fix: make test_watsonx_gpt_oss_prompt_transformation deterministic (#17597)

- Remove network dependency by mocking HuggingFace template fetch
- Use mock template that produces correct format for test validation
- Test now focuses on transformation logic, not network calls
- Fixes flaky test failures due to network timeouts/rate limits

The test verifies that prompt transformation occurs (not simple
concatenation), which doesn't require the actual HuggingFace template.
Mocking makes the test deterministic and faster while still validating
the core behavior.

tests/test_litellm/llms/watsonx/test_watsonx.py

@@ -255,8 +255,41 @@ async def test_watsonx_gpt_oss_prompt_transformation(monkeypatch):
         "model_id": "openai/gpt-oss-120b",
     }
 
+    # Mock HuggingFace template fetch to make test deterministic and avoid network flakiness.
+    # The test verifies that prompt transformation occurs (not simple concatenation), not the exact
+    # HuggingFace template format. Using a mock template that produces the correct format is sufficient.
+    from unittest.mock import patch
+
+    # Mock template that produces gpt-oss-120b-like format.
+    # Note: This is a simplified version of the actual template. The real template is more complex
+    # (adds metadata, handles tools, thinking messages, etc.), but this captures the key aspects:
+    # - Converts system role to developer (matching real template behavior)
+    # - Uses the same tag structure (<|start|>, <|message|>, <|end|>)
+    # - Preserves message content
+    mock_tokenizer_config = {
+        "status": "success",
+        "tokenizer": {
+            "chat_template": "{% for message in messages %}{% if message['role'] == 'system' %}<|start|>developer<|message|>{% else %}<|start|>{{ message['role'] }}<|message|>{% endif %}{{ message['content'] }}<|end|>{% endfor %}",
+            "bos_token": None,
+            "eos_token": None,
+        },
+    }
+
+    async def mock_aget_tokenizer_config(hf_model_name: str):
+        return mock_tokenizer_config
+
+    async def mock_aget_chat_template_file(hf_model_name: str):
+        # Return failure to use tokenizer_config instead
+        return {"status": "failure"}
+
     with patch.object(client, "post") as mock_post, patch.object(
         litellm.module_level_client, "post", return_value=mock_token_response
+    ), patch(
+        "litellm.litellm_core_utils.prompt_templates.huggingface_template_handler._aget_tokenizer_config",
+        side_effect=mock_aget_tokenizer_config,
+    ), patch(
+        "litellm.litellm_core_utils.prompt_templates.huggingface_template_handler._aget_chat_template_file",
+        side_effect=mock_aget_chat_template_file,
     ):
         # Set the mock to return the completion response
         mock_post.return_value = mock_completion_response
@@ -298,6 +331,12 @@ async def test_watsonx_gpt_oss_prompt_transformation(monkeypatch):
     assert "input" in json_data, "Request should have 'input' field"
     transformed_prompt = json_data["input"]
 
+    # Verify transformation occurred
+    assert transformed_prompt is not None, (
+        "Prompt transformation failed - the template should have been applied to transform "
+        "messages into the correct format for gpt-oss-120b."
+    )
+
     print(f"Transformed prompt: {repr(transformed_prompt)}")
     print(f"Prompt length: {len(transformed_prompt)}")