[Inference Providers] check inference provider mapping for HF Inference API (#2948)

* check inference provider mapping for hf inference

* custom check for supported tasks for hf inference

* nit

* comments

* re-record cassettes

* fix tests

* Apply suggestion

Co-authored-by: Lucain <[email protected]>

* Update src/huggingface_hub/inference/_providers/hf_inference.py

---------

Co-authored-by: Lucain <[email protected]>

Author: hanouticelina

src/huggingface_hub/inference/_client.py

@@ -1286,7 +1286,7 @@ def image_segmentation(
         [ImageSegmentationOutputElement(score=0.989008, label='LABEL_184', mask=<PIL.PngImagePlugin.PngImageFile image mode=L size=400x300 at 0x7FDD2B129CC0>), ...]
         ```
         """
-        provider_helper = get_provider_helper(self.provider, task="audio-classification")
+        provider_helper = get_provider_helper(self.provider, task="image-segmentation")
         request_parameters = provider_helper.prepare_request(
             inputs=image,
             parameters={

src/huggingface_hub/inference/_generated/_async_client.py

@@ -1331,7 +1331,7 @@ async def image_segmentation(
         [ImageSegmentationOutputElement(score=0.989008, label='LABEL_184', mask=<PIL.PngImagePlugin.PngImageFile image mode=L size=400x300 at 0x7FDD2B129CC0>), ...]
         ```
         """
-        provider_helper = get_provider_helper(self.provider, task="audio-classification")
+        provider_helper = get_provider_helper(self.provider, task="image-segmentation")
         request_parameters = provider_helper.prepare_request(
             inputs=image,
             parameters={

src/huggingface_hub/inference/_providers/hf_inference.py

@@ -24,15 +24,16 @@ def _prepare_api_key(self, api_key: Optional[str]) -> str:
         return api_key or get_token()  # type: ignore[return-value]
 
     def _prepare_mapped_model(self, model: Optional[str]) -> str:
-        if model is not None:
+        if model is not None and model.startswith(("http://", "https://")):
             return model
-        model = _fetch_recommended_models().get(self.task)
-        if model is None:
+        model_id = model if model is not None else _fetch_recommended_models().get(self.task)
+        if model_id is None:
             raise ValueError(
                 f"Task {self.task} has no recommended model for HF Inference. Please specify a model"
                 " explicitly. Visit https://huggingface.co/tasks for more info."
             )
-        return model
+        _check_supported_task(model_id, self.task)
+        return model_id
 
     def _prepare_url(self, api_key: str, mapped_model: str) -> str:
         # hf-inference provider can handle URLs (e.g. Inference Endpoints or TGI deployment)
@@ -120,3 +121,39 @@ def _fetch_recommended_models() -> Dict[str, Optional[str]]:
     response = get_session().get(f"{constants.ENDPOINT}/api/tasks", headers=build_hf_headers())
     hf_raise_for_status(response)
     return {task: next(iter(details["widgetModels"]), None) for task, details in response.json().items()}
+
+
+@lru_cache(maxsize=None)
+def _check_supported_task(model: str, task: str) -> None:
+    from huggingface_hub.hf_api import HfApi
+
+    model_info = HfApi().model_info(model)
+    pipeline_tag = model_info.pipeline_tag
+    tags = model_info.tags or []
+    is_conversational = "conversational" in tags
+    if task in ("text-generation", "conversational"):
+        if pipeline_tag == "text-generation":
+            # text-generation + conversational tag -> both tasks allowed
+            if is_conversational:
+                return
+            # text-generation without conversational tag -> only text-generation allowed
+            if task == "text-generation":
+                return
+            raise ValueError(f"Model '{model}' doesn't support task '{task}'.")
+
+    if pipeline_tag == "text2text-generation":
+        if task == "text-generation":
+            return
+        raise ValueError(f"Model '{model}' doesn't support task '{task}'.")
+
+    if pipeline_tag == "image-text-to-text":
+        if is_conversational and task == "conversational":
+            return  # Only conversational allowed if tagged as conversational
+        raise ValueError("Non-conversational image-text-to-text task is not supported.")
+
+    # For all other tasks, just check pipeline tag
+    if pipeline_tag != task:
+        raise ValueError(
+            f"Model '{model}' doesn't support task '{task}'. Supported tasks: '{pipeline_tag}', got: '{task}'"
+        )
+    return

tests/cassettes/TestInferenceClient.test_audio_classification[hf-inference,audio-classification].yaml

