About supervised fine-tuning for Translation LLM models

Supervised fine-tuning is a good option when you have a translation task with available labeled text data. It's particularly effective for domain-specific applications where the translation significantly differs from the general data the large model was originally trained on.

Supervised fine-tuning adapts model behavior with a labeled dataset. This process adjusts the model's weights to minimize the difference between its predictions and the actual labels.

Supported models

The following Translation LLM models support supervised tuning:

  • translation-llm-002 (In Public Preview, supports text only)

Limitations

  • Maximum input and output tokens:
    • Serving: 1,000 (~4000 characters)
  • Validation dataset size: 1024 examples
  • Training dataset file size: Up to 1GB for JSONL
  • Training example length: 1,000 (~4000 characters)
  • Adapter size:
    • Translation LLM V2: Supported value is only 4. Using any other values (e.g., 1 or 8) will result in failure.

Use cases for using supervised fine-tuning

General pretrained translation model works well when the text to be translated is based on general commonplace text structures that the model learned from. If you want a model to learn something niche or domain-specific that deviates from general translation, then you might want to consider tuning that model. For example, you can use model tuning to teach the model the following:

  • Specific content of an industry domain with jargon or style
  • Specific structures or formats for generating output.
  • Specific behaviors such as when to provide a terse or verbose output.
  • Specific customized outputs for specific types of inputs.

Configure a tuning job region

User data, such as the transformed dataset and the tuned model, is stored in the tuning job region. The only supported region is us-central1.

  • If you use the Vertex AI SDK, you can specify the region at initialization. For example:

    import vertexai
vertexai.init(project='myproject', location='us-central1')

  • If you create a supervised fine-tuning job by sending a POST request using the tuningJobs.create method, then you use the URL to specify the region where the tuning job runs. For example, in the following URL, you specify a region by replacing both instances of TUNING_JOB_REGION with the region where the job runs.

     https://TUNING_JOB_REGION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/TUNING_JOB_REGION/tuningJobs

  • If you use the Google Cloud console, you can select the region name in the Region drop down field on the Model details page. This is the same page where you select the base model and a tuned model name.

Quota

Quota is enforced on the number of concurrent tuning jobs. Every project comes with a default quota to run at least one tuning job. This is a global quota, shared across all available regions and supported models. If you want to run more jobs concurrently, you need to request additional quota for Global concurrent tuning jobs.

Pricing

Supervised fine-tuning for translation-llm-002 is in Preview. While tuning is in Preview, there is no charge to tune a model or to use it for inference.

Training tokens are calculated by the total number of tokens in your training dataset, multiplied by your number of epochs.

What's next