diff --git a/tests/test_litellm/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py b/tests/test_litellm/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py deleted file mode 100644 index e269e782061..00000000000 --- a/tests/test_litellm/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py +++ /dev/null @@ -1,337 +0,0 @@ -""" -Tests for IBM WatsonX Audio Transcription. - -Validates that litellm.transcription transforms requests correctly for WatsonX. -""" - -import json -from unittest.mock import AsyncMock, MagicMock, patch - -import pytest - - -import litellm -from litellm.llms.watsonx.audio_transcription.transformation import ( - IBMWatsonXAudioTranscriptionConfig, -) -from litellm.types.utils import TranscriptionResponse - - -class TestWatsonXAudioTranscription: - """Tests for WatsonX audio transcription via litellm.transcription.""" - - @pytest.mark.asyncio - async def test_watsonx_transcription_url_and_headers(self): - """ - Test that litellm.transcription sends request to correct WatsonX URL with proper headers. - """ - captured_request = {} - - async def mock_post(*args, **kwargs): - captured_request["url"] = str(kwargs.get("url", args[0] if args else None)) - captured_request["headers"] = kwargs.get("headers", {}) - captured_request["data"] = kwargs.get("data", {}) - captured_request["files"] = kwargs.get("files", {}) - - mock_response = MagicMock() - mock_response.json.return_value = { - "text": "test transcription", - "duration": 1.0, - } - mock_response.status_code = 200 - return mock_response - - with patch( - "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post", - new=mock_post, - ): - try: - await litellm.atranscription( - model="watsonx/whisper-large-v3-turbo", - file=b"fake_audio_data", - api_base="https://us-south.ml.cloud.ibm.com", - api_key="test-api-key", - project_id="test-project-123", - token="test-bearer-token", - ) - except Exception: - pass # We just want to capture the request - - # Validate URL contains WatsonX audio transcription endpoint - assert "/ml/v1/audio/transcriptions" in captured_request["url"] - assert "version=" in captured_request["url"] - # project_id should NOT be in URL (it should be in form data instead) - assert "project_id=test-project-123" not in captured_request["url"] - - # Validate headers contain WatsonX auth - assert "Authorization" in captured_request["headers"] - assert ( - "Bearer test-bearer-token" in captured_request["headers"]["Authorization"] - ) - - # Validate Content-Type is NOT set (httpx sets multipart/form-data automatically) - assert "Content-Type" not in captured_request["headers"] - - # Validate project_id is in form data, not URL - assert captured_request["data"].get("project_id") == "test-project-123" - - # Validate file is in files dict - assert "file" in captured_request["files"] - - @pytest.mark.asyncio - async def test_watsonx_transcription_request_body(self): - """ - Test that litellm.transcription sends correct request body for WatsonX. - - Validates that: - - Request uses multipart/form-data (data + files) - - Model name has watsonx/ prefix removed - - project_id is in form data, not URL - - Audio file is in files dict - - OpenAI params are included in form data - """ - captured_request = {} - - async def mock_post(*args, **kwargs): - captured_request["data"] = kwargs.get("data", {}) - captured_request["files"] = kwargs.get("files", {}) - - mock_response = MagicMock() - mock_response.json.return_value = { - "text": "test transcription", - "duration": 1.0, - } - mock_response.status_code = 200 - return mock_response - - with patch( - "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post", - new=mock_post, - ): - try: - await litellm.atranscription( - model="watsonx/whisper-large-v3-turbo", - file=b"fake_audio_data", - api_base="https://us-south.ml.cloud.ibm.com", - api_key="test-api-key", - project_id="test-project-123", - token="test-bearer-token", - language="en", - temperature=0.5, - ) - except Exception: - pass # We just want to capture the request - - # Validate form data contains expected fields - data = captured_request.get("data", {}) - - print("JSON DUMPS captured_request:") - print(json.dumps(captured_request, indent=4, default=str)) - - # Model name should NOT have watsonx/ prefix - assert data.get("model") == "whisper-large-v3-turbo" - - # project_id should be in form data - assert data.get("project_id") == "test-project-123" - - # OpenAI params should be in form data - assert data.get("language") == "en" - assert data.get("temperature") == 0.5 - # response_format should NOT be set by default - only send what user specifies - assert "response_format" not in data - - # Validate file is in files dict (multipart/form-data) - files = captured_request.get("files", {}) - assert "file" in files - assert isinstance( - files["file"], tuple - ) # Should be (filename, content, content_type) - - @pytest.mark.asyncio - async def test_watsonx_transcription_only_user_params_sent_with_project_id(self): - """ - Test that only user-specified params are sent in request body to WatsonX. - - LiteLLM should NOT add extra params like response_format if user didn't specify them. - """ - captured_request = {} - - async def mock_post(*args, **kwargs): - captured_request["data"] = kwargs.get("data", {}) - captured_request["files"] = kwargs.get("files", {}) - - mock_response = MagicMock() - mock_response.json.return_value = { - "text": "test transcription", - "duration": 1.0, - } - mock_response.status_code = 200 - return mock_response - - with patch( - "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post", - new=mock_post, - ): - try: - # Minimal request - only required params - await litellm.atranscription( - model="watsonx/whisper-large-v3-turbo", - file=b"fake_audio_data", - api_base="https://us-south.ml.cloud.ibm.com", - api_key="test-api-key", - project_id="test-project-123", - token="test-bearer-token", - ) - except Exception: - pass # We just want to capture the request - - data = captured_request.get("data", {}) - - # These are the ONLY keys that should be in data - expected_keys = {"model", "project_id"} - actual_keys = set(data.keys()) - - assert actual_keys == expected_keys, ( - f"Request body should only