MeUtils 2025.3.4.15.59.15__py3-none-any.whl → 2025.3.5.19.55.22__py3-none-any.whl

meutils/llm/completions/qwenllm.py

@@ -17,18 +17,19 @@ AttributeError: 'str' object has no attribute 'choices'
 from openai import AsyncOpenAI
 
 from meutils.pipe import *
-from meutils.io.files_utils import to_bytes
 from meutils.decorators.retry import retrying
+from meutils.io.files_utils import to_bytes, guess_mime_type
+from meutils.caches import rcache
 
 from meutils.llm.clients import qwen_client
 from meutils.llm.openai_utils import to_openai_params
 
 from meutils.config_utils.lark_utils import get_next_token_for_polling
-from meutils.schemas.openai_types import chat_completion, chat_completion_chunk, ChatCompletionRequest, CompletionUsage
+from meutils.schemas.openai_types import chat_completion, chat_completion_chunk, CompletionRequest, CompletionUsage
 
 FEISHU_URL = "https://xchatllm.feishu.cn/sheets/Bmjtst2f6hfMqFttbhLcdfRJnNf?sheet=PP1PGr"
 
-base_url = "https://chat.qwenlm.ai/api"
+base_url = "https://chat.qwen.ai/api"
 
 from fake_useragent import UserAgent
 
@@ -36,14 +37,18 @@ ua = UserAgent()
 
 
 @retrying()
-async def to_file(file, mime_type: str = "application/octet-stream"):
-    file_bytes = await to_bytes(file)
-    file = ("filename", file_bytes, mime_type)
+@rcache(ttl=3600, serializer='pickle')
+async def to_file(file):
+    filename = Path(file).name if isinstance(file, str) else 'untitled'
+    mime_type = guess_mime_type(file)
+    file_bytes: bytes = await to_bytes(file)
+    file = (filename, file_bytes, mime_type)
     file_object = await qwen_client.files.create(file=file, purpose="file-extract")
+    logger.debug(file_object)
     return file_object
 
 
-async def create(request: ChatCompletionRequest, token: Optional[str] = None):  # ChatCompletionRequest 重构
+async def create(request: CompletionRequest, token: Optional[str] = None):  # ChatCompletionRequest 重构
 
     token = token or await get_next_token_for_polling(feishu_url=FEISHU_URL)
 
@@ -54,11 +59,12 @@ async def create(request: ChatCompletionRequest, token: Optional[str] = None):
     )
 
     # qwen结构
-    if any(i in request.model.lower() for i in ("search",)):
+    model = request.model.lower()
+    if any(i in model for i in ("search",)):
         request.model = "qwen-max-latest"
         request.messages[-1]['chat_type'] = "search"
 
-    if any(i in request.model.lower() for i in ("qwq", "think")):
+    if any(i in model for i in ("qwq", "think")):  # qwq-max-search
         request.model = "qwen-max-latest"
         request.messages[-1]['feature_config'] = {"thinking_enabled": True}
 
@@ -67,6 +73,7 @@ async def create(request: ChatCompletionRequest, token: Optional[str] = None):
     #     # await to_file
     last_message = request.messages[-1]
     logger.debug(last_message)
+
     if last_message.get("role") == "user":
         user_content = last_message.get("content")
         if isinstance(user_content, list):
@@ -79,10 +86,26 @@ async def create(request: ChatCompletionRequest, token: Optional[str] = None):
 
                 elif content.get("type") == 'image_url':
                     url = content.get(content.get("type")).get("url")
-                    file_object = await to_file(url, "image/png")
+                    file_object = await to_file(url)
 
                     user_content[i] = {"type": "image", "image": file_object.id}
 
+        elif user_content.startswith("http"):
+            file_url, user_content = user_content.split(maxsplit=1)
+
+            user_content = [{"type": "text", "text": user_content}]
+
+            file_object = await to_file(file_url)
+
+            content_type = file_object.meta.get("content_type", "")
+            if content_type.startswith("image"):
+                user_content.append({"type": "image", "image": file_object.id})
+            else:
+                user_content.append({"type": "file", "file": file_object.id})
+
+        request.messages[-1]['content'] = user_content
+
+    logger.debug(request)
     data = to_openai_params(request)
     if request.stream:
         _chunk = ""
@@ -133,17 +156,19 @@ if __name__ == '__main__':
         {
             "type": "file_url",
             "file_url": {
-                "url": "https://oss.ffire.cc/files/%E6%8B%9B%E6%A0%87%E6%96%87%E4%BB%B6%E5%A4%87%E6%A1%88%E8%A1%A8%EF%BC%88%E7%AC%AC%E4%BA%8C%E6%AC%A1%EF%BC%89.pdf"
+                "url": "https://oss.ffire.cc/files/AIGC.pdf"
             }
         }
 
