Last updated:February 4, 2025

How to create and use a fine-tuned model?

In Fine-tuning Mode, you can create a custom AI model by refining a base model using your own dataset. Fine-tuning enhances the model’s knowledge in specific areas, allowing it to generate more accurate and relevant responses for your use case.

Step 1: Prepare Your Dataset for Fine-tuning

To fine-tune a model, you need to prepare your dataset in JSONL (JSON Lines) format. Each line in the JSONL file represents a conversation with a "prompt" and its corresponding "ideal response." This enables the model to learn from different types of interactions.

Example of JSONL Format

Here’s an example of a well-structured JSONL dataset for a technical support agent:

Json

Structure Breakdown:

System: Defines the assistant’s behavior and context. This part helps the model understand its role.
User: The user’s question or query.
Assistant: The model’s ideal response, which it should aim to replicate during fine-tuning.

Additional Examples

Example 1: Customer Support - Order Tracking
```
Json
```
Example 2: Healthcare Support - Symptom Inquiry
```
Json
```

You can add multiple examples like these (at least 20 recommended) in your JSONL file to train the model effectively.

Step 2: Upload Your Fine-Tuning Dataset

upload dataset

Once your JSONL file is ready, follow these steps to upload and start fine-tuning your model:

Go to the AskHandle Dashboard.
In the Command Center, select Fine-tuned Model.
Click on + Add Model to create a new fine-tuned model.
Choose a base model you want to fine-tune (e.g., GPT-4o-mini, GPT-4o, etc.).
Click Choose File to upload your JSONL dataset.
Click Save Model to start the fine-tuning process.

The fine-tuning process consists of two phases: validation and running. This typically takes a few minutes. Once the running phase is complete, your fine-tuned model will be ready for testing.

If the fine-tuning process fails, the issue is likely related to the data structure or file format. Double-check the format of your JSONL file, ensuring it's properly structured and saved as .jsonl.

Step 3: Use Your Fine-Tuned Model

Once your fine-tuned model is ready, you can choose from several modes to use it in: Select your model in these modes:

Mode	Use Case
Fine-Tuned Only	Pure custom interactions (e.g., brand-specific support).
Combined Data	Blend fine-tuned responses + your data sources + pre-trained AI knowledge base.
General Data	Only use AskHandle’s pre-trained AI knowledge base.
Your Own Data	Augment with your internal documents only (e.g., PDFs, FAQs).

Troubleshooting

Validation Failures: If the fine-tuning process fails during validation, check your JSONL file for errors. Ensure all data lines are correctly formatted, and that you're following the structure outlined in the examples above.
Slow Model Training: If fine-tuning takes too long, try reducing the size of your dataset and ensure that your file format is correct.