contain {expected_keys}, " - f"but got {actual_keys}. " - f"Extra keys: {actual_keys - expected_keys}" - ) - - # Specifically verify response_format is NOT added - assert ( - "response_format" not in data - ), "response_format should NOT be added by default" - - # Verify file is sent separately - files = captured_request.get("files", {}) - assert "file" in files - - @pytest.mark.asyncio - async def test_watsonx_transcription_only_user_params_sent_with_space_id(self): - """ - Test that only user-specified params are sent in request body to WatsonX. - - LiteLLM should NOT add extra params like response_format if user didn't specify them. - """ - captured_request = {} - - async def mock_post(*args, **kwargs): - captured_request["data"] = kwargs.get("data", {}) - captured_request["files"] = kwargs.get("files", {}) - - mock_response = MagicMock() - mock_response.json.return_value = { - "text": "test transcription", - "duration": 1.0, - } - mock_response.status_code = 200 - return mock_response - - with patch( - "litellm.llms.custom_httpx.http_handler.AsyncHTTPHandler.post", - new=mock_post, - ): - try: - # Minimal request - only required params - await litellm.atranscription( - model="watsonx/whisper-large-v3-turbo", - file=b"fake_audio_data", - api_base="https://us-south.ml.cloud.ibm.com", - api_key="test-api-key", - space_id="test-space_id-123", - token="test-bearer-token", - ) - except Exception: - pass # We just want to capture the request - - data = captured_request.get("data", {}) - - # These are the ONLY keys that should be in data - expected_keys = {"model", "space_id"} - actual_keys = set(data.keys()) - - assert actual_keys == expected_keys, ( - f"Request body should only contain {expected_keys}, " - f"but got {actual_keys}. " - f"Extra keys: {actual_keys - expected_keys}" - ) - - # Specifically verify response_format is NOT added - assert ( - "response_format" not in data - ), "response_format should NOT be added by default" - - # Verify file is sent separately - files = captured_request.get("files", {}) - assert "file" in files - - def test_transform_audio_transcription_response_removes_model_field(self): - """ - Test that transform_audio_transcription_response removes the 'model' field - from WatsonX response before creating TranscriptionResponse. - - This test ensures that when WatsonX returns a response with a 'model' field, - it is removed before creating the TranscriptionResponse object, since - TranscriptionResponse doesn't accept a 'model' parameter. - """ - handler = IBMWatsonXAudioTranscriptionConfig() - - # Mock response with 'model' field (as WatsonX may return) - mock_response = MagicMock() - mock_response.json.return_value = { - "text": "Hello, this is a test transcription.", - "model": "whisper-large-v3-turbo", # This field should be removed - "duration": 5.5, - } - mock_response.text = '{"text": "Hello, this is a test transcription.", "model": "whisper-large-v3-turbo", "duration": 5.5}' - - # This should not raise a TypeError - model field should be removed - result = handler.transform_audio_transcription_response(mock_response) - - # Verify the result is a TranscriptionResponse - assert isinstance(result, TranscriptionResponse) - - # Verify the text is correct - assert result.text == "Hello, this is a test transcription." - - # Verify duration is set via dictionary assignment - assert result["duration"] == 5.5 - - # Verify the model field is NOT in the serialized result - # Check via model_dump() or dict() to ensure it's not in the output - try: - result_dict = result.model_dump() - except AttributeError: - # Fallback for pydantic v1 - result_dict = result.dict() - - # The 'model' field should not be in the result - assert "model" not in result_dict, "Model field should be removed from response" - - def test_transform_audio_transcription_response_without_model_field(self): - """ - Test that transform_audio_transcription_response works correctly - when WatsonX response doesn't include a 'model' field. - """ - handler = IBMWatsonXAudioTranscriptionConfig() - - # Mock response without 'model' field - mock_response = MagicMock() - mock_response.json.return_value = { - "text": "Hello, this is a test transcription.", - "duration": 5.5, - } - mock_response.text = ( - '{"text": "Hello, this is a test transcription.", "duration": 5.5}' - ) - - result = handler.transform_audio_transcription_response(mock_response) - - # Verify the result is a TranscriptionResponse - assert isinstance(result, TranscriptionResponse) - - # Verify the text is correct - assert result.text == "Hello, this is a test transcription." - - # Verify duration is set via dictionary assignment - assert result["duration"] == 5.5 diff --git a/tests/test_litellm/llms/watsonx/test_watsonx.py b/tests/test_litellm/llms/watsonx/test_watsonx.py deleted file mode 100644 index 285afffefc0..00000000000 --- a/tests/test_litellm/llms/watsonx/test_watsonx.py +++ /dev/null @@ -1,577 +0,0 @@ -import json - -from typing import Optional -from unittest.mock import Mock, patch - -import pytest - -import litellm -from litellm import completion -from litellm.llms.custom_httpx.http_handler import HTTPHandler - - -@pytest.fixture -def watsonx_chat_completion_call(): - def _call( - model="watsonx/my-test-model", - messages=None, - api_key="test_api_key", - space_id: Optional[str] = None, - headers=None, - client=None, - patch_token_call=True, - ): - if messages is None: - messages = [{"role": "user", "content": "Hello, how are you?"