Fine-tuning tutorial

This tutorial will help you get started with the Gemini API tuning service using either the Python SDK or the REST API using curl. The examples show how to tune the text model behind the Gemini API text generation service.

View on ai.google.dev

Try a Colab notebook

View notebook on GitHub

Limitations

Before tuning a model, you should be aware of the following limitations:

Fine-tuning datasets

Fine-tuning datasets for Gemini 1.5 Flash have the following limitations:

The maximum input size per example is 40,000 characters.
The maximum output size per example is 5,000 characters.
Only input-output pair examples are supported. Chat-style multi-turn conversations are not supported.

Tuned models

Tuned models have the following limitations:

The input limit of a tuned Gemini 1.5 Flash model is 40,000 characters.
JSON mode is not supported with tuned models.
Only text input is supported.

Before you begin: Set up your project and API key

Before calling the Gemini API, you need to set up your project and configure your API key.

Expand to view how to set up your project and API key

Get and secure your API key

You need an API key to call the Gemini API. If you don't already have one, create a key in Google AI Studio.

Get an API key

It's strongly recommended that you do not check an API key into your version control system.

You should store your API key in a secrets store such as Google Cloud Secret Manager.

This tutorial assumes that you're accessing your API key as an environment variable.

Install the SDK package and configure your API key

The Python SDK for the Gemini API is contained in the google-generativeai package.

Install the dependency using pip:
```
pip install -U google-generativeai
```

Import the package and configure the service with your API key:

import os
import google.generativeai as genai

genai.configure(api_key=os.environ['API_KEY'])

List tuned models

You can check your existing tuned models with the tunedModels.list method.

for model_info in genai.list_tuned_models():
    print(model_info.name)tuned_models.py

Create a tuned model

To create a tuned model, you need to pass your dataset to the model in the tunedModels.create method.

For this example, you will tune a model to generate the next number in the sequence. For example, if the input is 1, the model should output 2. If the input is one hundred, the output should be one hundred one.

import time

base_model = "models/gemini-1.5-flash-001-tuning"
training_data = [
    {"text_input": "1", "output": "2"},
    # ... more examples ...
    # ...
    {"text_input": "seven", "output": "eight"},
]
operation = genai.create_tuned_model(
    # You can use a tuned model here too. Set `source_model="tunedModels/..."`
    display_name="increment",
    source_model=base_model,
    epoch_count=20,
    batch_size=4,
    learning_rate=0.001,
    training_data=training_data,
)

for status in operation.wait_bar():
    time.sleep(10)

result = operation.result()
print(result)
# # You can plot the loss curve with:
# snapshots = pd.DataFrame(result.tuning_task.snapshots)
# sns.lineplot(data=snapshots, x='epoch', y='mean_loss')

model = genai.GenerativeModel(model_name=result.name)
result = model.generate_content("III")
print(result.text)  # IVtuned_models.py

The optimal values for epoch count, batch size, and learning rate are dependent on your dataset and other constraints of your use case. To learn more about these values, see Advanced tuning settings and Hyperparameters.

Since tuning a model can take significant time, this API doesn't wait for the tuning to complete. Instead, it returns a google.api_core.operation.Operation object that lets you check on the status of the tuning job, or wait for it to complete, and check the result.

Your tuned model is immediately added to the list of tuned models, but its state is set to "creating" while the model is tuned.

Check tuning progress

You can check on the progress of the tuning operation using the wait_bar() method:

for status in operation.wait_bar():
    time.sleep(10)

You can also use operation.metadata to check the total number of tuning steps and operation.update() to refresh the status of the operation.

You can cancel your tuning job any time using the cancel() method.

operation.cancel()

Try the model

You can use the tunedModels.generateContent method and specify the name of the tuned model to test its performance.

model = genai.GenerativeModel(model_name="tunedModels/my-increment-model")
result = model.generate_content("III")
print(result.text)  # "IV"tuned_models.py

Update the description

You can update the description of your tuned model any time using the genai.update_tuned_model method.

genai.update_tuned_model('tunedModels/my-increment-model', {"description":"This is my model."})

Delete the model

You can clean up your tuned model list by deleting models you no longer need. Use the tunedModels.delete method to delete a model. If you canceled any tuning jobs, you may want to delete those as their performance may be unpredictable.

genai.delete_tuned_model("tunedModels/my-increment-model")