Fine-tuning
Agenthost now supports fine-tuning in beta. Please reach out to [email protected] should you experience any issues.
Fine-tuning allows you to train a custom model based on your own specific needs and data. Generally fine-tuning is a way to create a better experience for your users. You can reach more about fine tuning by reading OpenAI's documentation. Fine-tuning is only available to users on Pro and Growth paid plans.
Fine-tunes use 2x the message limits as gpt-3.5-turbo. This includes running the training as well as users chatting with the fine tuned model after training is complete. Fine-tuning is only supported with v1 agents.
To create a new fine-tuned model, follow these steps.
- Select Fine-tuning from the left side menu in the agent project page.
- Select Create on the top right side of the page. This will bring up a modal view.
- Enter a value for the name of the fine-tune. This can be anything you wish and is just for your referencing. Only gpt-3.5-turbo is available for fine tuning currently.
- Click Create to create the model.
- Creating the model is the first part. Now we'll need to select the messages we want to use for fine-tuning our model.
Now we need to select the messages we want to train our new fine-tuned model on. You must select atleast 10 messages to fine tune your agent.
- Select Manage on the newly created fine-tune job
- After selecting manage, click the Add messages button near the center of the modal that appears.
- This will take you to the Conversations view. From there, open a thread with messages you want your model trained on.
- On all messages from your agent, you'll now see a Train option. Click that button to add the message to your fine-tuned model training list.
- You must select atleast 10 messages to start fine-tuning.
Once happy with all the messages you've added to your training, go back to the Fine-tuning section.
Then click Run to begin the fine-tune process. The status wil then change to running
How long the fine-tune process runs depends but it can be anywhere from a couple of minutes to hours. Once fine-tuning completes, you'll be notified and the status will change to completed.
Congratulations on creating your fine-tuned model. Now the final piece is to update your agent to use your newly created fine-tuned model.
- Go to the Agent tab from the left menu.
- Scroll down to Advanced Settings and find GPT model
- Click the dropdown and select use fine tuned model
- A modal should popup asking you to select your fine tuned model. Select the fine-tuned model you just created.
- After selecting th model, be sure to scroll up and click Save to save your changes.
Now you can go to the chat page and ask your agent questions. Fine-tuned models do sometimes take slightly longer to responded compared to the base models.