}] - if client is None: - client = HTTPHandler() - - if patch_token_call: - mock_response = Mock() - mock_response.json.return_value = { - "access_token": "mock_access_token", - "expires_in": 3600, - } - mock_response.raise_for_status = Mock() # No-op to simulate no exception - - with ( - patch.object(client, "post") as mock_post, - patch.object( - litellm.module_level_client, "post", return_value=mock_response - ) as mock_get, - ): - try: - completion( - model=model, - messages=messages, - api_key=api_key, - headers=headers or {}, - client=client, - space_id=space_id, - ) - except Exception as e: - print(e) - - return mock_post, mock_get - else: - with patch.object(client, "post") as mock_post: - try: - completion( - model=model, - messages=messages, - api_key=api_key, - headers=headers or {}, - client=client, - space_id=space_id, - ) - except Exception as e: - print(e) - return mock_post, None - - return _call - - -def test_watsonx_deployment_model_id_not_in_payload( - monkeypatch, watsonx_chat_completion_call -): - """Test that deployment models do not include 'model_id' in the request payload""" - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - model = "watsonx/deployment/test-deployment-id" - messages = [{"role": "user", "content": "Test message"}] - - mock_post, _ = watsonx_chat_completion_call(model=model, messages=messages) - - assert mock_post.call_count == 1 - json_data = json.loads(mock_post.call_args.kwargs["data"]) - # Ensure model_id is not in the payload for deployment models - assert "model_id" not in json_data or json_data["model_id"] is None - # Ensure project_id is also not in the payload for deployment models - assert "project_id" not in json_data or json_data["project_id"] is None - - -def test_watsonx_regular_model_includes_model_id( - monkeypatch, watsonx_chat_completion_call -): - """Test that regular models include 'model_id' in the request payload""" - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - model = "watsonx/regular-model" - messages = [{"role": "user", "content": "Test message"}] - - mock_post, _ = watsonx_chat_completion_call(model=model, messages=messages) - - assert mock_post.call_count == 1 - json_data = json.loads(mock_post.call_args.kwargs["data"]) - # Ensure model_id is included in the payload for regular models - assert "model_id" in json_data - assert json_data["model_id"] == "regular-model" # Provider prefix is stripped - # Ensure project_id is also included for regular models - assert "project_id" in json_data - - -@pytest.fixture -def watsonx_completion_call(): - def _call( - model="watsonx_text/my-test-model", - prompt="Hello, how are you?", - api_key="test_api_key", - space_id: Optional[str] = None, - headers=None, - client=None, - patch_token_call=True, - ): - if client is None: - client = HTTPHandler() - - if patch_token_call: - mock_response = Mock() - mock_response.json.return_value = { - "access_token": "mock_access_token", - "expires_in": 3600, - } - mock_response.raise_for_status = Mock() - - with ( - patch.object(client, "post") as mock_post, - patch.object( - litellm.module_level_client, "post", return_value=mock_response - ) as mock_get, - ): - try: - litellm.text_completion( - model=model, - prompt=prompt, - api_key=api_key, - headers=headers or {}, - client=client, - space_id=space_id, - ) - except Exception as e: - print(e) - - return mock_post, mock_get - else: - with patch.object(client, "post") as mock_post: - try: - litellm.text_completion( - model=model, - prompt=prompt, - api_key=api_key, - headers=headers or {}, - client=client, - space_id=space_id, - ) - except Exception as e: - print(e) - return mock_post, None - - return _call - - -def test_watsonx_completion_deployment_model_id_not_in_payload( - monkeypatch, watsonx_completion_call -): - """Test that deployment models do not include 'model_id' in completion request payload""" - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - model = "watsonx_text/deployment/test-deployment-id" - prompt = "Test prompt" - - mock_post, _ = watsonx_completion_call(model=model, prompt=prompt) - - assert mock_post.call_count == 1 - json_data = json.loads(mock_post.call_args.kwargs["data"]) - # Ensure model_id is not in the payload for deployment models - assert "model_id" not in json_data - # Ensure project_id is also not in the payload for deployment models - assert "project_id" not in json_data - - -def test_watsonx_completion_regular_model_includes_model_id( - monkeypatch, watsonx_completion_call -): - """Test that regular models include 'model_id' in completion request payload""" - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - model = "watsonx_text/regular-model" - prompt = "Test prompt" - - mock_post, _ = watsonx_completion_call(model=model, prompt=prompt) - - assert mock_post.call_count == 1 - json_data = json.loads(mock_post.call_args.kwargs["data"]) - # Ensure model_id is included in the payload for regular models - assert "model_id" in json_data - assert json_data["model_id"] == "regular-model" # Provider prefix is stripped - # Ensure project_id is also included for regular models - assert "project_id" in json_data - - -def test_watsonx_gpt_oss_prompt_transformation(monkeypatch): - """ - Test that gpt-oss-120b model transforms messages to proper format instead of simple concatenation. - - This test calls litellm.completion (sync) and verifies what gets sent in the final POST request body. - Input messages should be transformed using the HuggingFace chat template from openai/gpt-oss-120b, - not just concatenated as "You are chatgpt Hi there". - """ - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - - # Test with gpt-oss model using watsonx_text provider (text generation endpoint) - model = "watsonx_text/openai/gpt-oss-120b" - - # Input messages - messages = [ - {"role": "system", "content": "You are chatgpt"}, - {"role": "user", "content": "Hi there"}, - ] - - client = HTTPHandler() - - # 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. - # - # 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, - }, - } - - # Isolate known_tokenizer_config so parallel tests don't interfere. - # monkeypatch.setitem restores the original value on teardown. - hf_model = "openai/gpt-oss-120b" - monkeypatch.setitem(litellm.known_tokenizer_config, hf_model, mock_tokenizer_config) - - # Mock IAM token generation to avoid real HTTP calls. - mock_token_response = Mock() - mock_token_response.json.return_value = { - "access_token": "mock_access_token", - "expires_in": 3600, - } - mock_token_response.raise_for_status = Mock() - - with ( - patch.object(client, "post") as mock_post, - patch.object( - litellm.module_level_client, "post", return_value=mock_token_response - ), - ): - try: - completion( - model=model, - messages=messages, - api_key="test_api_key", - client=client, - ) - except Exception as e: - print(f"Caught expected exception: {e}") - - # Verify the POST was called - assert ( - mock_post.call_count == 1 - ), f"POST should have been called exactly once, got {mock_post.call_count}" - - # Get the request body - call_args = mock_post.call_args - assert "data" in call_args.kwargs, "call_args.kwargs should contain 'data'" - json_data = json.loads(call_args.kwargs["data"]) - - # Verify the transformed input is in the request - assert "input" in json_data, "Request should have 'input' field" - transformed_prompt = json_data["input"] - - # Verify it's NOT simple concatenation - simple_concat = "You are chatgpt Hi there" - assert transformed_prompt != simple_concat, ( - f"Prompt should not be simple concatenation.