From fcf917df6d8c4eb790acfa61963c3a448e56093c Mon Sep 17 00:00:00 2001 From: "Zark ." <87560774+Alpha-Zark@users.noreply.github.com> Date: Thu, 23 Apr 2026 11:03:46 +0800 Subject: [PATCH] Feat(dashscope): add image generation support for qwen-image-2.0 and qwen-image-2.0-pro (#25672) * feat: add dashscope/qwen-image-2.0 and qwen-image-2.0-pro to model cost map * feat: implement DashScope image generation transformation class * feat: register DashScope in ProviderConfigManager for image generation * feat: add DashScope to image generation provider routing * feat: auto-route qwen-image /chat/completions requests to /images/generations * test: add unit tests for DashScope image generation (22 cases) * refactor: remove proxy-layer qwen-image auto-routing * feat: auto-redirect image_generation models in acompletion() * test: add acompletion auto-redirect test for image_generation models * fix: remove unused Union import in DashScope transformation * fix: scope acompletion redirect to dashscope and narrow exception handler * fix: move get_str_from_messages to module-level import and forward n param to aimage_generation * refactor: remove acompletion image_generation auto-redirect for dashscope * test: remove acompletion auto-redirect test for dashscope image models --------- Co-authored-by: zark.lin --- litellm/images/main.py | 1 + .../dashscope/image_generation/__init__.py | 9 + .../image_generation/transformation.py | 187 ++++++++++ ...odel_prices_and_context_window_backup.json | 16 + litellm/proxy/proxy_server.py | 1 + litellm/utils.py | 6 + model_prices_and_context_window.json | 16 + .../test_dashscope_image_generation.py | 328 ++++++++++++++++++ 8 files changed, 564 insertions(+) create mode 100644 litellm/llms/dashscope/image_generation/__init__.py create mode 100644 litellm/llms/dashscope/image_generation/transformation.py create mode 100644 tests/test_litellm/test_dashscope_image_generation.py diff --git a/litellm/images/main.py b/litellm/images/main.py index 0d3b2e9729..d95b7287d2 100644 --- a/litellm/images/main.py +++ b/litellm/images/main.py @@ -410,6 +410,7 @@ def image_generation( # noqa: PLR0915 litellm.LlmProviders.RUNWAYML, litellm.LlmProviders.VERTEX_AI, litellm.LlmProviders.OPENROUTER, + litellm.LlmProviders.DASHSCOPE, ): if image_generation_config is None: raise ValueError( diff --git a/litellm/llms/dashscope/image_generation/__init__.py b/litellm/llms/dashscope/image_generation/__init__.py new file mode 100644 index 0000000000..9fdb46586e --- /dev/null +++ b/litellm/llms/dashscope/image_generation/__init__.py @@ -0,0 +1,9 @@ +from litellm.llms.base_llm.image_generation.transformation import BaseImageGenerationConfig + +from .transformation import DashScopeImageGenerationConfig + +__all__ = ["DashScopeImageGenerationConfig"] + + +def get_dashscope_image_generation_config(model: str) -> BaseImageGenerationConfig: + return DashScopeImageGenerationConfig() diff --git a/litellm/llms/dashscope/image_generation/transformation.py b/litellm/llms/dashscope/image_generation/transformation.py new file mode 100644 index 0000000000..feac811df8 --- /dev/null +++ b/litellm/llms/dashscope/image_generation/transformation.py @@ -0,0 +1,187 @@ +""" +DashScope Image Generation Configuration + +Handles transformation between OpenAI-compatible format and DashScope multimodal-generation API. + +API endpoint: POST https://dashscope-intl.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation + +Request format: +{ + "model": "qwen-image-2.0-pro", + "input": { + "messages": [{"role": "user", "content": [{"text": ""}]}] + }, + "parameters": {"size": "1024*1024", ...} +} + +Response format: +{ + "output": { + "choices": [{"message": {"content": [{"image": ""}]}}] + }, + "usage": {"input_tokens": 0, "output_tokens": 0, "width": 1024, "height": 1024, "image_count": 1} +} +""" + +from typing import TYPE_CHECKING, Any, List, Optional + +import httpx + +from litellm.llms.base_llm.image_generation.transformation import BaseImageGenerationConfig +from litellm.secret_managers.main import get_secret_str +from litellm.types.llms.openai import AllMessageValues, OpenAIImageGenerationOptionalParams +from litellm.types.utils import ImageObject, ImageResponse + +if TYPE_CHECKING: + from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj + + LiteLLMLoggingObj = _LiteLLMLoggingObj +else: + LiteLLMLoggingObj = Any + +DEFAULT_API_BASE = "https://dashscope-intl.