Skip to main content
Version: v0.21.x

Experiment (model) flow

Overview​

The flow of an H2O Enterprise LLM Studio experiment from creation to deployment can be summarized in the following sequential steps:

Step 1: Upload/select a dataset​

As the first step in the experiment flow, import your experiment dataset to H2O Enterprise LLM Studio. Optionally, you can also use an existing sample dataset that is already available on H2O Enterprise LLM Studio, to begin creating the first experiment.

Step 2: Data generation (optional)​

If required, H2O Enterprise LLM Studio enables you to generate synthetic data to augment your dataset. This feature is useful for scenarios such as filling missing labels, creating new categories, or adding additional columns to enhance your data for model training or experimentation.

Step 3: Configure experiment​

H2O Enterprise LLM Studio offers several hyperparameter settings that you can adjust for your experiment. It is not mandatory to configure all of them and you can choose to proceed with default settings, but you do have the option to configure advanced settings as you wish at this stage. See Experiments for details on each setting.

Step 4: Improve the experiment with AutoML (optional)​

At this stage, you can also enable AutoML capabilities for your experiment so that H2O Enterprise LLM Studio will analyze the model's performance, score it based on metrics such as accuracy and efficiency, and suggest hyperparameter changes to improve your model based on previous results.

Step 5: Deploy an experiment's model​

After understanding and inspecting your built model, you can deploy your model using the H2O Enterprise LLM Studio user interface (UI). You will receive a valid endpoint for your deployed model which you can then use to make RESTful API calls for this model. If your model is a chatbot, you can test it out within the H2O Enterprise LLM Studio UI and receive live responses.


Feedback