\n" - f"Expected: Chat template with <|start|> tags\n" - f"Got: {transformed_prompt}" - ) - - # Verify it contains proper chat template formatting - assert "<|start|>" in transformed_prompt, "Prompt should contain <|start|> tag" - assert "<|message|>" in transformed_prompt, "Prompt should contain <|message|> tag" - assert "<|end|>" in transformed_prompt, "Prompt should contain <|end|> tag" - assert ( - "You are chatgpt" in transformed_prompt - ), "Prompt should contain system message content" - assert ( - "Hi there" in transformed_prompt - ), "Prompt should contain user message content" - - -@pytest.mark.asyncio -@pytest.mark.xdist_group("watsonx_heavy") -async def test_watsonx_gpt_oss_uses_async_http_handler(): - """ - Test that verifies async HTTP client is used when fetching HuggingFace templates. - """ - from unittest.mock import AsyncMock, MagicMock, patch - - from litellm.litellm_core_utils.prompt_templates.huggingface_template_handler import ( - _aget_chat_template_file, - ) - - # Mock the async HTTP client - mock_async_client = MagicMock() - mock_get = AsyncMock() - mock_async_client.get = mock_get - - # Create mock response for chat template file - mock_response = MagicMock() - mock_response.status_code = 200 - mock_response.content = b"test template content" - mock_get.return_value = mock_response - - # Test the async function directly - with patch( - "litellm.litellm_core_utils.prompt_templates.huggingface_template_handler.get_async_httpx_client", - return_value=mock_async_client, - ): - result = await _aget_chat_template_file(hf_model_name="test/model") - - # Verify async HTTP client was called - assert mock_get.called, "Async HTTP client's get method should be called" - assert mock_get.await_count > 0, "Async HTTP client's get should be awaited" - - # Verify it was called with HuggingFace URL - call_args = mock_get.call_args - assert call_args is not None, "get should have been called with arguments" - called_url = call_args.kwargs.get("url", "") - assert ( - "huggingface.co/test/model" in called_url - ), f"Should call HuggingFace API for test/model, got: {called_url}" - assert result["status"] == "success", "Should return success status" - - -@pytest.mark.parametrize("tokenizer_config_cached", [False, True], ids=["tokenizer_config", "cached_config_jinja"]) -async def test_watsonx_text_gpt_oss_async_completion_fetches_hf_template_off_the_event_loop( - monkeypatch, tokenizer_config_cached -): - import httpx - - from litellm._uuid import uuid - from litellm.litellm_core_utils.prompt_templates import huggingface_template_handler - from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler - - hf_model = f"openai/gpt-oss-{uuid.uuid4()}" - chat_template = "{% for m in messages %}<|{{ m['role'] }}|>{{ m['content'] }}{% endfor %}" - if tokenizer_config_cached: - cached_config = {"status": "success", "tokenizer": {"bos_token": None, "eos_token": None}} - monkeypatch.setattr(litellm, "known_tokenizer_config", {hf_model: cached_config}) - expected_fetch = f"https://huggingface.co/{hf_model}/raw/main/chat_template.jinja" - else: - monkeypatch.setattr(litellm, "known_tokenizer_config", {}) - expected_fetch = f"https://huggingface.co/{hf_model}/raw/main/tokenizer_config.json" - hf_fetched = [] - captured = {} - - def forbid_sync_client(): - raise AssertionError("sync HuggingFace fetch ran on the request path") - - async def serve_hf_file(url, **kwargs): - hf_fetched.append(url) - if url.endswith(".jinja"): - return httpx.Response(200, content=chat_template.encode()) - return httpx.Response(200, json={"chat_template": chat_template, "bos_token": None, "eos_token": None}) - - monkeypatch.setattr(huggingface_template_handler, "_get_httpx_client", forbid_sync_client) - monkeypatch.setattr(huggingface_template_handler, "get_async_httpx_client", lambda **kwargs: Mock(get=serve_hf_file)) - - def handle(request): - captured["body"] = json.loads(request.content) - return httpx.Response( - 200, - json={ - "model_id": hf_model, - "results": [ - { - "generated_text": "Hi", - "generated_token_count": 1, - "input_token_count": 1, - "stop_reason": "eos_token", - } - ], - }, - ) - - client = AsyncHTTPHandler() - client.client = httpx.AsyncClient(transport=httpx.MockTransport(handle)) - - response = await litellm.acompletion( - model=f"watsonx_text/{hf_model}", - messages=[{"role": "user", "content": "Hi there"}], - api_base="https://test-api.watsonx.ai", - project_id="test-project-id", - token="test-token", - client=client, - ) - - assert response.choices[0].message.content == "Hi" - assert hf_fetched == [expected_fetch] - assert captured["body"]["input"] == "<|user|>Hi there" - - -def test_watsonx_chat_completion_with_reasoning_effort(monkeypatch): - """ - Test that 'reasoning_effort' is correctly passed through to the WatsonX API payload. - """ - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - - model = "watsonx/openai/gpt-oss-120b" - messages = [{"role": "user", "content": "Test message"}] - - client = HTTPHandler() - - # Mock the token generation call - mock_token_response = Mock() - mock_token_response.json.return_value = { - "access_token": "mock_access_token", - "expires_in": 3600, - } - mock_token_response.raise_for_status = Mock() - - # Call litellm.completion with the new parameter - with ( - patch.object(client, "post") as mock_post, - patch.object( - litellm.module_level_client, "post", return_value=mock_token_response - ), - ): - try: - completion( - model=model, - messages=messages, - api_key="test_api_key", - client=client, - reasoning_effort="low", - ) - except Exception as e: - print(f"Caught