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" + +# Maps OpenAI size strings (WxH) to DashScope size strings (W*H) +OPENAI_TO_DASHSCOPE_SIZE: dict = { + "256x256": "256*256", + "512x512": "512*512", + "1024x1024": "1024*1024", + "1792x1024": "1792*1024", + "1024x1792": "1024*1792", + "2048x2048": "2048*2048", +} + + +class DashScopeImageGenerationConfig(BaseImageGenerationConfig): + """ + Configuration for DashScope image generation (qwen-image-2.0, qwen-image-2.0-pro). + """ + + def get_supported_openai_params( + self, model: str + ) -> List[OpenAIImageGenerationOptionalParams]: + return ["n", "size"] + + def map_openai_params( + self, + non_default_params: dict, + optional_params: dict, + model: str, + drop_params: bool, + ) -> dict: + supported_params = self.get_supported_openai_params(model) + mapped: dict = {} + for k, v in non_default_params.items(): + if k in optional_params: + continue + if k not in supported_params: + continue + if k == "size": + # Convert "WxH" → "W*H" + mapped["size"] = OPENAI_TO_DASHSCOPE_SIZE.get(v, v.replace("x", "*")) + elif k == "n": + mapped["image_count"] = v + return mapped + + def get_complete_url( + self, + api_base: Optional[str], + api_key: Optional[str], + model: str, + optional_params: dict, + litellm_params: dict, + stream: Optional[bool] = None, + ) -> str: + return ( + api_base + or get_secret_str("DASHSCOPE_API_BASE_IMAGE") + or DEFAULT_API_BASE + ) + + def validate_environment( + self, + headers: dict, + model: str, + messages: List[AllMessageValues], + optional_params: dict, + litellm_params: dict, + api_key: Optional[str] = None, + api_base: Optional[str] = None, + ) -> dict: + final_api_key = api_key or get_secret_str("DASHSCOPE_API_KEY") + if not final_api_key: + raise ValueError("DASHSCOPE_API_KEY is not set") + headers["Authorization"] = f"Bearer {final_api_key}" + headers["Content-Type"] = "application/json" + return headers + + def transform_image_generation_request( + self, + model: str, + prompt: str, + optional_params: dict, + litellm_params: dict, + headers: dict, + ) -> dict: + """ + Transform OpenAI-style image generation request to DashScope multimodal-generation format. + """ + parameters: dict = {} + for k, v in optional_params.items(): + parameters[k] = v + + return { + "model": model, + "input": { + "messages": [ + { + "role": "user", + "content": [{"text": prompt}], + } + ] + }, + "parameters": parameters, + } + + def transform_image_generation_response( + self, + model: str, + raw_response: httpx.Response, + model_response: ImageResponse, + logging_obj: LiteLLMLoggingObj, + request_data: dict, + optional_params: dict, + litellm_params: dict, + encoding: Any, + api_key: Optional[str] = None, + json_mode: Optional[bool] = None, + ) -> ImageResponse: + """ + Transform DashScope response to litellm ImageResponse. + + DashScope response: output.choices[0].message.content[0].image + OpenAI response: data[0].url + """ + try: + response_data = raw_response.json() + except Exception as e: + raise self.get_error_class( + error_message=f"Failed to parse DashScope image generation response: {e}", + status_code=raw_response.status_code, + headers=raw_response.headers, + ) + + if not model_response.data: + model_response.data = [] + + choices = response_data.get("output", {}).get("choices", []) + for choice in choices: + content_list = ( + choice.get("message", {}).get("content", []) + ) + for content_item in content_list: + image_url = content_item.get("image") + if image_url: + model_response.data.append(ImageObject(url=image_url)) + + return model_response diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 05b59d45f9..5f6f433167 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -10383,6 +10383,22 