support for CPU/GPU choice and initialization before starting the app

[Vokturz]

• The device can be defined by env variable DEVICE
• created an initialize_embeddings function which initialize the embeddings before starting the app

[limcheekin]

Thanks for send me the PR.

It make sense to load the text embeddings model before the first request. I leave the code of loading the embedding model on-demand back then due to I haven't make up my mind should the API serve multiple models or single model. I think it is clear now that the API serve single text embeddings model is the way to go, the code is simpler. If another text embeddings model is needed, the user can simply deploy another instance of the API.

Hence, the model parameter no longer needed here:

Suggested change

	def initialize_embeddings(model: Optional[str] = None):
	def initialize_embeddings():

[limcheekin]

I will move the initialize_embeddings() to create_app() as the current approach will break the code of aws.py.

[limcheekin]

I will merge the PR and make some minor changes as per my comments.

[limcheekin]

Just release the package at https://pypi.org/project/open-text-embeddings/.

Kindly let's me know if you think there are ways to improve it. This is the first python package I publish to pypi.