Skip to content

Fireworks AI

With Sibila you can use the models hosted by Fireworks AI, for which you'll need an API key (which is initially free). As in other providers, although you can pass this key when you create the model object, it's more secure to define an env variable with this information:

export FIREWORKS_API_KEY="..."

setx FIREWORKS_API_KEY "..."

Another possibility is to store your API key in .env files, which has many advantages: see the dotenv-python package.

Creating models#

Models served by Fireworks AI can be used by Sibila through the FireworksModel class.

Example

from sibila import FireworksModel

model = FireworksModel("accounts/fireworks/models/gemma-7b-it")

model("I think that I shall never see.")

Result (model is hallucinating)

The poem "I think that I shall never see" is a poem by William Blake. 
It is a poem about the loss of sight. The speaker is saying that they 
will never be able to see again. The poem is a reflection on the beauty 
of sight and the sadness of blindness.

You can also create a Fireworks AI model in the Models factory by using the "fireworks:" provider prefix:

from sibila import Models

model = Models.create("fireworks:accounts/fireworks/models/gemma-7b-it")

Model list#

The available Fireworks text inference models models are listed here.

Unfortunately Fireworks AI doesn't provide an API to list the models, so FireworksModel.known_models() will return None.

JSON Schema models#

All the Fireworks AI models should support JSON Schema generation, which is required for structured data extraction.