feat(config): 新增流式输出优化器开关配置

在环境变量示例文件(.env.example)和配置类(config.py)中新增 STREAM_OPTIMIZER_ENABLED 配置项，用于控制流式输出优化器的启用状态，默认设为 false

调整 Gemini 和 OpenAI 聊天服务的流式响应处理逻辑：
- 仅在流式优化器启用时（settings.STREAM_OPTIMIZER_ENABLED 为 true）
- 才会对文本内容执行流式输出优化处理
- 保持原有文本提取逻辑不变，仅增加配置条件判断

该变更使流式输出优化器变为可选功能，方便根据实际需求进行开关控制

.env.example

@@ -10,6 +10,8 @@ SHOW_SEARCH_LINK=true
 SHOW_THINKING_PROCESS=true
 BASE_URL=https://generativelanguage.googleapis.com/v1beta
 MAX_FAILURES=10
+# 请求超时时间（秒）
+TIME_OUT=300
 #########################image_generate 相关配置###########################
 PAID_KEY=AIzaSyxxxxxxxxxxxxxxxxxxx
 CREATE_IMAGE_MODEL=imagen-3.0-generate-002
@@ -20,6 +22,7 @@ CLOUDFLARE_IMGBED_URL=https://xxxxxxx.pages.dev/upload
 CLOUDFLARE_IMGBED_AUTH_CODE=xxxxxxxxx
 ##########################################################################
 #########################stream_optimizer 相关配置########################
+STREAM_OPTIMIZER_ENABLED=false
 STREAM_MIN_DELAY=0.016
 STREAM_MAX_DELAY=0.024
 STREAM_SHORT_TEXT_THRESHOLD=10

.github/workflows/docker-publish.yml

@@ -2,8 +2,6 @@ name: Docker Image CI
 
 on:
   push:
-    # branches: [ "main" ]
-    tags: [ 'v*.*.*' ]
   pull_request:
     branches: [ "main" ]
 
@@ -43,20 +41,30 @@ jobs:
         with:
           images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
           tags: |
-            type=raw,value=latest,enable={{is_default_branch}}
+            # https://github.com/docker/metadata-action/tree/v5/?tab=readme-ov-file#semver
+            # Event: push,     Ref: refs/head/main,       Tags: main
+            # Event: push tag, Ref: refs/tags/v1.2.3,     Tags: 1.2.3, 1.2, 1, latest
+            # Event: push tag, Ref: refs/tags/v2.0.8-rc1, Tags: 2.0.8-rc1
+            type=ref,event=branch
             type=semver,pattern={{version}}
             type=semver,pattern={{major}}.{{minor}}
-            type=sha,format=long
+            type=semver,pattern={{major}}
           labels: |
             org.opencontainers.image.description=OpenAI API Compatible Server
             org.opencontainers.image.source=${{ github.event.repository.html_url }}
 
-      - name: Build and push Docker image
-        uses: docker/build-push-action@v5
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v3
+
+      - name: Build and push
+        uses: docker/build-push-action@v6
         with:
+          file: Dockerfile
           context: .
+          platforms: linux/amd64,linux/arm64
           push: ${{ github.event_name != 'pull_request' }}
+          load: false
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
+          cache-from: type=gha,scope=${{ github.workflow }}
+          cache-to: type=gha,scope=${{ github.workflow }}

README.md

@@ -1,4 +1,4 @@
-# 🚀 FastAPI OpenAI (Gemini) 代理服务
+# 🚀 Gemini 代理服务（支持openai/gemini格式）
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 

app/config/config.py

@@ -4,7 +4,7 @@
 from typing import List
 from pydantic_settings import BaseSettings
 
-from app.core.constants import API_VERSION, DEFAULT_CREATE_IMAGE_MODEL, DEFAULT_FILTER_MODELS, DEFAULT_MODEL, DEFAULT_STREAM_CHUNK_SIZE, DEFAULT_STREAM_LONG_TEXT_THRESHOLD, DEFAULT_STREAM_MAX_DELAY, DEFAULT_STREAM_MIN_DELAY, DEFAULT_STREAM_SHORT_TEXT_THRESHOLD
+from app.core.constants import API_VERSION, DEFAULT_CREATE_IMAGE_MODEL, DEFAULT_FILTER_MODELS, DEFAULT_MODEL, DEFAULT_STREAM_CHUNK_SIZE, DEFAULT_STREAM_LONG_TEXT_THRESHOLD, DEFAULT_STREAM_MAX_DELAY, DEFAULT_STREAM_MIN_DELAY, DEFAULT_STREAM_SHORT_TEXT_THRESHOLD, DEFAULT_TIMEOUT
 
 
 class Settings(BaseSettings):
@@ -16,6 +16,7 @@ class Settings(BaseSettings):
     AUTH_TOKEN: str = ""
     MAX_FAILURES: int = 3
     TEST_MODEL: str = DEFAULT_MODEL
+    TIME_OUT: int = DEFAULT_TIMEOUT
     
