.1 and 7.5

1. Go to Manage Sites.

2. Inside the Settings tab, go to the Configuration tab.

3. In the config.json file set the OpenAI key and OpenAI model fine-tuned name.

Syntax

"QnA": {
        "Enable": true,
        "OpenAiKey": "<:API_KEY>",
        "OpenAiModelName": "<:MODEL_NAME>"
      }

4. Replace <:API_KEY> and <:MODEL_NAME> with your apiKey and fine-tuned model name.

5. Restart the Internet Information Services (IIS) Manager and refresh the dashboard.

Bold BI version 7.5 or later

1. Go to Manage Sites.

2. Inside Settings tab and go the Configuration tab.

3. In the config.json file you will find by default OpenAI Active as true and AzureAI Active as false.

4. Replace <:API_KEY> and <:MODEL_NAME> with your apiKey and fine-tuned model name.

Syntax

   "OpenAI":{
     "Active":true,
     "OpenAiKey": "<:API_KEY>",
     "OpenAiModelName": "<:MODEL_NAME>"
   }

NOTE Ensure the Active is true for OpenAI and false for AzureAI.

5. Restart the Bold BI services and open the dashboard.

NOTE: Fine-tuning once will cost approximately $25 USD. Each search in the Q&A widget will incur a cost based on the pricing of the OpenAI fine-tuned model (https://openai.com/pricing), where the number of input and output tokens depends on the question and the data sources connected to the dashboard.