@@ -1,4 +1,74 @@
 interactions:
+- request:
+    body: null
+    headers:
+      Accept:
+      - '*/*'
+      Accept-Encoding:
+      - gzip, deflate
+      Connection:
+      - keep-alive
+      X-Amzn-Trace-Id:
+      - bb0168cc-7f1c-4d14-b411-41f64f5cc7b1
+    method: GET
+    uri: https://huggingface.co/api/models/jonatasgrosman/wav2vec2-large-xlsr-53-english
+  response:
+    body:
+      string: '{"_id":"621ffdc136468d709f17cdb3","id":"jonatasgrosman/wav2vec2-large-xlsr-53-english","private":false,"pipeline_tag":"automatic-speech-recognition","library_name":"transformers","tags":["transformers","pytorch","jax","safetensors","wav2vec2","automatic-speech-recognition","audio","en","hf-asr-leaderboard","mozilla-foundation/common_voice_6_0","robust-speech-event","speech","xlsr-fine-tuning-week","dataset:common_voice","dataset:mozilla-foundation/common_voice_6_0","doi:10.57967/hf/3569","license:apache-2.0","model-index","endpoints_compatible","region:us"],"downloads":18234759,"likes":466,"modelId":"jonatasgrosman/wav2vec2-large-xlsr-53-english","author":"jonatasgrosman","sha":"569a6236e92bd5f7652a0420bfe9bb94c5664080","lastModified":"2023-03-25T10:56:55.000Z","gated":false,"disabled":false,"model-index":[{"name":"XLSR
+        Wav2Vec2 English by Jonatas Grosman","results":[{"task":{"name":"Automatic
+        Speech Recognition","type":"automatic-speech-recognition"},"dataset":{"name":"Common
+        Voice en","type":"common_voice","args":"en"},"metrics":[{"name":"Test WER","type":"wer","value":19.06,"verified":false},{"name":"Test
+        CER","type":"cer","value":7.69,"verified":false},{"name":"Test WER (+LM)","type":"wer","value":14.81,"verified":false},{"name":"Test
+        CER (+LM)","type":"cer","value":6.84,"verified":false}]},{"task":{"name":"Automatic
+        Speech Recognition","type":"automatic-speech-recognition"},"dataset":{"name":"Robust
+        Speech Event - Dev Data","type":"speech-recognition-community-v2/dev_data","args":"en"},"metrics":[{"name":"Dev
+        WER","type":"wer","value":27.72,"verified":false},{"name":"Dev CER","type":"cer","value":11.65,"verified":false},{"name":"Dev
+        WER (+LM)","type":"wer","value":20.85,"verified":false},{"name":"Dev CER (+LM)","type":"cer","value":11.01,"verified":false}]}]}],"config":{"architectures":["Wav2Vec2ForCTC"],"model_type":"wav2vec2"},"cardData":{"language":"en","datasets":["common_voice","mozilla-foundation/common_voice_6_0"],"metrics":["wer","cer"],"tags":["audio","automatic-speech-recognition","en","hf-asr-leaderboard","mozilla-foundation/common_voice_6_0","robust-speech-event","speech","xlsr-fine-tuning-week"],"license":"apache-2.0","model-index":[{"name":"XLSR
+        Wav2Vec2 English by Jonatas Grosman","results":[{"task":{"name":"Automatic
+        Speech Recognition","type":"automatic-speech-recognition"},"dataset":{"name":"Common
+        Voice en","type":"common_voice","args":"en"},"metrics":[{"name":"Test WER","type":"wer","value":19.06,"verified":false},{"name":"Test