     # 模型相关配置
     SEARCH_MODELS: List[str] = ["gemini-2.0-flash-exp"]
@@ -35,6 +36,7 @@ class Settings(BaseSettings):
     CLOUDFLARE_IMGBED_AUTH_CODE: str = ""
     
     # 流式输出优化器配置
+    STREAM_OPTIMIZER_ENABLED: bool = False
     STREAM_MIN_DELAY: float = DEFAULT_STREAM_MIN_DELAY
     STREAM_MAX_DELAY: float = DEFAULT_STREAM_MAX_DELAY
     STREAM_SHORT_TEXT_THRESHOLD: int = DEFAULT_STREAM_SHORT_TEXT_THRESHOLD

app/domain/gemini_models.py

@@ -1,6 +1,8 @@
 from typing import List, Optional, Dict, Any, Literal, Union
 from pydantic import BaseModel
 
+from app.core.constants import DEFAULT_TEMPERATURE, DEFAULT_TOP_K, DEFAULT_TOP_P
+
 
 class SafetySetting(BaseModel):
     category: Optional[Literal["HARM_CATEGORY_HATE_SPEECH", "HARM_CATEGORY_DANGEROUS_CONTENT", "HARM_CATEGORY_HARASSMENT", "HARM_CATEGORY_SEXUALLY_EXPLICIT", "HARM_CATEGORY_CIVIC_INTEGRITY"]] = None
@@ -13,9 +15,9 @@ class GenerationConfig(BaseModel):
     responseSchema: Optional[Dict[str, Any]] = None
     candidateCount: Optional[int] = 1
     maxOutputTokens: Optional[int] = None
-    temperature: Optional[float] = None
-    topP: Optional[float] = None
-    topK: Optional[int] = None
+    temperature: Optional[float] = DEFAULT_TEMPERATURE
+    topP: Optional[float] = DEFAULT_TOP_P
+    topK: Optional[int] = DEFAULT_TOP_K
     presencePenalty: Optional[float] = None
     frequencyPenalty: Optional[float] = None
     responseLogprobs: Optional[bool] = None

app/domain/openai_models.py

@@ -1,7 +1,7 @@
 from pydantic import BaseModel
 from typing import List, Optional, Union
 
-from app.core.constants import DEFAULT_MAX_TOKENS, DEFAULT_MODEL, DEFAULT_TEMPERATURE, DEFAULT_TOP_K, DEFAULT_TOP_P
+from app.core.constants import DEFAULT_MODEL, DEFAULT_TEMPERATURE, DEFAULT_TOP_K, DEFAULT_TOP_P
 
 
 class ChatRequest(BaseModel):
@@ -10,7 +10,7 @@ class ChatRequest(BaseModel):
     temperature: Optional[float] = DEFAULT_TEMPERATURE
     stream: Optional[bool] = False
     tools: Optional[List[dict]] = []
-    max_tokens: Optional[int] = DEFAULT_MAX_TOKENS
+    max_tokens: Optional[int] = None
     top_p: Optional[float] = DEFAULT_TOP_P
     top_k: Optional[int] = DEFAULT_TOP_K
     stop: Optional[List[str]] = []

app/handler/stream_optimizer.py

@@ -107,15 +107,15 @@ class StreamOptimizer:
 
         # 计算智能延迟时间
         delay = self.calculate_delay(len(text))
-        if self.logger:
-            self.logger.info(f"Text length: {len(text)}, delay: {delay:.4f}s")
+        # if self.logger:
+        #     self.logger.info(f"Text length: {len(text)}, delay: {delay:.4f}s")
 
         # 根据文本长度决定输出方式
         if len(text) >= self.long_text_threshold:
             # 长文本：分块输出
             chunks = self.split_text_into_chunks(text)
-            if self.logger:
-                self.logger.info(f"Long text: splitting into {len(chunks)} chunks")
+            # if self.logger:
+            #     self.logger.info(f"Long text: splitting into {len(chunks)} chunks")
             for chunk_text in chunks:
                 chunk_response = create_response_chunk(chunk_text)
                 yield format_chunk(chunk_response)

app/router/gemini_routes.py

@@ -41,7 +41,7 @@ async def list_models(
     logger.info("-" * 50 + "list_gemini_models" + "-" * 50)
     logger.info("Handling Gemini models list request")
     
