Add Kimi provider, add vision support to LMArenaBeta

.gitignore

@@ -37,3 +37,4 @@ projects/windows/
 *.bak
 *.backup
 .env
+g4f.dev/

docker/Dockerfile

@@ -19,7 +19,7 @@ RUN if [ "$G4F_VERSION" = "" ] ; then \
 RUN apt-get -qqy update \
   && apt-get -qqy upgrade \
   && apt-get -qyy autoremove \
-  && apt-get -qqy install python3 python-is-python3 pip ffmpeg flac \
+  && apt-get -qqy install python3 python-is-python3 pip ffmpeg flac libavcodec-extra \
   && apt-get -qyy remove openjdk-11-jre-headless \
   && apt-get -qyy autoremove \
   && apt-get -qyy clean \

g4f.dev

@@ -1 +0,0 @@
-Subproject commit b3a9831dd9b10e90f17bcf6524ff48863ac8112d

g4f/Provider/Kimi.py

@@ -0,0 +1,104 @@
+from __future__ import annotations
+
+import random
+from typing import AsyncIterator
+
+from .base_provider import AsyncAuthedProvider, ProviderModelMixin
+from ..providers.helper import get_last_user_message
+from ..requests import StreamSession, see_stream
+from ..providers.response import AuthResult, TitleGeneration, JsonConversation, FinishReason
+from ..typing import AsyncResult, Messages
+
+class Kimi(AsyncAuthedProvider, ProviderModelMixin):
+    url = "https://www.kimi.com"
+    working = True
+    active_by_default = True
+    default_model = "kimi-k2"
+    models = [default_model]
+
+    @classmethod
+    async def on_auth_async(cls, proxy: str = None, **kwargs) -> AsyncIterator:
+        device_id = str(random.randint(1000000000000000, 9999999999999999))
+        async with StreamSession(proxy=proxy, impersonate="chrome") as session:
+            async with session.post(
+                "https://www.kimi.com/api/device/register",
+                json={},
+                headers={
+                    "x-msh-device-id": device_id,
+                    "x-msh-platform": "web",
+                    "x-traffic-id": device_id
+                }
+            ) as response:
+                if response.status != 200:
+                    raise Exception("Failed to register device")
+            data = await response.json()
+            if not data.get("access_token"):
+                raise Exception("No access token received")
+        yield AuthResult(
+            api_key=data.get("access_token"),
+            device_id=device_id,
+        )
+
+    @classmethod
+    async def create_authed(
+        cls,
+        model: str,
+        messages: Messages,
+        auth_result: AuthResult,
+        proxy: str = None,
+        conversation: JsonConversation = None,
+        web_search: bool = False,
+        **kwargs
+    ) -> AsyncResult:
+        pass
+        async with StreamSession(
+            proxy=proxy,
+            impersonate="chrome",
+            headers={
+                "Authorization": f"Bearer {auth_result.api_key}",
+            }
+        ) as session:
+            if conversation is None:
+                async with session.post("https://www.kimi.com/api/chat", json={
+                    "name":"未命名会话",
+                    "born_from":"home",
+                    "kimiplus_id":"kimi",
+                    "is_example":False,
+                    "source":"web",
+                    "tags":[]
+                }) as response:
+                    if response.status != 200:
+                        raise Exception("Failed to create chat")
+                    chat_data = await response.json()
+                conversation = JsonConversation(chat_id=chat_data.get("id"))
+            data = {
+                "kimiplus_id": "kimi",
+                "extend": {"sidebar": True},
+                "model": model,
+                "use_search": web_search,
+                "messages": [
+                    {
+                        "role": "user",
+                        "content": get_last_user_message(messages)
+                    }
+                ],
+                "refs": [],
+                "history": [],
+                "scene_labels": [],
+                "use_semantic_memory": False,
+                "use_deep_research": False
+            }
+            async with session.post(
+                f"https://www.kimi.com/api/chat/{conversation.chat_id}/completion/stream",
+                json=data
+            ) as response:
+                if response.status != 200:
+                    raise Exception("Failed to start chat completion")
+                async for line in see_stream(response):
+                    if line.get("event") == "cmpl":
+                        yield line.get("text")
+                    elif line.get("event") == "rename":
+                        yield TitleGeneration(line.get("text"))
+                    elif line.get("event") == "all_done":
+                        yield FinishReason("stop")
+                        break

