FAQs

🙋🏻 Common questions about Gradient

What is Gradient?

Gradient is a developer API that allows you to customize open-source Large Language Models (LLMs) for your needs. We host the models so you can easily access, fine-tune, and deploy the model for your use case.

How is Gradient different from other products for training models?

Gradient is the only developer API that gives you easy access to state-of-the-art open-source LLMs and allows you to run inference and fine-tune these models. You have full ownership of private instances of base models and the model weights generated in the fine-tuning process.

How are you able to guarantee privacy?

When you create a model in a Gradient workspace no one outside of the workspace can access it. You have full ownership and control over your data and model.

We also hold ourselves to the highest enterprise security standards, taking best practices from the industry (i.e., from experience at Google) and applying them here.

Are you SOC 2 certified?

Yes, we are SOC 2 certified as of August 2023. We know this is important to developers with high data security requirements, and we have prioritized security early.

🙋🏽‍♀️ Common questions about using the Gradient platform

What models are available on Gradient?

You can see the list of available models here. We are actively working to host more open-source models for you to fine-tune.

What is a model adapter?

In the Gradient platform, when you are preparing to fine-tune a base model, you first create a model adapter. An adapter is a private instance of the model that you own and are able to fine-tune.

We can further classify these model adapters:

• A Type 1 adapter is a private instance of the base model without fine-tuning
• A Type 2 adapter is a fine-tuned version of a Type 1 adapter

What's your maximum content length?

Per sample limits:

What are you default completion generation settings?

By default, our maximum token length is set to 20 and our temperature is set to 0. You can learn more about these parameters in the Tips & Tricks page.

Do you have versioning?

We do not currently have model versioning. We are actively working on supporting this to make it easier to use your models in production.

Can I download the model and host it locally?

The models are currently only available via API. Gradient hosts the models so that you have perpetual access to inference on demand, so the models are not yet easily exportable.

Who do you integrate with?

We are currently working on integrations with LangChain, LlamaIndex, MongoDB, and Haystack.