Skip to main content

Model fine-tuning

Not OnPrem

Fine-tuning allows you to adapt a base model to a specific use case by training it further on a curated dataset. This can significantly improve model performance on domain-specific tasks like legal summarization, biomedical text generation, product recommendation, or customer support automation.

aiXplain supports fine-tuning for both hosted models and passthrough models (those hosted on third-party infrastructure). Once fine-tuned, your model is deployed and accessible like any other model on the platform.

Why Fine-Tune?​

Fine-tuning is especially useful when:

  • You have a specialized domain (e.g., legal, healthcare, education) where generic LLMs struggle with accuracy.

  • You want to improve output consistency or align it with a brand voice.

  • You're building a product that requires specific task behavior, such as question answering, classification, or summarization.

  • You have training data that reflects your task more closely than a general-purpose dataset.

How to fine-tune a model

This guide walks you through the process of fine-tuning models using the aiXplain SDK. Learn how to select datasets and configure fine-tuning settings.

Generic Example (Template)​

from aixplain.factories import DatasetFactory, ModelFactory, FinetuneFactory

dataset = DatasetFactory.get("...") # specify Data ID
model = ModelFactory.get("...") # specify Model ID

finetune = FinetuneFactory.create(
  "finetuned_model",
  [dataset],
  model
)

finetuned_model = finetune.start()

finetuned_model.check_finetune_status()

FineTune Examples​

The following examples cover the four supported FineTune use cases. View them in this documentation or by opening their corresponding Google Colab notebooks.

Text generation (passthrough)
Open In Colab

Text generation (hosted)
Open In Colab

note

Passthrough: models hosted on third-party infrastructure
Hosted: models hosted on aiXplain's infrastructure

Imports​

from aixplain.factories import DatasetFactory, ModelFactory, FinetuneFactory
from aixplain.enums import Function, Language # for search
from aixplain.modules.finetune import Hyperparameters # for hosted models

Select Model & Datasets​

info

Datasets are currently private, so you must first onboard the datasets in the examples below (or similar) to follow along.
See our guide on How to upload a dataset.

Model​

# Choose 'exactly one' model
model_list = ModelFactory.list(
    function=Function.TEXT_GENERATION,
    is_finetunable=True
)["results"]

for model in model_list:
    print(model.__dict__)
Show output
selected_model = ModelFactory.get("640b517694bf816d35a59125")
selected_model.__dict__
Show output

Dataset​

# Choose 'one or more' datasets
dataset_list = DatasetFactory.list(
  function=Function.TEXT_GENERATION,
  page_size=5
)["results"]

for dataset in dataset_list:
  print(dataset.__dict__)
Show output
selected_dataset = DatasetFactory.get("6501ea64b61fed7fe5976c49")
selected_dataset.__dict__
Show output

Create a FineTune​

Use FinetuneFactory to create a FineTune object and the cost method to check the estimated training, hosting and inference costs.

finetune = FinetuneFactory.create(
  "<UNIQUE_FINETUNE_NAME>",
  [selected_dataset],
  selected_model
)

finetune.__dict__
Show output

Cost​

finetune.cost.to_dict()
Show output

Starting a FineTune​

Call the start method to begin fine-tuning and the check_finetune_status method to check its status.

finetune_model = finetune.start()
status = finetune_model.check_finetune_status()

Status can be one of the following: onboarding, onboarded, hidden, training, deleted, enabling, disabled, failed, deleting.

tip

You can use a loop to check the status.

import time

while status != "onboarded":
  status = finetune_model.check_finetune_status()
  print(f"Current status: {status}")
  time.sleep(10)

Once onboarded, you are ready to use the model as any other which can be integrated into your agents, providing customized solutions! 🥳