     ]
 
-    request = ChatCompletionRequest(
+    request = CompletionRequest(
         # model="qwen-turbo-2024-11-01",
-        # model="qwen-max-latest",
+        model="qwen-max-latest",
         # model="qwen-max-latest-search",
-        model="qwq-max",
+        # model="qwq-max",
+        # model="qwq-max-search",
+
         # model="qwen2.5-vl-72b-instruct",
 
         # model="qwen-plus-latest",
@@ -152,8 +177,8 @@ if __name__ == '__main__':
             {
                 'role': 'user',
                 # 'content': '今天南京天气',
-                'content': "9.8 9.11哪个大",
-                # 'content': '总结下',
+                # 'content': "9.8 9.11哪个大",
+                'content': 'https://oss.ffire.cc/files/AIGC.pdf 总结下',
 
                 # "chat_type": "search",
 
@@ -174,6 +199,22 @@ if __name__ == '__main__':
                 #     }
                 # ]
 
+                # "content": [
+                #     {
+                #         "type": "text",
+                #         "text": "总结下",
+                #         "chat_type": "t2t",
+                #         "feature_config": {
+                #             "thinking_enabled": False
+                #         }
+                #     },
+                #     {
+                #         "type": "file_url",
+                #         "file_url": {
+                #           "url": 'xxxxxxx'
+                #         }
+                #     }
+                # ]
                 # "content": [
                 #     {
                 #         "type": "text",

meutils/llm/completions/yuanbao.py

@@ -9,13 +9,14 @@
 # @Description  :
 import asyncio
 
+import pandas as pd
 from aiostream import stream
 
 from meutils.pipe import *
 from meutils.io.image import image2nowatermark_image
 
 from meutils.llm.utils import oneturn2multiturn
-from meutils.schemas.openai_types import ChatCompletionRequest
+from meutils.schemas.openai_types import CompletionRequest
 from meutils.schemas.image_types import HunyuanImageRequest
 
 from meutils.schemas.yuanbao_types import FEISHU_URL, SSEData, YUANBAO_BASE_URL, API_CHAT, API_GENERATE_ID, \
@@ -41,7 +42,7 @@ class Completions(object):
 
     async def create(
             self,
-            request: Optional[ChatCompletionRequest] = None,
+            request: Optional[CompletionRequest] = None,
             image_request: Optional[HunyuanImageRequest] = None,
             token: Optional[str] = None
     ):
@@ -111,6 +112,7 @@ class Completions(object):
                 logger.debug(response.status_code)
                 response.raise_for_status()
 
+                reasoning = "<think>\n"  # </think>
                 async for chunk in response.aiter_lines():
                     sse = SSEData(chunk=chunk)
                     if image_request and sse.image:
@@ -118,6 +120,24 @@ class Completions(object):
                         yield sse.image
 
                     if request:
+                        if sse.reasoning_content:
+                            yield reasoning
+                            yield sse.reasoning_content
+                            reasoning = ""
+                        elif sse.content and reasoning == "":
+                            reasoning = "\n</think>"
+                            yield reasoning
+
+                        if sse.search_content:
+                            df = pd.DataFrame(sse.search_content)
+                            df['title'] = "[" + df['title'] + "](" + df['url'] + ")"
+                            df['sourceName'] = "![" + df['sourceName'] + "](" + df['icon_url'] + ")"
+                            df = df[['title', 'web_site_name', 'sourceName', "publish_time"]]
+                            df.index += 1
+                            yield '> Search\n'
+                            yield df.to_markdown()
+                            yield '\n\n'
+
                         yield sse.content
 
     def generate_id(self, random: bool = True):
@@ -166,9 +186,15 @@ if __name__ == '__main__':
 
