Import a language model into Foundry [Sunset]
The language model import functionality described on this page is in the sunset phase of development and will be deprecated at a future date. The list of supported models will not be extended and Hugging Face language models will not support more recent features, such as direct deployments. As an alternative, we recommend using one of the LLMs natively supported by Palantir AIP or importing model files from Hugging Face instead.
Foundry allows you to import open-source language models to your Foundry workflows with no code. These language models can be submitted to a modeling objective for use in batch and live deployments.
To use other language models, you can also easily add support for a new language model. For details on the model adapters used below, see the documentation on language model adapters.
Import a language model
A supported open-source model can be integrated without writing any code by following the steps below:
- Create a new or open an existing modeling objective to which you will add the language model. In the modeling objective, select Add model.
- Select the option Import an open-source model.
- Select one of the available models, then select Next.
-
Choose where the created model resource will be saved in Foundry.
Foundry provides a default model adapter for available open-source language models. However, you can also configure a custom model adapter.
The usage instructions for each language model will depend on the specific model adapter used.
Learn more about the model adapters that Foundry provides for the above open-source language models.