Hologres offers various built-in AI models for diverse AI applications. You can deploy these models from the Hologres console based on your business needs. This document explains which AI models are available and how to deploy them.
Supported models
These built-in models require Hologres V3.2 or later versions.
Model name | Category | Recommended minimum vCPUs for single-replica deployment | Recommended minimum memory for single-replica deployment (GB) | Recommended minimum number of GPUs for single-replica deployment | Recommended minimum GPU memory for single-replica deployment (GB) | Required instance versions | Notes |
PDF conversion model | 20 | 100 | 1 or more | 48 | V4.0 and later | None | |
Multimodal model | 7 | 24 | 1 or more | 24 | V4.0 and later | None | |
Multimodal model | 7 | 30 | 1 or more | 48 | V4.0 and later | None | |
Multimodal model | 7 | 30 | 1 or more | 96 | V4.0 and later | None | |
Image embedding model | 7 | 24 | 1 | 24 | V4.0 and later |
| |
Multilingual embedding model for images | 7 | 24 | 1 | 24 | V4.0 and later |
| |
Image embedding model | 7 | 24 | 1 | 24 | V4.0 and later |
| |
Image embedding model | 7 | 24 | 1 | 24 | V4.0 and later |
| |
LLM | 7 | 30 | 1 or more | 8 | V3.2 and later | None | |
LLM | 7 | 30 | 1 or more | 16 | V3.2 and later | None | |
LLM | 7 | 30 | 1 or more | 32 | V3.2 and later | None | |
LLM | 7 | 30 | 1 or more | 48 | V3.2 and later | None | |
LLM | 7 | 30 | 1 or more | 96 | V3.2 and later | None | |
Sentiment classification | 7 | 30 | 1 | 4 | V3.2 and later | None | |
Text embedding model | 7 | 30 | 1 | 12 | V3.2 and later | Output vector dimensions: 768 | |
Text embedding model | 7 | 30 | 1 | 16 | V3.2 and later | Output vector dimensions: 1024 | |
Text embedding model | 7 | 30 | 1 | 8 | V3.2 and later | Output vector dimensions: 512 | |
Text embedding model | 7 | 30 | 1 | 8 | V3.2 and later | None | |
Text embedding model | 7 | 30 | 1 | 32 | V3.2 and later | None | |
Text embedding model | 7 | 30 | 1 | 48 | V3.2 and later | None | |
recursive-character-text-splitter | Text chunking | 15 | 30 | 0 | 0 | V3.2 and later | Select CPU specifications as needed. Setting the number of GPUs is not required. |
Long text embedding | 7 | 30 | 1 | 12 | V3.2 and later | Output vector dimensions: 768 | |
Long text embedding | 7 | 30 | 1 | 12 | V3.2 and later | Output vector dimensions: 768 | |
Long text embedding | 7 | 30 | 1 | 16 | V3.2 and later | Output vector dimensions: 1024 | |
Long text embedding | 7 | 30 | 1 | 16 | V3.2 and later | Output vector dimensions: 1024 | |
Long text embedding | 7 | 30 | 1 | 8 | V3.2 and later | Output vector dimensions: 384 | |
Long text embedding | 7 | 30 | 1 | 8 | V3.2 and later | Output vector dimensions: 512 |
Prerequisites
You have purchased AI resources.
Notes
Select and deploy models from the list above. Each model requires specified minimum AI resources.
You can deploy multiple models on one instance, provided the total resource consumption does not exceed your purchased quota. Scale up if resources are insufficient.
For primary/secondary instances: Model deployment and management (modifying resources, deleting) are exclusive to the primary instance. Secondary instances can view the primary instance's models and call them via AI functions.
Deploy a model
Log on to the Hologres console and select a region.
In the left navigation menu, click Instances. Then, click the target instance ID.
On the Instance Details page, click AI Node.
In the Models section, click Deploy Model.
In the Deploy Model dialog box, set Model Name and Model Type.
The parameters for Resource Configurations are automatically populated based on the selected Model Type.
After you complete the configurations, click OK to deploy the model.
In the Models section, view the deployment status and perform the following operations:
Adjust model configurations: In the Actions column of the target model, click Adjust Configurations.
Delete the model: In the Actions column of the target model, click Delete.
NoteHologres does not check for dependent services when deleting a model. Exercise extreme caution to prevent service downtime.
Next step
After deploying a model, you can call it via AI functions. For more information, see AI functions.