     # async2sync_generator(Completions(api_key).achat('画条狗')) | xprint
     # request = HunyuanImageRequest(prompt='画条狗', size='16:9')
+    # deep_seek deep_seek_v3 hunyuan_t1 hunyuan_gpt_175B_0404
+    # model = 'deep_seek_v3-search'
+    model = 'deep_seek-search'
 
     arun(Completions().create(
-        ChatCompletionRequest(messages=[{'role': 'user', 'content': '南京天气如何'}]),
+        CompletionRequest(
+            model=model,
+            messages=[{'role': 'user', 'content': '南京天气如何'}]
+        ),
         # image_request=request,
         # token=token
     ))

meutils/llm/openai_utils/common.py

@@ -21,7 +21,7 @@ from meutils.apis.oneapi.user import get_user_money, get_api_key_log
 from meutils.apis.oneapi.token import get_api_key_money
 
 from meutils.schemas.oneapi import MODEL_PRICE
-from meutils.schemas.openai_types import ChatCompletionRequest, TTSRequest, STTRequest
+from meutils.schemas.openai_types import CompletionRequest, ChatCompletionRequest, TTSRequest, STTRequest
 from meutils.schemas.openai_types import ChatCompletion, ChatCompletionChunk, CompletionUsage
 from meutils.schemas.openai_types import chat_completion, chat_completion_chunk, chat_completion_chunk_stop  # todo
 
@@ -37,7 +37,7 @@ AUDIO_TRANSCRIPTIONS_PARAMS = get_function_params(fn=OpenAI(api_key='').audio.tr
 
 
 def to_openai_params(
-        request: Union[dict, ChatCompletionRequest, ImageRequest, TTSRequest, STTRequest],
+        request: Union[dict, CompletionRequest, ChatCompletionRequest, ImageRequest, TTSRequest, STTRequest],
         redirect_model: Optional[str] = None,
 ) -> dict:
     data = copy.deepcopy(request)

meutils/schemas/oneapi/common.py

@@ -280,28 +280,28 @@ MODEL_PRICE = {
     # MJ
     "mj-chat": 0.3,
 
-    "mj_fast_blend": 0.08,
+    "mj_fast_blend": 0.1,
     "mj_fast_custom_oom": 0,
-    "mj_fast_describe": 0.04,
-    "mj_fast_high_variation": 0.08,
-    "mj_fast_imagine": 0.08,
+    "mj_fast_describe": 0.05,
+    "mj_fast_high_variation": 0.1,
+    "mj_fast_imagine": 0.1,
     "mj_fast_inpaint": 0,
-    "mj_fast_low_variation": 0.08,
-    "mj_fast_modal": 0.08,
-    "mj_fast_pan": 0.08,
+    "mj_fast_low_variation": 0.1,
+    "mj_fast_modal": 0.1,
+    "mj_fast_pan": 0.1,
     "mj_fast_pic_reader": 0,
     "mj_fast_prompt_analyzer": 0,
     "mj_fast_prompt_analyzer_extended": 0,
-    "mj_fast_reroll": 0.08,
-    "mj_fast_shorten": 0.08,
-    "mj_fast_upload": 0.01,
-    "mj_fast_upscale": 0.04,
-    "mj_fast_upscale_creative": 0.08,
-    "mj_fast_upscale_subtle": 0.08,
-    "mj_fast_variation": 0.08,
-    "mj_fast_zoom": 0.08,
+    "mj_fast_reroll": 0.1,
+    "mj_fast_shorten": 0.1,
+    "mj_fast_upload": 0.1,
+    "mj_fast_upscale": 0.05,
+    "mj_fast_upscale_creative": 0.1,
+    "mj_fast_upscale_subtle": 0.1,
+    "mj_fast_variation": 0.1,
+    "mj_fast_zoom": 0.1,
 