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "dashscope/qwen-image-2.0": { + "litellm_provider": "dashscope", + "mode": "image_generation", + "source": "https://www.alibabacloud.com/help/en/model-studio/models", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "dashscope/qwen-image-2.0-pro": { + "litellm_provider": "dashscope", + "mode": "image_generation", + "source": "https://www.alibabacloud.com/help/en/model-studio/models", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, "databricks/databricks-bge-large-en": { "input_cost_per_token": 1.0003e-07, "input_dbu_cost_per_token": 1.429e-06, diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py index 0efa1d452d..ebe38705d9 100644 --- a/litellm/proxy/proxy_server.py +++ b/litellm/proxy/proxy_server.py @@ -7246,6 +7246,7 @@ async def chat_completion( # noqa: PLR0915 and user_api_key_dict.agent_id is not None ): data["metadata"]["agent_id"] = user_api_key_dict.agent_id + base_llm_response_processor = ProxyBaseLLMRequestProcessing(data=data) try: result = await base_llm_response_processor.base_process_llm_request( diff --git a/litellm/utils.py b/litellm/utils.py index 7a9f62afa0..e1ad1db63e 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -8952,6 +8952,12 @@ class ProviderConfigManager: ) return get_openrouter_image_generation_config(model) + elif LlmProviders.DASHSCOPE == provider: + from litellm.llms.dashscope.image_generation import ( + get_dashscope_image_generation_config, + ) + + return get_dashscope_image_generation_config(model) return None @staticmethod diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 8a28235f98..98723b80aa 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -10397,6 +10397,22 @@ "supports_reasoning": true, "supports_tool_choice": true }, + "dashscope/qwen-image-2.0": { + "litellm_provider": "dashscope", + "mode": "image_generation", + "source": "https://www.alibabacloud.com/help/en/model-studio/models", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, + "dashscope/qwen-image-2.0-pro": { + "litellm_provider": "dashscope", + "mode": "image_generation", + "source": "https://www.alibabacloud.com/help/en/model-studio/models", + "supported_endpoints": [ + "/v1/images/generations" + ] + }, "databricks/databricks-bge-large-en": { "input_cost_per_token": 1.0003e-07, "input_dbu_cost_per_token": 1.429e-06, diff --git a/tests/test_litellm/test_dashscope_image_generation.py b/tests/test_litellm/test_dashscope_image_generation.py new file mode 100644 index 0000000000..b7680a7e7f --- /dev/null +++ b/tests/test_litellm/test_dashscope_image_generation.py @@ -0,0 +1,328 @@ +""" +Unit tests for DashScope image generation support (qwen-image-2.0, qwen-image-2.0-pro). + +Run in docker: pytest tests/test_litellm/test_dashscope_image_generation.py -v +""" + +import json +from unittest.mock import MagicMock, patch + +import httpx +import pytest + +import litellm +from litellm.llms.dashscope.image_generation.transformation import ( + DashScopeImageGenerationConfig, + DEFAULT_API_BASE, +) +from litellm.types.utils import ImageObject, ImageResponse +from litellm.utils import get_llm_provider + + +# --------------------------------------------------------------------------- +# 1. Provider detection +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + "model_string", + [ + "dashscope/qwen-image-2.0", + "dashscope/qwen-image-2.0-pro", + ], +) +def test_get_llm_provider_returns_dashscope(model_string: str): + model, provider, _, _ = get_llm_provider(model_string) + assert provider == "dashscope", f"Expected 'dashscope', got '{provider}'" + assert "qwen-image" in model + + +# --------------------------------------------------------------------------- +# 2. Model info: mode == "image_generation" +# --------------------------------------------------------------------------- + + +@pytest.mark.parametrize( + "model_string, custom_provider", + [ + ("dashscope/qwen-image-2.0", "dashscope"), + ("dashscope/qwen-image-2.0-pro", "dashscope"), + ], +) +def test_get_model_info_mode_is_image_generation(model_string: str, custom_provider: str): + import os + + prev_env = os.environ.get("LITELLM_LOCAL_MODEL_COST_MAP") + prev_model_cost = litellm.model_cost + try: + os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True" + litellm.model_cost = litellm.get_model_cost_map(url="") + + info = litellm.get_model_info(model=model_string, custom_llm_provider=custom_provider) + assert info["mode"] == "image_generation", ( + f"Expected mode='image_generation', got '{info['mode']}'" + ) + finally: + if prev_env is None: + os.environ.pop("LITELLM_LOCAL_MODEL_COST_MAP", None) + else: + os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = prev_env + litellm.model_cost = prev_model_cost + + +# --------------------------------------------------------------------------- +# 3. Request transformation +# --------------------------------------------------------------------------- + + +class TestDashScopeImageGenerationConfig: + def setup_method(self): + self.cfg = DashScopeImageGenerationConfig() + + def test_get_complete_url_default(self): + url = self.cfg.get_complete_url(None, None, "qwen-image-2.0", {}, {}) + assert url == DEFAULT_API_BASE + + def test_get_complete_url_custom(self): + custom = "https://custom.endpoint/generate" + url = self.cfg.get_complete_url(custom, None, "qwen-image-2.0", {}, {}) + assert url == custom + + def test_validate_environment_sets_auth_header(self): + headers = self.cfg.validate_environment( + headers={}, + model="qwen-image-2.0", + messages=[], + optional_params={}, + litellm_params={}, + api_key="sk-test-key", + ) + assert headers["Authorization"] == "Bearer sk-test-key" + assert headers["Content-Type"] == "application/json" + + def test_validate_environment_raises_without_key(self): + with patch("litellm.llms.dashscope.image_generation.transformation.get_secret_str", return_value=None): + with pytest.raises(ValueError, match="DASHSCOPE_API_KEY"): + self.cfg.validate_environment( + headers={}, + model="qwen-image-2.0", + messages=[], + optional_params={}, + litellm_params={}, + api_key=None, + ) + + def test_transform_request_structure(self): + req = self.cfg.transform_image_generation_request( + model="qwen-image-2.0", + prompt="a puppy on green grass", + optional_params={"size": "1024*1024"}, + litellm_params={}, + headers={}, + ) + assert req["model"] == "qwen-image-2.0" + messages = req["input"]["messages"] + assert len(messages) == 1 + assert messages[0]["role"] == "user" + assert messages[0]["content"][0]["text"] == "a puppy on green grass" + assert req["parameters"]["size"] == "1024*1024" + + def test_transform_request_empty_params(self): + req = self.cfg.transform_image_generation_request( + model="qwen-image-2.0-pro", + prompt="sunset over the ocean", + optional_params={}, + litellm_params={}, + headers={}, + ) + assert req["parameters"] == {} + + # --------------------------------------------------------------------------- + # 4. Response transformation + # --------------------------------------------------------------------------- + + def _make_mock_response(self, image_url: str) -> httpx.Response: + body = { + "status_code": 200, + "request_id": "test-request-id", + "output": { + "choices": [ + { + "finish_reason": "stop", + "message": { + "role": "assistant", + "content": [{"image": image_url}], + }, + } + ] + }, + "usage": { + "input_tokens": 0, + "output_tokens": 0, + "width": 1024, + "height": 1024, + "image_count": 1, + }, + } + mock_resp = MagicMock(spec=httpx.Response) + mock_resp.status_code = 200 + mock_resp.headers = {} + mock_resp.json.return_value = body + return mock_resp + + def test_transform_response_extracts_url(self): + image_url = "https://example.oss.aliyuncs.com/generated/test.png" + mock_resp = self._make_mock_response(image_url) + model_response = ImageResponse() + result = self.cfg.transform_image_generation_response( + model="qwen-image-2.0", + raw_response=mock_resp, + model_response=model_response, + logging_obj=MagicMock(), + request_data={}, + optional_params={}, + litellm_params={}, + encoding=None, + ) + assert result.data is not None + assert len(result.data) == 1 + assert result.data[0].url == image_url + + def test_transform_response_multiple_images(self): + body = { + "output": { + "choices": [ + {"finish_reason": "stop", "message": {"role": "assistant", "content": [{"image": "https://example.com/img1.png"}]}}, + {"finish_reason": "stop", "message": {"role": "assistant", "content": [{"image": "https://example.com/img2.png"}]}}, + ] + }, + "usage": {}, + } + mock_resp = MagicMock(spec=httpx.Response) + mock_resp.status_code = 200 + mock_resp.headers = {} + mock_resp.json.return_value = body + + model_response = ImageResponse() + result = self.cfg.transform_image_generation_response( + model="qwen-image-2.0", + raw_response=mock_resp, + model_response=model_response, + logging_obj=MagicMock(), + request_data={}, + optional_params={}, + litellm_params={}, + encoding=None, + ) + assert len(result.data) == 2 + assert result.data[0].url == "https://example.com/img1.png" + assert result.data[1].url == "https://example.com/img2.png" + + # --------------------------------------------------------------------------- + # 5. OpenAI → DashScope parameter mapping + # --------------------------------------------------------------------------- + + def test_map_openai_params_size_conversion(self): + mapped = self.cfg.map_openai_params( + non_default_params={"size": "1024x1024"}, + optional_params={}, + model="qwen-image-2.0", + drop_params=False, + ) + assert mapped["size"] == "1024*1024" + + def test_map_openai_params_n_to_image_count(self): + mapped = self.cfg.map_openai_params( + non_default_params={"n": 2}, + optional_params={}, + model="qwen-image-2.0", + drop_params=False, + ) + assert mapped["image_count"] == 2 + + def test_map_openai_params_unknown_size_uses_asterisk(self): + mapped = self.cfg.map_openai_params( + non_default_params={"size": "768x768"}, + optional_params={}, + model="qwen-image-2.0", + drop_params=False, + ) + assert mapped["size"] == "768*768" + + @pytest.mark.parametrize( + "openai_size, expected", + [ + ("256x256", "256*256"), + ("512x512", "512*512"), + ("1024x1024", "1024*1024"), + ("1792x1024", "1792*1024"), + ("1024x1792", "1024*1792"), + ("2048x2048", "2048*2048"), + ], + ) + def test_map_openai_params_size_table(self, openai_size: str, expected: str): + mapped = self.cfg.map_openai_params( + non_default_params={"size": openai_size}, + optional_params={}, + model="qwen-image-2.0", + drop_params=False, + ) + assert mapped["size"] == expected + + +# --------------------------------------------------------------------------- +# 6. End-to-end flow via litellm.image_generation (HTTP mocked) +# --------------------------------------------------------------------------- + + +def test_litellm_image_generation_dashscope_end_to_end(): + mock_response_body = { + "output": { + "choices": [ + { + "finish_reason": "stop", + "message": { + "role": "assistant", + "content": [ + {"image": "https://dashscope-result.oss.aliyuncs.com/test.png"} + ], + }, + } + ] + }, + "usage": {"input_tokens": 0, "output_tokens": 0, "width": 1024, "height": 1024, "image_count": 1}, + } + + with patch( + "litellm.llms.custom_httpx.llm_http_handler.HTTPHandler.post" + ) as mock_post: + mock_http_response = MagicMock() + mock_http_response.json.return_value = mock_response_body + mock_http_response.status_code = 200 + mock_http_response.headers = {} + mock_post.return_value = mock_http_response + + response = litellm.image_generation( + model="dashscope/qwen-image-2.0", + prompt="a puppy playing on green grass", + api_key="sk-test-key", + size="1024x1024", + ) + + assert response is not None + assert response.data is not None + assert len(response.data) == 1 + assert response.data[0].url == "https://dashscope-result.oss.aliyuncs.com/test.png" + + # Verify the HTTP call was made to the DashScope endpoint + call_args = mock_post.call_args + called_url = call_args[0][0] if call_args[0] else call_args.kwargs.get("url", "") + assert "dashscope" in called_url or "aliyuncs" in called_url + + # Verify request body contains DashScope format + call_kwargs = call_args[1] if call_args[1] else {} + if "json" in call_kwargs: + body = call_kwargs["json"] + assert "input" in body + assert "messages" in body["input"] +