Skip to content

Fine-Tuning UI

After selecting Fine Tuning from the sidebar menu, you’ll see a list of existing problems and the option to create new ones.

Fine-Tuning Home Page

1. Creating a New Fine-Tuning Problem

  1. Click Create new problem.
  2. Enter the Label and Name for your fine-tuning job.
  3. Select one of the following task types:

Create New Fine-Tuning Problem

1.1 Task Types

  1. text_causal_classification_modeling

    • Definition – Classify text into categories (e.g., sentiment analysis, topic classification).
    • Dataset Structure
      {"text": "This product is amazing!", "label": "positive"}
{"text": "I'm not sure how I feel about this.", "label": "neutral"}

  2. text_causal_language_modeling

    • Definition – Predict next tokens in a sequence (text generation, autocompletion).
    • Dataset Structure
      {"system_prompt": "You are a helpful assistant."}
{"sample_question": "Can I fine-tune LLMs here?"}
{"sample_answer": "Yes, select 'Fine-Tuning' in the sidebar to begin."}

  3. text_dpo_modeling (Direct Preference Optimization)

    • Definition – Align model output with human preferences.
    • Dataset Structure
      {"prompt": "Write a product description.", "chosen": "Better descriptive text", "rejected": "Less ideal text"}

  4. text_rlhf_modeling (Reinforcement Learning with Human Feedback)

    • Definition – Use RL + human feedback to refine model output.
    • Dataset Structure
      {"prompt": "What is MLOps?", "response": "It's an abbreviation", "reward": 0}
{"prompt": "What is MLOps?", "response": "It operationalizes ML workflows", "reward": 1}

  5. text_seq_to_seq (Sequence-to-Sequence)

    • Definition – Transform one text sequence into another (e.g., machine translation).
    • Dataset Structure
      {"input_text": "Bonjour!", "target_text": "Hello!"}
{"input_text": "What is the capital of France?", "target_text": "Paris"}

2. Configuring Your Problem

Once created, open the Overview page to adjust specific configurations for your fine-tuning task.

Fine-Tuning Problem Overview

3. Starting a New Fine-Tuning Run

  1. Click Start new run in the problem’s overview.
  2. Choose your preferred instance type.
  3. Launch the fine-tuning job.

Fine-Tuning Start New Run

You can track the status of your runs in the Overview tab:

Fine-Tuning Runs

3.1 Monitoring a Run

  • Logs – Click on a running instance to see real-time logs.
  • Stop and Push – Halts the run and uploads the checkpoint to Hugging Face or S3 (provided a checkpoint exists).
  • Stop Run – Stops the run entirely, discarding any progress.

Fine-Tuning Progress

4. Data Frames

Fine-tuning requires a dataset. Navigate to the Data frames tab under Fine Tuning to manage or upload these datasets.

Fine Tuning Data Frames

4.1 Uploading a Data Frame

  1. Click Upload dataframe.
  2. Select your file.
  3. Use this data in your subsequent fine-tuning problems.

Fine Tuning Upload Data Frame

Next Steps