-    api_key = await key_manager.get_next_working_key()
+    api_key = await key_manager.get_first_valid_key()
     logger.info(f"Using API key: {api_key}")
     
     models_json = model_service.get_gemini_models(api_key)
@@ -86,7 +86,7 @@ async def list_models(
 async def generate_content(
     model_name: str,
     request: GeminiRequest,
-    _=Depends(security_service.verify_goog_api_key),
+    _=Depends(security_service.verify_key_or_goog_api_key),
     api_key: str = Depends(get_next_working_key),
     key_manager: KeyManager = Depends(get_key_manager)
 ):
@@ -118,7 +118,7 @@ async def generate_content(
 async def stream_generate_content(
     model_name: str,
     request: GeminiRequest,
-    _=Depends(security_service.verify_goog_api_key),
+    _=Depends(security_service.verify_key_or_goog_api_key),
     api_key: str = Depends(get_next_working_key),
     key_manager: KeyManager = Depends(get_key_manager)
 ):

app/router/openai_routes.py

@@ -44,7 +44,7 @@ async def list_models(
 ):
     logger.info("-" * 50 + "list_models" + "-" * 50)
     logger.info("Handling models list request")
-    api_key = await key_manager.get_next_working_key()
+    api_key = await key_manager.get_first_valid_key()
     logger.info(f"Using API key: {api_key}")
     try:
         return model_service.get_gemini_openai_models(api_key)

app/service/chat/gemini_chat_service.py

@@ -89,6 +89,11 @@ def _get_safety_settings(model: str) -> List[Dict[str, str]]:
 def _build_payload(model: str, request: GeminiRequest) -> Dict[str, Any]:
     """构建请求payload"""
     request_dict = request.model_dump()
+    if request.generationConfig:
+        if request.generationConfig.maxOutputTokens is None:
+            # 如果未指定最大输出长度，则不传递该字段，解决截断的问题
+            request_dict["generationConfig"].pop("maxOutputTokens")
+    
     payload = {
         "contents": request_dict.get("contents", []),
         "tools": _build_tools(model, request_dict),
@@ -162,9 +167,8 @@ class GeminiChatService:
                             json.loads(line), model, stream=True
                         )
                         text = self._extract_text_from_response(response_data)
-
-                        # 如果有文本内容，使用流式输出优化器处理
-                        if text:
+                        # 如果有文本内容，且开启了流式输出优化器，则使用流式输出优化器处理
+                        if text and settings.STREAM_OPTIMIZER_ENABLED:
                             # 使用流式输出优化器处理文本输出
                             async for (
                                 optimized_chunk

app/service/chat/openai_chat_service.py

@@ -115,7 +115,6 @@ def _build_payload(
         "contents": messages,
         "generationConfig": {
             "temperature": request.temperature,
-            "maxOutputTokens": request.max_tokens,
             "stopSequences": request.stop,
             "topP": request.top_p,
             "topK": request.top_k,
@@ -123,6 +122,8 @@ def _build_payload(
         "tools": _build_tools(request, messages),
         "safetySettings": _get_safety_settings(request.model),
     }
+    if request.max_tokens is not None:
+        payload["generationConfig"]["maxOutputTokens"] = request.max_tokens
     if request.model.endswith("-image") or request.model.endswith("-image-generation"):
         payload["generationConfig"]["responseModalities"] = ["Text", "Image"]
 
@@ -214,7 +215,7 @@ class OpenAIChatService:
                         if openai_chunk:
                             # 提取文本内容
                             text = self._extract_text_from_openai_chunk(openai_chunk)
-                            if text:
+                            if text and settings.STREAM_OPTIMIZER_ENABLED:
                                 # 使用流式输出优化器处理文本输出
                                 async for (
                                     optimized_chunk

app/service/key/key_manager.py

@@ -81,6 +81,13 @@ class KeyManager:
 
         return {"valid_keys": valid_keys, "invalid_keys": invalid_keys}
 
+    async def get_first_valid_key(self) -> str:
+        """获取第一个有效的API key"""
+        async with self.failure_count_lock:
+            for key in self.key_failure_counts:
+                if self.key_failure_counts[key] < self.MAX_FAILURES:
+                    return key
+        return self.api_keys[0]
 
 _singleton_instance = None
 _singleton_lock = asyncio.Lock()