`HuggingFaceAPIGenerator` is not working

[anakin87]

Failing tests: https://github.com/deepset-ai/haystack/actions/runs/16038492313/job/45255328842

FAILED test/components/generators/test_hugging_face_api.py::TestHuggingFaceAPIGenerator::test_run_serverless - huggingface_hub.errors.BadRequestError: (Request ID: Root=1-6865cc56-4ecddb912cf025587594bc7e;5e267090-bcd9-4192-bf25-338e1f1f71ba)

Bad Request: The endpoint is paused, ask a maintainer to restart it

Investigation

This error occurs because Text generation models are no longer available via HF Inference API: https://huggingface.co/models?pipeline_tag=text-generation&inference_provider=hf-inference&sort=downloads
Related thread on HF forum

Attempts to switch to another inference provider fail because text_generation is only supported by the HF Inference API. See huggingface/huggingface_hub#3196.

Next steps

• We can initially skip the failing tests to not block PRs
• If Text generation models become available again on HF Inference API, we can revert and proceed as before
• Otherwise, we might use chat_completion method instead of text_generation in the HuggingFaceAPIGenerator: this would mean changing the behavior of the component and no longer allowing raw prompts (like <|user|>\nWhat is the capital of France?<|end|>\n<|assistant|>\n)

[anakin87]

It now seems that also embedding models are no longer available on HF Inference API.

[anakin87]

We temporarily skipped related tests in #9575.
I'll keep the situation monitored to understand if it makes sense to change our existing components.

[anakin87]

Almost official news from HF DevRel on Discord

yes! these models were sunset as part of us closing down hugging.chat unfortunately
however there's quite a lot of models that you can use through our inference providers as a replacement