HuggingFace Provider

[rgr]

Hi,

as a user I would benefit having HuggingFace as a standard provider. There are many models we could use there and switch easily. Would this feature be able to make it to the next release ?

Thanks !

Raphaël

[ilvalerione]

Hi @rgr thank you for sharing your request. It's not on the roadmap for the short term because the focus now is on let developers explore all the possible interactions with the current structure of the framework, and fix emerging architectural week points and bugs.

[ilvalerione]

You can vote here #26

[ilvalerione]

For what I can read in the HF documentation their API are OpenAI compatible. You can choose one of the provider supported by HuggingFace and talk with the model using the appropriate endpoint.

https://huggingface.co/docs/inference-providers/tasks/chat-completion#code-snippet-example-for-conversational-llms

That means you can create the HuggingFace class just extending the default OpenAI provider class:

use NeuronAI\Providers\OpenAI\OpenAI;

class HuggingFace extends OpenAI 
{
    protected string $baseUri = "https://router.huggingface.co/together/v1";
}

The example above supposes you use HF with Together AI.

You can use it in your agent:

class MyAgent extends Agent 
{
    public function provider(): AIProviderInterface
    {
        return new HuggingFace(
            key: 'HF_API_KEY',
            model: 'HF_MODEL'
        );
    }
}

Let me know if it works.