+        CER","type":"cer","value":7.69,"verified":false},{"name":"Test WER (+LM)","type":"wer","value":14.81,"verified":false},{"name":"Test
+        CER (+LM)","type":"cer","value":6.84,"verified":false}]},{"task":{"name":"Automatic
+        Speech Recognition","type":"automatic-speech-recognition"},"dataset":{"name":"Robust
+        Speech Event - Dev Data","type":"speech-recognition-community-v2/dev_data","args":"en"},"metrics":[{"name":"Dev
+        WER","type":"wer","value":27.72,"verified":false},{"name":"Dev CER","type":"cer","value":11.65,"verified":false},{"name":"Dev
+        WER (+LM)","type":"wer","value":20.85,"verified":false},{"name":"Dev CER (+LM)","type":"cer","value":11.01,"verified":false}]}]}]},"transformersInfo":{"auto_model":"AutoModelForCTC","pipeline_tag":"automatic-speech-recognition","processor":"AutoProcessor"},"siblings":[{"rfilename":".gitattributes"},{"rfilename":"README.md"},{"rfilename":"alphabet.json"},{"rfilename":"config.json"},{"rfilename":"eval.py"},{"rfilename":"flax_model.msgpack"},{"rfilename":"full_eval.sh"},{"rfilename":"language_model/attrs.json"},{"rfilename":"language_model/lm.binary"},{"rfilename":"language_model/unigrams.txt"},{"rfilename":"log_mozilla-foundation_common_voice_6_0_en_test_predictions.txt"},{"rfilename":"log_mozilla-foundation_common_voice_6_0_en_test_predictions_greedy.txt"},{"rfilename":"log_mozilla-foundation_common_voice_6_0_en_test_targets.txt"},{"rfilename":"log_speech-recognition-community-v2_dev_data_en_validation_predictions.txt"},{"rfilename":"log_speech-recognition-community-v2_dev_data_en_validation_predictions_greedy.txt"},{"rfilename":"log_speech-recognition-community-v2_dev_data_en_validation_targets.txt"},{"rfilename":"model.safetensors"},{"rfilename":"mozilla-foundation_common_voice_6_0_en_test_eval_results.txt"},{"rfilename":"mozilla-foundation_common_voice_6_0_en_test_eval_results_greedy.txt"},{"rfilename":"preprocessor_config.json"},{"rfilename":"pytorch_model.bin"},{"rfilename":"special_tokens_map.json"},{"rfilename":"speech-recognition-community-v2_dev_data_en_validation_eval_results.txt"},{"rfilename":"speech-recognition-community-v2_dev_data_en_validation_eval_results_greedy.txt"},{"rfilename":"vocab.json"}],"spaces":["bertin-project/bertin-gpt-j-6B","Gradio-Blocks/Alexa-NLU-Clone","qanastek/Alexa-NLU-Clone","Gradio-Blocks/poor-mans-duplex","awacke1/ASR-High-Accuracy-Test","trysem/Spleeter_and_ASR","Detomo/audio-stream-translate","yashsrivastava/speech-to-text-yash","RealTimeLiveAIForHealth/ASR-High-Accuracy-Test","manmeetkaurbaxi/YouTube-Video-Summarizer","baaastien/Spleeter_and_ASR","GeekedReals/jonatasgrosman-wav2vec2-large-xlsr-53-english","Charles95/gradio-tasks","Gna1L/jonatasgrosman-wav2vec2-large-xlsr-53-english","Mintiny/Customer_Review_Audio_Analysis","awacke1/STT-TTS-ASR-AI-NLP-Pipeline","Detomo/audio-translate","Amrrs/yt-video-summarizer","JPLTedCas/TedCasSpeechRecognition","stanciu/jonatasgrosman-wav2vec2-large-xlsr-53-english","raunak627887/jonatasgrosman-wav2vec2-large-xlsr-53-english","Hrsh-Venket/Corrected-Speech-to-Text","Rhyolite/jonatasgrosman-wav