Hi,
I’ve used this https://huggingface.co/blog/getting-started-with-embeddings as reference to extract embeddings using the bigscience/bloom models.
Using the url: https://api-inference.huggingface.co/pipeline/feature-extraction/bigscience/bloom-3b the response is correct, returning a list of arrays. However when trying the bigscience/bloom model using the same request: https://api-inference.huggingface.co/pipeline/feature-extraction/bigscience/bloom, the response is [{'generated_text': '...'}] which appears to be a text-generation response.
Given the size of the model, I cannot check locally if the issue is with the model or the api endpoint. According to the documentation, both models should behave the same way as far as I can tell: https://huggingface.co/bigscience/bloom.
I think this is undesired behavior from this endpoint?