g4f/Provider/__init__.py

@@ -41,6 +41,7 @@ from .DeepInfraChat        import DeepInfraChat
 from .DuckDuckGo           import DuckDuckGo
 from .Free2GPT             import Free2GPT
 from .ImageLabs            import ImageLabs
+from .Kimi                 import Kimi
 from .LambdaChat           import LambdaChat
 from .LegacyLMArena        import LegacyLMArena
 from .OIVSCodeSer2         import OIVSCodeSer2

g4f/Provider/needs_auth/LMArenaBeta.py

@@ -4,12 +4,14 @@ import time
 import uuid
 import json
 import asyncio
+import os
 
-from ...typing import AsyncResult, Messages
+from ...typing import AsyncResult, Messages, MediaListType
 from ...requests import StreamSession, get_args_from_nodriver, raise_for_status, merge_cookies
 from ...requests import DEFAULT_HEADERS, has_nodriver
 from ...errors import ModelNotFoundError
 from ...providers.response import FinishReason, Usage, JsonConversation, ImageResponse
+from ...tools.media import merge_media
 from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin,AuthFileMixin
 from ..helper import get_last_user_message
 from ... import debug
@@ -114,6 +116,7 @@ models = [
 ]
 text_models = {model["publicName"]: model["id"] for model in models if "text" in model["capabilities"]["outputCapabilities"]}
 image_models = {model["publicName"]: model["id"] for model in models if "image" in model["capabilities"]["outputCapabilities"]}
+vision_models = [model["publicName"] for model in models if "image" in model["capabilities"]["inputCapabilities"]]
 
 class LMArenaBeta(AsyncGeneratorProvider, ProviderModelMixin, AuthFileMixin):
     label = "LMArena (New)"
@@ -124,6 +127,7 @@ class LMArenaBeta(AsyncGeneratorProvider, ProviderModelMixin, AuthFileMixin):
     default_model = list(text_models.keys())[0]
     models = list(text_models) + list(image_models)
     image_models = list(image_models)
+    vision_models = vision_models
 
     @classmethod
     async def create_async_generator(
@@ -131,6 +135,7 @@ class LMArenaBeta(AsyncGeneratorProvider, ProviderModelMixin, AuthFileMixin):
         model: str,
         messages: Messages,
         conversation: JsonConversation = None,
+        media: MediaListType = None,
         proxy: str = None,
         timeout: int = None,
         **kwargs
@@ -181,7 +186,15 @@ class LMArenaBeta(AsyncGeneratorProvider, ProviderModelMixin, AuthFileMixin):
                     "id": userMessageId,
                     "role": "user",
                     "content": prompt,
-                    "experimental_attachments": [],
+                    "experimental_attachments": [
+                        {
+                            "name": name or os.path.basename(url),
+                            "contentType": get_content_type(url),
+                            "url": url
+                        }
+                        for url, name in list(merge_media(media, messages))
+                        if url.startswith("https://")
+                    ],
                     "parentMessageIds": [] if conversation is None else conversation.message_ids,
                     "participantPosition": "a",
                     "modelId": None,
@@ -233,3 +246,13 @@ class LMArenaBeta(AsyncGeneratorProvider, ProviderModelMixin, AuthFileMixin):
         # Save the args to cache file
         with cache_file.open("w") as f:
             json.dump(args, f)
+
+def get_content_type(url: str) -> str:
+    if url.endswith(".webp"):
+        return "image/webp"
+    elif url.endswith(".png"):
+        return "image/png"
+    elif url.endswith(".jpg") or url.endswith(".jpeg"):
+        return "image/jpeg"
+    else:
+        return "application/octet-stream"

g4f/tools/run_tools.py

@@ -129,6 +129,7 @@ class AuthManager:
         "GeminiPro": "Gemini",
         "PollinationsAI": "Pollinations",
         "OpenaiAPI": "Openai",
+        "PuterJS": "Puter",
     }
 
     @classmethod