Partial revert #2851: allow task override on sentence-similarity (#2861)

Author: Wauplin

src/huggingface_hub/inference/_providers/hf_inference.py

@@ -38,7 +38,13 @@ def _prepare_url(self, api_key: str, mapped_model: str) -> str:
         # hf-inference provider can handle URLs (e.g. Inference Endpoints or TGI deployment)
         if mapped_model.startswith(("http://", "https://")):
             return mapped_model
-        return f"{self.base_url}/models/{mapped_model}"
+        return (
+            # Feature-extraction and sentence-similarity are the only cases where we handle models with several tasks.
+            f"{self.base_url}/pipeline/{self.task}/{mapped_model}"
+            if self.task in ("feature-extraction", "sentence-similarity")
+            # Otherwise, we use the default endpoint
+            else f"{self.base_url}/models/{mapped_model}"
+        )
 
     def _prepare_payload(self, inputs: Any, parameters: Dict, mapped_model: str) -> Optional[Dict]:
         if isinstance(inputs, bytes):

tests/cassettes/TestInferenceClient.test_sentence_similarity[hf-inference,sentence-similarity].yaml

@@ -17,7 +17,7 @@ interactions:
       X-Amzn-Trace-Id:
       - 0434ff33-56fe-49db-9380-17b81e41f756
     method: POST
-    uri: https://router.huggingface.co/hf-inference/models/sentence-transformers/all-MiniLM-L6-v2
+    uri: https://router.huggingface.co/hf-inference/pipeline/sentence-similarity/sentence-transformers/all-MiniLM-L6-v2
   response:
     body:
       string: '[0.7785724997520447,0.4587624967098236,0.29062220454216003]'

tests/cassettes/test_async_sentence_similarity.yaml

@@ -3,7 +3,7 @@ interactions:
     body: null
     headers: {}
     method: POST
-    uri: https://router.huggingface.co/hf-inference/models/sentence-transformers/all-MiniLM-L6-v2
+    uri: https://router.huggingface.co/hf-inference/pipeline/sentence-similarity/sentence-transformers/all-MiniLM-L6-v2
   response:
     body:
       string: '[0.7785724997520447,0.4587624967098236,0.29062220454216003]'