-    "mj_relax_imagine": 0.08 * 0.5,
+    "mj_relax_imagine": 0.05 * MJ_RELAX,
 
     "mj_relax_blend": 0.08,
     "mj_relax_custom_oom": 0,
@@ -401,12 +401,13 @@ MODEL_RATIO = {
     # 阿里千问 https://dashscope.console.aliyun.com/billing
     "qwen-long": 0.25,
     "qwen-turbo": 0.05,
-    "qwen-plus": 2,
+    "qwen-plus": 0.8,
     "qwen-max": 1.2,
     "qwen-max-longcontext": 20,
     "qwen-turbo-2024-11-01": 0.15,
     "qwen-max-latest": 1.2,
     "qwen2.5-max": 1.2,
+    "qwen-max-2025-01-25": 1.2,
 
     "qwen-vl-max-latest": 1.5,
     "qwen-vl-plus-latest": 0.75,
@@ -421,7 +422,6 @@ MODEL_RATIO = {
     "qwen/qwq-32b-preview": 1,
     "Qwen/QwQ-32B-Preview": 1,
 
-
     "qwq-max": 1.2,
     "qwq-max-search": 1.2,
     "qwen-max-search": 1.2,
@@ -771,7 +771,7 @@ COMPLETION_RATIO = {
     "gpt-4-all": 4,
     "gpt-4-gizmo-*": 4,
     "gpt-4o-all": 4,
-    "gpt-4.5-preview-2025-02-27":2,
+    "gpt-4.5-preview-2025-02-27": 2,
 
     "o1-mini": 4,
     "o1-preview": 4,
@@ -862,6 +862,9 @@ COMPLETION_RATIO = {
     "qwen2-vl-72b-instruct": 5,
     "qwen-max-latest": 4,
     "qwen2.5-max": 4,
+    "qwen-max-2025-01-25": 4,
+
+    "qwen-plus": 2.5,
 
     "qwq-max": 4,
     "qwq-max-search": 4,
@@ -1109,7 +1112,6 @@ REDIRECT_MODEL = {
 
 }
 
-
 GROUP_RATIO = {
     "chatfire": 1,
 

meutils/schemas/openai_types.py

@@ -118,25 +118,9 @@ class CompletionRequest(BaseModel):
 
     # 拓展字段
 
-    user_content: Optional[Any] = None  # str dict
-
-    system_messages: Optional[list] = None
-    last_content: Optional[Any] = None
-
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
 
-        # if self.last_message.get("role") == "user":
-        #     user_content = self.last_message.get("content")
-        #     if isinstance(user_content, list):  # 可能是多模态 todo: 多模态 'image_url','video_url' and 'video' 'file' audio
-        #         for i, c in enumerate(user_content):
-        #             if c.get("type") == "image_url":
-        #                 user_content[i]["type"] = "image_url"
-        #                 user_content[i]["image_url"] = user_content[i].get("image_url", {}).get("url", "")
-        #
-        # self.messages = self.messages or [{'role': 'user', 'content': 'hi'}]
-        # self.system_messages = [m for m in self.messages if m.get("role") == "system"]
-
     class Config:
         extra = "allow"
 
@@ -144,11 +128,37 @@ class CompletionRequest(BaseModel):
     def last_message(self):
         return self.messages and self.messages[-1]
 
-    # @cached_property
-    def last_user_content(self):
-        return "xxxxxxxx"
-        # if self.last_message.get("role") == "user":
-        #     return self.last_message.get("content")
+    @cached_property
+    def last_user_content(self) -> str:
+        for i, message in enumerate(self.messages[::-1], 1):
+            if message.get("role") == "user":
+                user_contents = message.get("content")
+                if isinstance(user_contents, list):
+                    for content in user_contents:
+                        return content.get('text', "")
+                else:
+                    return user_contents
+
+    @cached_property
+    def last_urls(self):  # file_url 多轮对话需要  sum(request.last_urls.values(), [])
+        content_types = {"image_url", "file", "file_url", "video_url", "audio_url"}
+        for i, message in enumerate(self.messages[::-1], 1):
+            data = {}
+            if message.get("role") == "user":  # 每一轮还要处理
+                user_contents = message.get("content")
+                if isinstance(user_contents, list):  # 用户 url
+                    for content in user_contents:
+                        content_type = content.get("type")
+                        if content_type in content_types:
+                            if _url := content.get(content_type, {}):  # {"type": "file", "file": fileid}
+                                if isinstance(_url, str):  # 兼容了spark qwenai
+                                    url = _url
+                                else:
+                                    url = _url.get("url")
+                                url and data.setdefault(content_type, []).append(url)
+
+            if data: return data
+        return {}
 
