Revert "Revert "fix tests (#12286)""

This reverts commit 12f157513b.

litellm/llms/huggingface/chat/transformation.py

@@ -109,22 +109,19 @@ class HuggingFaceChatConfig(OpenAIGPTConfig):
         # Default construction with provider
         else:
             # Parse provider and model
+            complete_url = "https://router.huggingface.co/v1/chat/completions"
             first_part, remaining = model.split("/", 1)
             if "/" in remaining:
                 provider = first_part
-            else:
-                provider = "hf-inference"
-
-            if provider == "hf-inference":
-                route = f"{provider}/models/{model}/v1/chat/completions"
-            elif provider == "novita":
-                route = f"{provider}/v3/openai/chat/completions"
-            elif provider == "fireworks-ai":
-                route = f"{provider}/inference/v1/chat/completions"
-            else:
-                route = f"{provider}/v1/chat/completions"
-            complete_url = f"{BASE_URL}/{route}"
-
+                if provider == "hf-inference":
+                    route = f"{provider}/models/{model}/v1/chat/completions"
+                elif provider == "novita":
+                    route = f"{provider}/v3/openai/chat/completions"
+                elif provider == "fireworks-ai":
+                    route = f"{provider}/inference/v1/chat/completions"
+                else:
+                    route = f"{provider}/v1/chat/completions"
+                complete_url = f"{BASE_URL}/{route}"
         # Ensure URL doesn't end with a slash
         complete_url = complete_url.rstrip("/")
         return complete_url
@@ -145,25 +142,24 @@ class HuggingFaceChatConfig(OpenAIGPTConfig):
             logger.warning("`max_retries` is not supported. It will be ignored.")
             optional_params.pop("max_retries", None)
         first_part, remaining = model.split("/", 1)
+        mapped_model = model
         if "/" in remaining:
             provider = first_part
             model_id = remaining
-        else:
-            provider = "hf-inference"
-            model_id = model
-        provider_mapping = _fetch_inference_provider_mapping(model_id)
-        if provider not in provider_mapping:
-            raise HuggingFaceError(
-                message=f"Model {model_id} is not supported for provider {provider}",
-                status_code=404,
-                headers={},
-            )
-        provider_mapping = provider_mapping[provider]
-        if provider_mapping["status"] == "staging":
-            logger.warning(
-                f"Model {model_id} is in staging mode for provider {provider}. Meant for test purposes only."
-            )
-        mapped_model = provider_mapping["providerId"]
+            provider_mapping = _fetch_inference_provider_mapping(model_id)
+            if provider not in provider_mapping:
+                raise HuggingFaceError(
+                    message=f"Model {model_id} is not supported for provider {provider}",
+                    status_code=404,
+                    headers={},
+                )
+            provider_mapping = provider_mapping[provider]
+            if provider_mapping["status"] == "staging":
+                logger.warning(
+                    f"Model {model_id} is in staging mode for provider {provider}. Meant for test purposes only."
+                )
+            mapped_model = provider_mapping["providerId"]
+
         messages = self._transform_messages(messages=messages, model=mapped_model)
         return dict(
             ChatCompletionRequest(

tests/llm_translation/test_huggingface_chat_completion.py

@@ -2,19 +2,21 @@
 Test HuggingFace LLM
 """
 
-from base_llm_unit_tests import BaseLLMChatTest
 import json
 import os
 import sys
-from unittest.mock import patch, MagicMock, AsyncMock
+from unittest.mock import AsyncMock, MagicMock, patch
+
+from base_llm_unit_tests import BaseLLMChatTest
 
 sys.path.insert(
     0, os.path.abspath("../..")
 )  # Adds the parent directory to the system path
 
-import litellm
 import pytest
-from litellm.types.utils import ModelResponseStream, ModelResponse
+
+import litellm
+from litellm.types.utils import ModelResponse, ModelResponseStream
 
 MOCK_COMPLETION_RESPONSE = {
     "id": "9115d3daeab10608",
@@ -361,31 +363,27 @@ class TestHuggingFace(BaseLLMChatTest):
             )
 
     @pytest.mark.parametrize(
-        "model, provider, expected_url",
+        "model, expected_url",
         [
             (
                 "meta-llama/Llama-3-8B-Instruct",
-                None,
-                "https://router.huggingface.co/hf-inference/models/meta-llama/Llama-3-8B-Instruct/v1/chat/completions",
+                "https://router.huggingface.co/v1/chat/completions",
             ),
             (
                 "together/meta-llama/Llama-3-8B-Instruct",
-                None,
                 "https://router.huggingface.co/together/v1/chat/completions",
             ),
             (
                 "novita/meta-llama/Llama-3-8B-Instruct",
-                None,
                 "https://router.huggingface.co/novita/v3/openai/chat/completions",
             ),
             (
                 "http://custom-endpoint.com/v1/chat/completions",
-                None,
                 "http://custom-endpoint.com/v1/chat/completions",
             ),
         ],
     )
-    def test_get_complete_url(self, model, provider, expected_url):
+    def test_get_complete_url(self, model, expected_url):
         """Test that the complete URL is constructed correctly for different providers"""
         from litellm.llms.huggingface.chat.transformation import HuggingFaceChatConfig
 

tests/local_testing/test_text_completion.py

@@ -3975,7 +3975,7 @@ def test_text_completion_stream():
     try:
         for _ in range(2):  # check if closed client used
             response = text_completion(
-                model="huggingface/sarvamai/sarvam-m",
+                model="huggingface/deepseek-ai/DeepSeek-R1",
                 prompt="good morning",
                 stream=True,
                 max_tokens=10,