Introduction to custom LLMs
To better meet your conversational search requirements, OpenSearch LLM-Based Conversational Search Edition provides custom large language models (LLMs). You can create and train custom LLMs based on your business requirements. This topic describes how to create and train a custom LLM.
Procedure
You can create a custom LLM only if your OpenSearch LLM-Based Conversational Search Edition instance is of the Professional Edition.
Create a custom LLM
1. After you create an OpenSearch LLM-Based Conversational Search Edition instance of the Professional Edition, go to the instance details page and click Custom LLM in the left-side navigation pane. Then, click Create.
2. Enter a model name, select a model from the Basic Model drop-down list, and upload a file in the Existing Q&A Pair section. Then, click OK.
Model Name: The name must be 1 to 30 characters in length and can contain letters, digits, and underscores (_). It must start with a letter and cannot be the same as other model names.
Basic Model: Only opensearch-llama2-13b is supported.
Existing Q&A Pair: Files in the xlsx format are supported. The size of a single file cannot exceed 128 MB. The file name can contain letters, digits, underscores (_), and hyphens (-). For more information, click Data Sample in the Import File panel.
Billing Method: For more information, see Billing overview of OpenSearch LLM-Based Conversational Search Edition.
Number of LLMs: You can create two LLMs at most.
3. After the LLM is created, click Train or Complete.
Train a custom LLM
1. After an LLM is created, its Model Status is Unavailable and Latest Version Status is Pending Training. Click Train in the Actions column. In the message that appears, click Submit to start model training.
Model Name: the name of the custom LLM.
Created At: the time when the custom LLM was created.
Model Status: the state of the custom LLM. Valid values: Available and Unavailable.
Last Training Started At: the time when the last training of the custom LLM started. You can train a custom LLM multiple times.
Latest Version Status: the state of the latest version of the custom LLM. A newly created LLM is in the Pending Training state, an LLM being trained is in the Scheduling state, and a trained LLM is in the Trained and Ready state.
View a custom LLM
1. After a custom LLM is trained, you can click Details in the Actions column to view the information about the LLM.
Basic Information | |
Created At | The time when the custom LLM was created. |
Model Status | The state of the custom LLM. Valid values: Available and Unavailable. |
Last Training Started At | The time when the last training of the custom LLM started. You can train a custom LLM multiple times. |
Latest Version Status | The state of the latest version of the custom LLM. A newly created LLM is in the Pending Training state, an LLM being trained is in the Scheduling state, and a trained LLM is in the Trained and Ready state. |
Configuration Information | |
Basic Model | Only opensearch-llama2-13b is supported. |
Existing Q&A Pair | Files in the xlsx format are supported. The size of a single file cannot exceed 128 MB. The file name can contain letters, digits, underscores (_), and hyphens (-). For more information, click Data Sample in the Import File panel. |
Training History | |
Model Version | The version of the LLM. |
Status | The state of the LLM version. |
Started At | The time when the training started. |
Ended At | The time when the training ended. |
2. Click Test Effect in the Actions column of a training record to view the training effect of the LLM.
On the Q&A Test page, use search keywords relevant to the training materials to check the performance of the custom LLM.