2vec2-large-xlsr-53-english","Grepper/jonatasgrosman-wav2vec2-large-xlsr-53-english","Yarumo/jonatasgrosman-wav2vec2-large-xlsr-53-english","jbraun19/ASR-High-Accuracy-Test","sankalphimself/pitchpal","Rajab123/jonatasgrosman-wav2vec2-large-xlsr-53-english","SteeleN1/jonatasgrosman-wav2vec2-large-xlsr-53-english","melazab1/jonatasgrosman-wav2vec2-large-xlsr-53-english","amaamas/jonatasgrosman-wav2vec2-large-xlsr-53-english","Gearijigu/jonatasgrosman-wav2vec2-large-xlsr-53-english","sebasjm/jonatasgrosman-wav2vec2-large-xlsr-53-english","codetopolymath/jonatasgrosman-wav2vec2-large-xlsr-53-english","lingdai/jonatasgrosman-wav2vec2-large-xlsr-53-english","nabdtran/jonatasgrosman-wav2vec2-large-xlsr-53-english","NourAlmolhem/jonatasgrosman-wav2vec2-large-xlsr-53-english","leetik/jonatasgrosman-wav2vec2-large-xlsr-53-english","mastere00/jonatasgrosman-wav2vec2-large-xlsr-53-english","pragyachik/jonatasgrosman-wav2vec2-large-xlsr-53-english","shubhsnow/jonatasgrosman-wav2vec2-large-xlsr-53-english","kushiel/jonatasgrosman-wav2vec2-large-xlsr-53-english","adarsh8986/jonatasgrosman-wav2vec2-large-xlsr-53-english","Baghdad99/eng-to-hau","Akkaris/jonatasgrosman-wav2vec2-large-xlsr-53-english","Baghdad99/english-to-hausa","anonymous4me/jonatasgrosman-wav2vec2-large-xlsr-53-english","LEWOPO/Voice_to_Text","Nikhil0987/speechrecho","neridonk/jonatasgrosman-wav2vec2-large-xlsr-53-english","ganeshkamath89/gradio-huggingface-pipeline-tasks-demo-all","quangnhan145/jonatasgrosman-wav2vec2-large-xlsr-53-english-demo-app","niveone/jonatasgrosman-wav2vec2-large-xlsr-53-english","dincali/jonatasgrosman-wav2vec2-large-xlsr-53-english","oryxsoftware/speech-to-text","Shashwat2528/Avishkaarak-ekta-new-audio","AVISHKAARAM/avishkarak-ekta-audio","IES-Rafael-Alberti/AudioToText","Mahmoud2020220/jonatasgrosman-wav2vec2-large-xlsr-53-english","saronium/jonatasgrosman-wav2vec2-large-xlsr-53-english","trizzzy/jonatasgrosman-wav2vec2-large-xlsr-53-english","guna-entrans/jonatasgrosman-wav2vec2-large-xlsr-53-english","pmiguelpds/jonatasgrosman-wav2vec2-large-xlsr-53-english","LordCoffee/transcript","manohar025/video-summarizer","maan2605/Youtube_Video_Summarizer_using_ASR","ABIDFAYAZ/meeting-transcription","derrideanlils/jonatasgrosman-wav2vec2-large-xlsr-53-english","saud-altuwaijri/demo4","Durganihantri/AI-Child-Behavior-Assessment","65rted6tfdjhgfjyrf/Gibberish-transcribr","aicodingfun/Alexa-NLU-Clone"],"createdAt":"2022-03-02T23:29:05.000Z","safetensors":{"parameters":{"F32":315472545},"total":315472545},"inference":"warm","usedStorage":7177619310}'
+    headers:
+      Access-Control-Allow-Origin:
+      - https://huggingface.co
+      Access-Control-Expose-Headers:
+      - X-Repo-Commit,X-Request-Id,X-Error-Code,X-Error-Message,X-Total-Count,ETag,Link,Accept-Ranges,Content-Range,X-Xet-Access-Token,X-Xet-Token-Expiration,X-Xet-Refresh-Route,X-Xet-Cas-Url,X-Xet-Hash
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '8113'
+      Content-Type:
+      - application/json; charset=utf-8
+      Date:
+      - Fri, 21 Mar 2025 11:29:30 GMT