expected exception: {e}") - - # Verify the parameter is in the final request payload - assert ( - mock_post.call_count == 1 - ), "The completion endpoint should have been called once." - - # Get the JSON data sent in the POST request - request_kwargs = mock_post.call_args.kwargs - json_data = json.loads(request_kwargs["data"]) - - print("\nRequest payload sent to WatsonX API:") - print(json.dumps(json_data, indent=2)) - - # Check for the parameter at the top level of the payload - assert ( - "reasoning_effort" in json_data - ), "'reasoning_effort' should be at the top level of the payload." - assert ( - json_data["reasoning_effort"] == "low" - ), "The value of 'reasoning_effort' should be 'low'." - - -def test_watsonx_zen_api_key_from_client(monkeypatch, watsonx_chat_completion_call): - """ - Test that zen_api_key can be passed from client code and is used in Authorization header. - """ - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - - model = "watsonx/ibm/granite-3-3-8b-instruct" - messages = [{"role": "user", "content": "What is your favorite color?"}] - - client = HTTPHandler() - - zen_api_key = "U1ZDLWQo=" - - # No need to patch token call since zen_api_key should skip token generation - with patch.object(client, "post") as mock_post: - try: - completion( - model=model, - messages=messages, - api_key="test_api_key", - client=client, - zen_api_key=zen_api_key, - ) - except Exception as e: - print(f"Caught expected exception: {e}") - - # Verify the request was made - assert ( - mock_post.call_count == 1 - ), "The completion endpoint should have been called once." - - # Get the headers sent in the POST request - request_kwargs = mock_post.call_args.kwargs - headers = request_kwargs["headers"] - - print("\nHeaders sent to WatsonX API:") - print(json.dumps(dict(headers), indent=2)) - - # Verify Authorization header uses ZenApiKey format - assert "Authorization" in headers, "Authorization header should be present." - assert headers["Authorization"] == f"ZenApiKey {zen_api_key}", ( - f"Authorization header should use ZenApiKey format. " - f"Expected: 'ZenApiKey {zen_api_key}', Got: '{headers['Authorization']}'" - ) - - -def test_watsonx_zen_api_key_from_env(monkeypatch, watsonx_chat_completion_call): - """ - Test that zen_api_key from environment variable is used in Authorization header. - """ - monkeypatch.setenv("WATSONX_PROJECT_ID", "test-project-id") - monkeypatch.setenv("WATSONX_API_BASE", "https://test-api.watsonx.ai") - - zen_api_key = "U1ZDLWxpdG--===" - monkeypatch.setenv("WATSONX_ZENAPIKEY", zen_api_key) - - model = "watsonx/ibm/granite-3-3-8b-instruct" - messages = [{"role": "user", "content": "What is your favorite color?"}] - - client = HTTPHandler() - - # No need to patch token call since zen_api_key should skip token generation - with patch.object(client, "post") as mock_post: - try: - completion( - model=model, - messages=messages, - api_key="test_api_key", - client=client, - ) - except Exception as e: - print(f"Caught expected exception: {e}") - - # Verify the request was made - assert ( - mock_post.call_count == 1 - ), "The completion endpoint should have been called once." - - # Get the headers sent in the POST request - request_kwargs = mock_post.call_args.kwargs - headers = request_kwargs["headers"] - - print("\nHeaders sent to WatsonX API:") - print(json.dumps(dict(headers), indent=2)) - - # Verify Authorization header uses ZenApiKey format - assert "Authorization" in headers, "Authorization header should be present." - assert headers["Authorization"] == f"ZenApiKey {zen_api_key}", ( - f"Authorization header should use ZenApiKey format. " - f"Expected: 'ZenApiKey {zen_api_key}', Got: '{headers['Authorization']}'" - ) diff --git a/tests/test_litellm/llms/xai/xai_responses/__init__.py b/tests/test_litellm/llms/xai/xai_responses/__init__.py deleted file mode 100644 index 330e9f5a560..00000000000 --- a/tests/test_litellm/llms/xai/xai_responses/__init__.py +++ /dev/null @@ -1 +0,0 @@ -# XAI Responses API tests diff --git a/tests/test_litellm/llms/xai/xai_responses/test_transformation.py b/tests/test_litellm/llms/xai/xai_responses/test_transformation.py deleted file mode 100644 index 3ea3fe631bd..00000000000 --- a/tests/test_litellm/llms/xai/xai_responses/test_transformation.py +++ /dev/null @@ -1,105 +0,0 @@ -""" -Tests for XAI Responses API transformation - -Tests the XAIResponsesAPIConfig class that handles XAI-specific -transformations for the Responses API. - -Source: litellm/llms/xai/responses/transformation.py -""" - - - -import pytest -from litellm.types.utils import LlmProviders -from litellm.utils import ProviderConfigManager -from litellm.llms.xai.responses.transformation import XAIResponsesAPIConfig -from litellm.types.llms.openai import ResponsesAPIOptionalRequestParams - - -class TestXAIResponsesAPITransformation: - """Test XAI Responses API configuration and transformations""" - - def test_xai_provider_config_registration(self): - """Test that XAI provider returns XAIResponsesAPIConfig""" - config = ProviderConfigManager.get_provider_responses_api_config( - model="xai/grok-4-fast", - provider=LlmProviders.XAI, - ) - - assert config is not None, "Config should not be None for XAI provider" - assert isinstance( - config, XAIResponsesAPIConfig - ), f"Expected XAIResponsesAPIConfig, got {type(config)}" - assert ( - config.custom_llm_provider == LlmProviders.XAI - ), "custom_llm_provider should be XAI" - - def test_code_interpreter_container_field_removed(self): - """Test that container field is removed from code_interpreter tools""" - config = XAIResponsesAPIConfig() - - params = ResponsesAPIOptionalRequestParams( - tools=[{"type": "code_interpreter", "container": {"type": "auto"}}] - ) - - result = config.map_openai_params( - response_api_optional_params=params, model="grok-4-fast", drop_params=False - ) - - assert "tools" in result - assert len(result["tools"]) == 1 - assert result["tools"][0]["type"] == "code_interpreter" - assert ( - "container" not in result["tools"][0] - ), "Container field should be removed" - - def test_instructions_parameter_forwarded(self): - """xAI supports 'instructions' on /v1/responses, so it must survive param mapping""" - config = XAIResponsesAPIConfig() - - params = ResponsesAPIOptionalRequestParams( - instructions="You are a helpful assistant.", temperature=0.7 - ) - - result = config.map_openai_params( - response_api_optional_params=params, model="grok-4-fast", drop_params=False - ) - - assert result.get("instructions") == "You are a helpful assistant." - assert result.get("temperature") == 0.7, "Other params should be preserved" - - def test_supported_params_includes_instructions(self): - """A system message bridged to 'instructions' must not be rejected for xAI""" - config = XAIResponsesAPIConfig() - supported = config.get_supported_openai_params("grok-4-fast") - - assert "instructions" in supported, "instructions should be supported" - assert "tools" in supported, "tools should be supported" - assert "temperature" in supported, "temperature should be supported" - assert "model" in supported, "model should be supported" - - def test_xai_responses_endpoint_url(self): - """Test that get_complete_url returns correct XAI endpoint""" - config = XAIResponsesAPIConfig() - - # Test with default XAI API base - url = config.get_complete_url(api_base=None, litellm_params={}) - assert ( - url == "https://api.x.ai/v1/responses" - ), f"Expected XAI responses endpoint, got {url}" - - # Test with custom api_base - custom_url = config.get_complete_url( - api_base="https://custom.x.ai/v1", litellm_params={} - ) - assert ( - custom_url == "https://custom.x.ai/v1/responses" - ), f"Expected custom endpoint, got {custom_url}" - - # Test with trailing slash - url_with_slash = config.get_complete_url( - api_base="https://api.x.ai/v1/", litellm_params={} - ) - assert ( - url_with_slash == "https://api.x.ai/v1/responses" - ), "Should handle trailing slash" diff --git a/tests/unit/llms/vertex_ai/__init__.py b/tests/unit/llms/vertex_ai/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/__init__.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/__init__.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py similarity index 98% rename from tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py rename to tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py index 4b710175a48..e2fb81bc240 100644 --- a/tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py +++ b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_location.py @@ -98,6 +98,8 @@ class TestCountTokensLocationResolution: self, counter, monkeypatch ): """Claude models without any location should default to us-east5.""" + monkeypatch.delenv("VERTEXAI_LOCATION", raising=False) + monkeypatch.delenv("VERTEX_LOCATION", raising=False) captured = {} async def fake_ensure_access_token( diff --git a/tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_no_vertexai_sdk.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_no_vertexai_sdk.py similarity index 100% rename from tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_no_vertexai_sdk.py rename to tests/unit/llms/vertex_ai/vertex_ai_partner_models/count_tokens/test_count_tokens_no_vertexai_sdk.py diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/llama3/__init__.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/llama3/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/llama3/test_vertex_ai_partner_models_llama3_transformation.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/llama3/test_vertex_ai_partner_models_llama3_transformation.py similarity index 100% rename from tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/llama3/test_vertex_ai_partner_models_llama3_transformation.py rename to tests/unit/llms/vertex_ai/vertex_ai_partner_models/llama3/test_vertex_ai_partner_models_llama3_transformation.py diff --git a/tests/unit/llms/vertex_ai/vertex_ai_partner_models/mistral/__init__.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/mistral/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/mistral/test_vertex_ai_partner_models_mistral_transformation.py b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/mistral/test_vertex_ai_partner_models_mistral_transformation.py similarity index 60% rename from tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/mistral/test_vertex_ai_partner_models_mistral_transformation.py rename to tests/unit/llms/vertex_ai/vertex_ai_partner_models/mistral/test_vertex_ai_partner_models_mistral_transformation.py index f7df4507651..15df8e47af3 100644 --- a/tests/test_litellm/llms/vertex_ai/vertex_ai_partner_models/mistral/test_vertex_ai_partner_models_mistral_transformation.py +++ b/tests/unit/llms/vertex_ai/vertex_ai_partner_models/mistral/test_vertex_ai_partner_models_mistral_transformation.py @@ -1,6 +1,21 @@ +import pytest + import litellm +@pytest.fixture +def local_model_cost_map(monkeypatch): + original_model_cost = litellm.model_cost + monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True") + litellm.model_cost = litellm.get_model_cost_map(url="") + litellm.get_model_info.cache_clear() + try: + yield + finally: + litellm.model_cost = original_model_cost + litellm.get_model_info.cache_clear() + + def test_reasoning_effort_stays_unsupported_on_vertex_partner_models(local_model_cost_map): assert "reasoning_effort" in litellm.get_supported_openai_params( model="mistral-medium-3", custom_llm_provider="mistral" diff --git a/tests/unit/llms/vertex_ai/vertex_gemma_models/__init__.py b/tests/unit/llms/vertex_ai/vertex_gemma_models/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/vertex_ai/vertex_gemma_models/test_vertex_gemma_transformation.py b/tests/unit/llms/vertex_ai/vertex_gemma_models/test_vertex_gemma_transformation.py similarity index 100% rename from tests/test_litellm/llms/vertex_ai/vertex_gemma_models/test_vertex_gemma_transformation.py rename to tests/unit/llms/vertex_ai/vertex_gemma_models/test_vertex_gemma_transformation.py diff --git a/tests/unit/llms/vertex_ai/videos/__init__.py b/tests/unit/llms/vertex_ai/videos/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/vertex_ai/videos/test_vertex_video_transformation.py b/tests/unit/llms/vertex_ai/videos/test_vertex_video_transformation.py similarity index 100% rename from tests/test_litellm/llms/vertex_ai/videos/test_vertex_video_transformation.py rename to tests/unit/llms/vertex_ai/videos/test_vertex_video_transformation.py