 
 class ChatCompletionRequest(BaseModel):
@@ -416,14 +426,29 @@ if __name__ == '__main__':
     #
     #
     # print(A(n=11))
-    # messages = [
-    #     {
-    #         "role": "user",
-    #         "content": [{'role': 'user', 'content': [{"type": "image_url", "image_url": "这是个图片链接"}]}]
-    #     },
-    #
-    #     # {'role': 'user', 'content': [{"type": "image_url", "image_url": {"url": "这是个图片链接"}}]},
-    # ]
+    messages = [
+        {'role': 'user',
+         'content': [{"type": "image_url", "image_url": "https://oss.ffire.cc/files/kling_watermark.png"}]},
+        {'role': 'assistant', 'content': [{"type": "image_url", "image_url": "这是个图片链接"}]},
+
+        {'role': 'assistant', 'content': [{"type": "image_url", "image_url": "这是个图片链接"}]},
+
+        {'role': 'user', 'content': [
+            {"type": "image_url", "image_url": {"url": "这是个图片链接1"}},
+            {"type": "file_url", "file_url": {"url": "这是个file_url"}},
+            {"type": "file_url", "file_url": {"url": "这是个file_url"}},
+            {"type": "file", "file": "这是个fileid"},
+
+            {"type": "audio_url", "audio_url": {"url": "这是个file_url"}},
+            {"type": "video_url", "video_url": {"url": "这是个video_url"}}
+        ]},
+
+        {'role': 'assistant', 'content': [{"type": "image_url", "image_url": "这是个图片链接"}]},
+        {'role': 'user', 'content': [
+            {"type": "text", "text": "这是个文本"},
+            {"type": "image_url", "image_url": "这是个图片链接"}
+        ]},
+    ]
     #
     # r = ChatCompletionRequest(model="gpt-3.5-turbo", messages=messages)
     # r.messages[-1]['content'] = [{"type": "image_url", "image_url": {"url": r.urls[-1]}}]
@@ -433,4 +458,5 @@ if __name__ == '__main__':
     # print(chat_completion)
     # print(chat_completion_chunk_stop)
 
-    print(CompletionRequest().last_user_content())
+    # print(CompletionRequest(messages=messages).last_urls)
+    print(CompletionRequest(messages=messages).last_user_content)

meutils/schemas/yuanbao_types.py

@@ -7,6 +7,7 @@
 # @WeChat       : meutils
 # @Software     : PyCharm
 # @Description  :
+import pandas as pd
 
 from meutils.pipe import *
 
@@ -221,6 +222,7 @@ class SSEData(BaseModel):
 
     content: str = ""
     reasoning_content: str = ""
+    search_content: str = ""
 
     image: Optional[str] = None
 
@@ -229,10 +231,11 @@ class SSEData(BaseModel):
     def __init__(self, **data):
         super().__init__(**data)
 
-        logger.debug(self.chunk)
+        # logger.debug(self.chunk)
 
         chunk = self.chunk.lstrip("data:")
 
+        content = ""
         if '"type":"progress"' in chunk:
             content = json.loads(chunk).get("msg", "")
 
@@ -254,14 +257,10 @@ class SSEData(BaseModel):
         # df['image'] = "![](" + df['image'] + ")"
 
         elif '{"type":"think"' in chunk:  # 思考中...
-            content = self.reasoning_content = json.loads(chunk).get("content", "")
+            self.reasoning_content = json.loads(chunk).get("content", "")
 
         elif '{"type":"searchGuid"' in chunk:  # 思考中...
-            content = self.reasoning_content = json.loads(chunk).get("docs", "")
-
-        else:
-            content = ""
-            # chunk.strip() or logger.debug(chunk)  # debug
+            self.search_content = json.loads(chunk).get("docs", "")
 
         self.content = content
 

meutils/types.py

@@ -35,3 +35,5 @@ def is_list_of_strings(lst):
 
 def is_list_of_ints(lst):
     return isinstance(lst, List) and all(isinstance(item, int) for item in lst)
+
+