Stay organized with collections
Save and categorize content based on your preferences.
This page describes how to use a textual large-language model (LLM) in the custom recommendation models. We train these models for you. You can enable the pretrained features in the custom recommendation models.
Recommendations uses the product description field to feed to LLMs and put them into your recommendations models.
New LLM textual features
While it's possible to get text embeddings by manually configuring a Vertex AI generative model, you might want to integrate the new LLM capabilities into your recommendations models to improve performance.
The text embeddings are more descriptive, longer, and are not repetitive, as well as have multilingual interpretation capabilities. This feature is based on an allowlist. Contact support for enabling this feature.
There's no charge for using the text embeddings and they are included in Vertex AI Search pricing.
The LLM-pretrained embeddings improve semantic understanding of long form text searches such as descriptions.
See the following resources for more information on how to use embeddings and generative AI alone in your own custom ML training:
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-08-29 UTC."],[],[],null,["# Utilizing pretrained LLM\n\nThis page describes how to use a textual large-language model (LLM) in the custom recommendation models. We train these models for you. You can enable the pretrained features in the [custom recommendation models](/retail/docs/models).\n\nRecommendations uses the product [`description`](/retail/docs/reference/rest/v2/projects.locations.catalogs.branches.products#Product.FIELDS.description) field to feed to LLMs and put them into your recommendations models.\n| **Note:** Creating and configuring models is only available for recommendations.\n\nNew LLM textual features\n------------------------\n\nWhile it's possible to get text embeddings by [manually configuring a Vertex AI generative model](/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings#get_text_embeddings_for_a_snippet_of_text), you might want to integrate the new LLM capabilities into your recommendations models to improve performance.\n\nThe text embeddings are more descriptive, longer, and are not repetitive, as well as have multilingual interpretation capabilities. This feature is based on an allowlist. Contact support for enabling this feature.\n\nThere's no charge for using the text embeddings and they are included in Vertex AI Search [pricing](/retail/docs/pricing).\n\nThe LLM-pretrained embeddings improve semantic understanding of long form text searches such as descriptions.\n\nSee the following resources for more information on how to use embeddings and generative AI alone in your own custom ML training:\n\n- [Vertex AI generative AI documentation](/retail/../generative-ai-app-builder/docs/about-checklists)\n- [Machine Learning Crash Course: Embeddings](https://developers.google.com/machine-learning/crash-course/embeddings)\n\nModel compatibility\n-------------------\n\nThe LLM feature is compatible with all ml model types and objectives, including:\n\n- OYML\n- FBT\n- and more.\n\nFor more information on the different types of recommendation models Vertex AI Search for commerce supports, see [About recommendations models](/retail/docs/models)."]]
