Define a large language model (LLM) source
Overview
H2O Model Validation lets you obtain large language model (LLM) insights into models, datasets, and validation tests. To receive insights, select and define one of the supported LLM sources. See Supported large language model (LLM) sources.
Insights generated heavily depend on the LLM source. With the ongoing advancements in research and the emergence of new models, this functionality is more akin to an experimental feature we continuously enhance. Depending on the specific LLM source, there is a significant possibility of encountering hallucinations or inaccurate judgments within the generated insights.
Instructions
To define an LLM source, consider the following instructions:
- In the H2O Model Validation navigation menu, click Settings.
- In the Active LLM source list, select one of the supported LLM sources.
- Depending on the selected LLM source, consider the following instructions:
- h2oGPT-GM
- h2oGPT
- H2O MLOps
- GPT-4
- In the URL box, enter the URL for one of the supported h2oGPT-GM models.
- This setting refers to the URL of one of the supported h2oGPT-GM models
- Select/clear the Enable prompt special tokens checkbox.
- A few model endpoints expect special tokens to be placed in prompts to generate successful (better) answers. H2O Model Validation adds certain special tokens to the prompt if the checkbox is checked
- Click Save.
- In the Gradio URL list, select one of the h2oGPT URL model endpoints.
- This setting refers to the Gradio URL to the h2oGPT model endpoint (that is either the public https://gpt.h2o.ai endpoint or one's local Gradio endpoint generated by the h2oGPT repo)
- Click Save.
- In the Name box, enter a name for the model.
- This setting refers to a name to attribute to an H2O MLOps model
- In the URL box, enter the endpoint URL of the deployed model in H2O MLOps.
- This setting refers to the endpoint URL of a deployed model (an LLM model) in H2O MLOps
- Click Save.
- In the OpenAI API key box, enter an OpenAI API key.
- This setting refers to an OpenAI API key associated with an OpenAI account
- Click Save.
Supported large language model (LLM) sources
Below, observe the supported LLMs in H2O Model Validation.
h2oGPT-GM
H2O Generative Pre-trained Transformer Grandmaster (h2oGPT-GM) is a Kaggle grandmaster platform where H2OAI provides multiple large language models (LLM).
- To learn more, see h2oGPT-GM (UI).
h2oGPT
H2O Generative Pre-trained Transformer (h2oGPT) is an H2OAI large language model (LLM) fine-tuning framework and chatbot UI with document(s) question-answer capabilities.
- To learn more, see h2oGPT (repo).
- To view the h2oGPT UI, see h2oGPT (UI).
H2O MLOps
H2O MLOps is an open, interoperable platform for model deployment, management, governance, monitoring, and alerting that features integration with H2O Driverless AI, H2O-3 open source, and third-party models.
GPT-4
Generative Pre-trained Transformer 4 (GPT-4) is a large language model (LLM) created by OpenAI.
- To learn more, see GPT-4.
- Submit and view feedback for this page
- Send feedback about H2O Model Validation to cloud-feedback@h2o.ai