diff --git a/tests/unit/llms/volcengine/__init__.py b/tests/unit/llms/volcengine/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/volcengine/responses/__init__.py b/tests/unit/llms/volcengine/responses/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/volcengine/responses/test_volcengine_responses_transformation.py b/tests/unit/llms/volcengine/responses/test_volcengine_responses_transformation.py similarity index 94% rename from tests/test_litellm/llms/volcengine/responses/test_volcengine_responses_transformation.py rename to tests/unit/llms/volcengine/responses/test_volcengine_responses_transformation.py index d42bf7b7a1c..5c8d67ecc70 100644 --- a/tests/test_litellm/llms/volcengine/responses/test_volcengine_responses_transformation.py +++ b/tests/unit/llms/volcengine/responses/test_volcengine_responses_transformation.py @@ -137,30 +137,6 @@ class TestVolcengineResponsesAPITransformation: with pytest.raises(ValueError, match='Volcengine API key is required\\. Set ARK_API_KEY /'): config.validate_environment(headers={}, model="volcengine/demo", litellm_params={}) - def test_unsupported_params_are_dropped_with_extra_body(self): - """Unknown fields (including extra_body) should be dropped before send.""" - config = VolcEngineResponsesAPIConfig() - - request = config.transform_responses_api_request( - model="volcengine/demo-model", - input="hi", - response_api_optional_request_params={ - "unsupported_custom_param": 0.1, - "temperature": 0.2, - "metadata": {"k": "v"}, - "extra_body": {"unsupported_custom_param": 1, "temperature": 0.3}, - }, - litellm_params=GenericLiteLLMParams(), - headers={}, - ) - - assert "unsupported_custom_param" not in request - assert "metadata" not in request - assert request["temperature"] == 0.2 - assert "extra_body" in request - assert "unsupported_custom_param" not in request["extra_body"] - assert request["extra_body"]["temperature"] == 0.3 - def test_valid_thinking_caching_and_expire_at_pass(self): """Documented params should pass through without validation errors.""" config = VolcEngineResponsesAPIConfig() diff --git a/tests/unit/llms/voyage/__init__.py b/tests/unit/llms/voyage/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/voyage/rerank/__init__.py b/tests/unit/llms/voyage/rerank/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/voyage/rerank/test_voyage_rerank_transformation.py b/tests/unit/llms/voyage/rerank/test_voyage_rerank_transformation.py similarity index 100% rename from tests/test_litellm/llms/voyage/rerank/test_voyage_rerank_transformation.py rename to tests/unit/llms/voyage/rerank/test_voyage_rerank_transformation.py diff --git a/tests/test_litellm/llms/voyage/test_voyage_contextual_embedding.py b/tests/unit/llms/voyage/test_voyage_contextual_embedding.py similarity index 100% rename from tests/test_litellm/llms/voyage/test_voyage_contextual_embedding.py rename to tests/unit/llms/voyage/test_voyage_contextual_embedding.py diff --git a/tests/test_litellm/llms/voyage/test_voyage_multimodal_embedding.py b/tests/unit/llms/voyage/test_voyage_multimodal_embedding.py similarity index 100% rename from tests/test_litellm/llms/voyage/test_voyage_multimodal_embedding.py rename to tests/unit/llms/voyage/test_voyage_multimodal_embedding.py diff --git a/tests/unit/llms/watsonx/__init__.py b/tests/unit/llms/watsonx/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/watsonx/audio_transcription/__init__.py b/tests/unit/llms/watsonx/audio_transcription/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py b/tests/unit/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py new file mode 100644 index 00000000000..efe592f515e --- /dev/null +++ b/tests/unit/llms/watsonx/audio_transcription/test_watsonx_audio_transcription_transformation.py @@ -0,0 +1,85 @@ +""" +Tests for IBM WatsonX Audio Transcription. + +Validates the WatsonX transcription response transformation. +""" + +from unittest.mock import MagicMock + +from litellm.llms.watsonx.audio_transcription.transformation import ( + IBMWatsonXAudioTranscriptionConfig, +) +from litellm.types.utils import TranscriptionResponse + + +class TestWatsonXAudioTranscription: + def test_transform_audio_transcription_response_removes_model_field(self): + """ + Test that transform_audio_transcription_response removes the 'model' field + from WatsonX response before creating TranscriptionResponse. + + This test ensures that when WatsonX returns a response with a 'model' field, + it is removed before creating the TranscriptionResponse object, since + TranscriptionResponse doesn't accept a 'model' parameter. + """ + handler = IBMWatsonXAudioTranscriptionConfig() + + # Mock response with 'model' field (as WatsonX may return) + mock_response = MagicMock() + mock_response.json.return_value = { + "text": "Hello, this is a test transcription.", + "model": "whisper-large-v3-turbo", # This field should be removed + "duration": 5.5, + } + mock_response.text = '{"text": "Hello, this is a test transcription.", "model": "whisper-large-v3-turbo", "duration": 5.5}' + + # This should not raise a TypeError - model field should be removed + result = handler.transform_audio_transcription_response(mock_response) + + # Verify the result is a TranscriptionResponse + assert isinstance(result, TranscriptionResponse) + + # Verify the text is correct + assert result.text == "Hello, this is a test transcription." + + # Verify duration is set via dictionary assignment + assert result["duration"] == 5.5 + + # Verify the model field is NOT in the serialized result + # Check via model_dump() or dict() to ensure it's not in the output + try: + result_dict = result.model_dump() + except AttributeError: + # Fallback for pydantic v1 + result_dict = result.dict() + + # The 'model' field should not be in the result + assert "model" not in result_dict, "Model field should be removed from response" + + def test_transform_audio_transcription_response_without_model_field(self): + """ + Test that transform_audio_transcription_response works correctly + when WatsonX response doesn't include a 'model' field. + """ + handler = IBMWatsonXAudioTranscriptionConfig() + + # Mock response without 'model' field + mock_response = MagicMock() + mock_response.json.return_value = { + "text": "Hello, this is a test transcription.", + "duration": 5.5, + } + mock_response.text = ( + '{"text": "Hello, this is a test transcription.", "duration": 5.5}' + ) + + result = handler.transform_audio_transcription_response(mock_response) + + # Verify the result is a TranscriptionResponse + assert isinstance(result, TranscriptionResponse) + + # Verify the text is correct + assert result.text == "Hello, this is a test transcription." + + # Verify duration is set via dictionary assignment + assert result["duration"] == 5.5 diff --git a/tests/unit/llms/watsonx/embed/__init__.py b/tests/unit/llms/watsonx/embed/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/watsonx/embed/test_watsonx_embedding_transformation.py b/tests/unit/llms/watsonx/embed/test_watsonx_embedding_transformation.py similarity index 100% rename from tests/test_litellm/llms/watsonx/embed/test_watsonx_embedding_transformation.py rename to tests/unit/llms/watsonx/embed/test_watsonx_embedding_transformation.py diff --git a/tests/unit/llms/watsonx/passthrough/__init__.py b/tests/unit/llms/watsonx/passthrough/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/watsonx/passthrough/test_watsonx_passthrough_transformation.py b/tests/unit/llms/watsonx/passthrough/test_watsonx_passthrough_transformation.py similarity index 100% rename from tests/test_litellm/llms/watsonx/passthrough/test_watsonx_passthrough_transformation.py rename to tests/unit/llms/watsonx/passthrough/test_watsonx_passthrough_transformation.py diff --git a/tests/unit/llms/watsonx/rerank/__init__.py b/tests/unit/llms/watsonx/rerank/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/watsonx/rerank/test_watsonx_rerank.py b/tests/unit/llms/watsonx/rerank/test_watsonx_rerank.py similarity index 100% rename from tests/test_litellm/llms/watsonx/rerank/test_watsonx_rerank.py rename to tests/unit/llms/watsonx/rerank/test_watsonx_rerank.py diff --git a/tests/unit/llms/watsonx/test_watsonx.py b/tests/unit/llms/watsonx/test_watsonx.py new file mode 100644 index 00000000000..077539c9acd --- /dev/null +++ b/tests/unit/llms/watsonx/test_watsonx.py @@ -0,0 +1,74 @@ +import json +from unittest.mock import Mock + +import pytest + +import litellm + + +@pytest.mark.parametrize("tokenizer_config_cached", [False, True], ids=["tokenizer_config", "cached_config_jinja"]) +async def test_watsonx_text_gpt_oss_async_completion_fetches_hf_template_off_the_event_loop( + monkeypatch, tokenizer_config_cached +): + import httpx + + from litellm._uuid import uuid + from litellm.litellm_core_utils.prompt_templates import huggingface_template_handler + from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler + + hf_model = f"openai/gpt-oss-{uuid.uuid4()}" + chat_template = "{% for m in messages %}<|{{ m['role'] }}|>{{ m['content'] }}{% endfor %}" + if tokenizer_config_cached: + cached_config = {"status": "success", "tokenizer": {"bos_token": None, "eos_token": None}} + monkeypatch.setattr(litellm, "known_tokenizer_config", {hf_model: cached_config}) + expected_fetch = f"https://huggingface.co/{hf_model}/raw/main/chat_template.jinja" + else: + monkeypatch.setattr(litellm, "known_tokenizer_config", {}) + expected_fetch = f"https://huggingface.co/{hf_model}/raw/main/tokenizer_config.json" + hf_fetched = [] + captured = {} + + def forbid_sync_client(): + raise AssertionError("sync HuggingFace fetch ran on the request path") + + async def serve_hf_file(url, **kwargs): + hf_fetched.append(url) + if url.endswith(".jinja"): + return httpx.Response(200, content=chat_template.encode()) + return httpx.Response(200, json={"chat_template": chat_template, "bos_token": None, "eos_token": None}) + + monkeypatch.setattr(huggingface_template_handler, "_get_httpx_client", forbid_sync_client) + monkeypatch.setattr(huggingface_template_handler, "get_async_httpx_client", lambda **kwargs: Mock(get=serve_hf_file)) + + def handle(request): + captured["body"] = json.loads(request.content) + return httpx.Response( + 200, + json={ + "model_id": hf_model, + "results": [ + { + "generated_text": "Hi", + "generated_token_count": 1, + "input_token_count": 1, + "stop_reason": "eos_token", + } + ], + }, + ) + + client = AsyncHTTPHandler() + client.client = httpx.AsyncClient(transport=httpx.MockTransport(handle)) + + response = await litellm.acompletion( + model=f"watsonx_text/{hf_model}", + messages=[{"role": "user", "content": "Hi there"}], + api_base="https://test-api.watsonx.ai", + project_id="test-project-id", + token="test-token", + client=client, + ) + + assert response.choices[0].message.content == "Hi" + assert hf_fetched == [expected_fetch] + assert captured["body"]["input"] == "<|user|>Hi there" diff --git a/tests/test_litellm/llms/watsonx/test_watsonx_common_utils.py b/tests/unit/llms/watsonx/test_watsonx_common_utils.py similarity index 100% rename from tests/test_litellm/llms/watsonx/test_watsonx_common_utils.py rename to tests/unit/llms/watsonx/test_watsonx_common_utils.py diff --git a/tests/unit/llms/xai/__init__.py b/tests/unit/llms/xai/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/unit/llms/xai/responses/__init__.py b/tests/unit/llms/xai/responses/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/xai/responses/test_xai_responses_transformation.py b/tests/unit/llms/xai/responses/test_xai_responses_transformation.py similarity index 100% rename from tests/test_litellm/llms/xai/responses/test_xai_responses_transformation.py rename to tests/unit/llms/xai/responses/test_xai_responses_transformation.py diff --git a/tests/unit/llms/you_com/__init__.py b/tests/unit/llms/you_com/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/you_com/test_you_com_search.py b/tests/unit/llms/you_com/test_you_com_search.py similarity index 100% rename from tests/test_litellm/llms/you_com/test_you_com_search.py rename to tests/unit/llms/you_com/test_you_com_search.py diff --git a/tests/unit/llms/zai/__init__.py b/tests/unit/llms/zai/__init__.py new file mode 100644 index 00000000000..e69de29bb2d diff --git a/tests/test_litellm/llms/zai/test_zai_provider.py b/tests/unit/llms/zai/test_zai_provider.py similarity index 100% rename from tests/test_litellm/llms/zai/test_zai_provider.py rename to tests/unit/llms/zai/test_zai_provider.py