Fine Tuning (OpenAI)

Fine-Tuning Enabler on RagaAI Catalyst

This guide shows how to fine-tune a model within RagaAI Catalyst using its built-in Fine Tuning Enabler. The Fine Tuning Enabler orchestrates the necessary dataset formatting, job creation, and interaction with third-party platforms (such as OpenAI) under the hood, so you can focus on defining your job configurations and data mappings. Refer to the annotated screenshots below for a visual walkthrough.

Fine Tuning Enabler

The Fine Tuning Enabler is RagaAI Catalyst’s behind-the-scenes orchestrator for:

• Dataset Conversion: Automatically transforms your dataset into the structure required by the selected platform.

• Job Creation: Safely uploads files and initiates fine-tuning jobs with your chosen hyperparameters.

• Progress Tracking: Polls job status from the provider and updates Catalyst in near real-time.

• Result Registration: Stores the final fine-tuned model reference so you can easily query or test it in the Playground.

You do not need to perform any extra steps to use the Fine Tuning Enabler; simply provide the correct configurations in the UI, and it handles the rest.

---

1. Accessing the Fine-Tuning Tab

1. In the Catalyst sidebar, click Finetuning.

2. If you have not created any jobs yet, you will see “No Job found.”

---

2. Creating a New Job

1. Click + New Job (see bottom-right corner in the “No Job found” page).

2. A dialog box titled “Create New Job” appears.

3. Provide a Job Name (such as New_job) and click Create.

Note After you click Create, you will be taken to the Job Detail page where you can configure your fine-tuning parameters.

---

3. Selecting a Dataset

Once you’re on the Job Detail page:

1. Use the Select Dataset dropdown at the top-left of the main area to choose from your project’s available datasets (e.g., new_test_data, pii_test_v).

2. If your project has many datasets, use the search bar within the dropdown to quickly locate the one you need.

Important The chosen dataset’s schema and rows will be used to create prompts, contexts, and responses during fine-tuning. Make sure your dataset is structured appropriately (e.g., columns that can map to prompt, instructions, or answer text).

---

4. Configuring Hyperparameters

On the right-hand side of the Job Detail page:

1. Under Platform, choose your provider (e.g., OpenAI).

2. Set your training config values:

   • Epoch: Number of training epochs.

   • Learning Rate: Learning rate for gradient updates.

   • Batch Size: Size of mini-batches during training.

---

5. Template and Preview

The middle panel displays a Template section, illustrating how your dataset columns will map to:

• System (overall instructions)

• User (prompt)

• Assistant (desired response)

You can also see a short Result summary at the bottom:

• JobID (placeholder if not yet created)

• Model (to be determined once you run the job)

• Status (e.g., Draft, Running, or Succeeded)

---

6. Executing the Fine-Tuning Job

1. Review your chosen dataset, hyperparameters, and the template.

2. Click Execute Job (in the bottom-right corner of the middle panel).

3. If needed, Save Changes (top-right corner) to preserve your configuration before or after execution.

Once executed, the Fine Tuning Enabler will:

• Convert your dataset into the appropriate JSONL or other required format.

• Upload it securely to the selected platform (e.g., OpenAI).

• Create the corresponding Fine-Tuning job on that platform.

• Poll for job status, updating it in RagaAI Catalyst.

---

7. Monitoring Job Status

1. Navigate back to Finetuning from Job Status Page

2. Each job card displays whether it is Running, Succeeded, or Failed.

3. Click on a job card to view detailed logs, error messages (if any), and final results (fine-tuned model ID).

Tip Once your job succeeds, you can reference the newly fine-tuned model in Playground or other parts of RagaAI Catalyst.

---