Model tuning

With Yandex Foundation Models, you can tune YandexGPT Lite and Llama 8B¹ text generation models and YandexGPT Lite-based classifiers using the LoRA (Low-Rank Adaptation of Large Language Models) method.

Model tuning in Yandex Foundation Models is at the Preview stage.

Fine-tuning text generation modelsFine-tuning text generation models

You cannot fine-tune a text generation model based on new data, e.g., the knowledge base of your support service. However, you can train the model to generate responses in a specific format or analyze texts. You can train the model to:

• Summarize and rewrite texts.
• Generate questions and answers from text input.
• Provide responses in a particular format or style.
• Classify texts, queries, and conversations.
• Extract entities from texts.

Fine-tuning in Foundation ModelsFine-tuning in Foundation Models

For more information on tuning data requirements, see Text generation datasets and Text classification datasets.

You need to upload the prepared data to Yandex Cloud as a dataset. By default, you can upload up to 5 GB of tuning data into one dataset. For all limitations, see Quotas and limits in Yandex Foundation Models.

After you upload a dataset, start tuning by specifying its type and parameters (optional). Tuning can take from 1 to 24 hours depending on the amount of data and system workload.

For a model tuning example, see Tuning a text generation model.

You will need an ai.editor role for model tuning in Foundation Models. This role allows you to upload data and start the tuning process.

Requests to fine-tuned modelsRequests to fine-tuned models

Once you complete tuning your model, you will get its ID. Provide this ID in the modelUri field of the request body. You can submit requests to a fine-tuned text generation model through the text generation API, AI Assistant API, or from Yandex DataSphere and other applications. To send a request to a fine-tuned classifier, use the classify Text Classification API method. You can also use Yandex Cloud ML SDK to work with fine-tuned models:

To send API requests in DataSphere notebooks, add the user or service account you are going to use for requests to the list of DataSphere project members. The account must have the ai.languageModels.user role.

ExamplesExamples

• Tuning a text generation model.
• Tuning a text classification model
• Model tuning in DataSphere

¹ Llama was created by Meta. Meta is designated as an extremist organization and its activities are prohibited in Russia.