+      ETag:
+      - W/"1fb1-kJSNr/PIoqRzIeGGFb80dxk+xKs"
+      Referrer-Policy:
+      - strict-origin-when-cross-origin
+      Vary:
+      - Origin
+      Via:
+      - 1.1 02ee9ebd8a83522edf11335f04975776.cloudfront.net (CloudFront)
+      X-Amz-Cf-Id:
+      - lUVe7zu2SXod2amCAy6NI4WliBYareXEaVPdf4NCasTTA5C5P_dj-Q==
+      X-Amz-Cf-Pop:
+      - CDG52-P4
+      X-Cache:
+      - Miss from cloudfront
+      X-Powered-By:
+      - huggingface-moon
+      X-Request-Id:
+      - Root=1-67dd4d9a-2b095a0137b6d23f0720dd7b;bb0168cc-7f1c-4d14-b411-41f64f5cc7b1
+      cross-origin-opener-policy:
+      - same-origin
+    status:
+      code: 200
+      message: OK
 - request:
     body: !!binary |
       ZkxhQwAAACIQABAAAAGOABkSA+gA8AABIqDeyGtH35+dVPygWCFAFS5GAwAAEgAAAAAAAAAAAAAA
@@ -1666,7 +1736,7 @@ interactions:
       Content-Length:
       - '94321'
       X-Amzn-Trace-Id:
-      - 3d462b35-26a7-4c68-959b-f6a7d0301caf
+      - d3d22a44-b253-43a2-8f4f-6a20b5aee642
     method: POST
     uri: https://router.huggingface.co/hf-inference/models/jonatasgrosman/wav2vec2-large-xlsr-53-english
   response:
@@ -1675,40 +1745,48 @@ interactions:
     headers:
       Access-Control-Allow-Origin:
       - '*'
-      Access-Control-Expose-Headers:
-      - X-Repo-Commit,X-Request-Id,X-Error-Code,X-Error-Message,X-Total-Count,ETag,Link,Accept-Ranges,Content-Range,X-Xet-Access-Token,X-Xet-Token-Expiration,X-Xet-Refresh-Route,X-Xet-Cas-Url,X-Xet-Hash
       Connection:
       - keep-alive
-      Content-Length:
-      - '49'
       Content-Type:
       - application/json
       Date:
-      - Thu, 13 Feb 2025 15:49:08 GMT
+      - Fri, 21 Mar 2025 11:29:31 GMT
       Referrer-Policy:
       - strict-origin-when-cross-origin
+      Transfer-Encoding:
+      - chunked
       Via:
-      - 1.1 fba88aca5f6c32c13257a59071f94248.cloudfront.net (CloudFront)
+      - 1.1 6676a739f016238678e391e91007cc98.cloudfront.net (CloudFront)
       X-Amz-Cf-Id:
-      - Ae6o7enRbofgvT3JVME57ywv8zStyC4fsspyhh_4QrpEvLi6ZYwLWA==
+      - hu8tduAfzg-IW6ZBolY67qPKbBnahkp3CYwuCBoJkR3u3Wn2BHygvQ==
       X-Amz-Cf-Pop:
       - CDG55-P3
       X-Cache:
       - Miss from cloudfront
       X-Powered-By:
       - huggingface-moon
+      X-Robots-Tag:
+      - none
       access-control-allow-credentials:
       - 'true'
+      access-control-expose-headers:
+      - x-compute-type, x-compute-time
       cross-origin-opener-policy:
       - same-origin
+      server:
+      - uvicorn
       vary:
       - Origin, Access-Control-Request-Method, Access-Control-Request-Headers
+      x-compute-audio-length:
+      - '4.65'
       x-compute-time:
-      - '0.638'
+      - '0.712'
       x-compute-type:
-      - cache
+      - cpu
+      x-inference-id:
+      - EUm9ZEn7NXWqjizXKR6bM
       x-request-id:
-      - Root=1-67ae1474-43a45db64b62db6251089c0d;3d462b35-26a7-4c68-959b-f6a7d0301caf
+      - Root=1-67dd4d9a-28cb8d3e43da94102f4cd965;d3d22a44-b253-43a2-8f4f-6a20b5aee642
       x-sha:
       - 569a6236e92bd5f7652a0420bfe9bb94c5664080
     status: