gemini-rproxy

Paused

App Files Files Community

Moonfanz commited on Jan 11

Commit

c06ccd5

verified ·

1 Parent(s): 0208b24

Upload 4 files

Browse files

Files changed (2) hide show

app.py +131 -87
func.py +2 -8

app.py CHANGED Viewed

@@ -1,5 +1,5 @@
 from flask import Flask, request, jsonify, Response, stream_with_context, render_template_string
-from google.generativeai.types import BlockedPromptException, StopCandidateException, generation_types
 from google.api_core.exceptions import InvalidArgument, ResourceExhausted, Aborted, InternalServerError, ServiceUnavailable, PermissionDenied
 import google.generativeai as genai
 import json
@@ -13,6 +13,12 @@ import time
 import requests
 from collections import deque
 import random
 os.environ['TZ'] = 'Asia/Shanghai'
@@ -38,7 +44,7 @@ request_counts = {}
 api_key_blacklist = set()
 api_key_blacklist_duration = 60
-# 核心优势
 safety_settings = [
     {
         "category": "HARM_CATEGORY_HARASSMENT",
@@ -75,6 +81,44 @@ safety_settings_g2 = [
         "threshold": "OFF"
     }
 ]
 class APIKeyManager:
     def __init__(self):
         self.api_keys = re.findall(r"AIzaSy[a-zA-Z0-9_-]{33}", os.environ.get('KeyArray'))
@@ -138,7 +182,7 @@ GEMINI_MODELS = [
 @app.route('/')
 def index():
-    main_content = "Moonfanz Reminiproxy v2.2.1 2025-01-11"
     html_template = """
 <!DOCTYPE html>
 <html>
@@ -303,113 +347,110 @@ def chat_completions():
     max_tokens = request_data.get('max_tokens', 8192)
     stream = request_data.get('stream', False)
     hint = "流式" if stream else "非流"
-    logger.info(f"\n{model} [{hint}] → {current_api_key[:11]}...")
-    gemini_history, user_message, system_instruction, error_response = func.process_messages_for_gemini(messages)
-    # r_g = json.dumps(gemini_history, indent=4, ensure_ascii=False).replace('\\n', '\n')
-    # r_u = json.dumps(user_message, indent=4, ensure_ascii=False).replace('\\n', '\n')
-    # r_s = json.dumps(system_instruction, indent=4, ensure_ascii=False).replace('\\n', '\n')
-    # logger.info(f"历史对话: {r_g}")
-    # logger.info(f"用户消息: {r_u}")
-    # logger.info(f"系统指令: {r_s}")
     if error_response:
         logger.error(f"处理输入消息时出错↙\n {error_response}")
         return jsonify(error_response), 400
     def do_request(current_api_key, attempt):
-        isok, time = is_within_rate_limit(current_api_key)
         if not isok:
-            logger.warning(f"{current_api_key[:11]} → 暂时超过限额，该API key将在 {time} 秒后启用...")
             switch_api_key()
             return 0, None
         increment_request_count(current_api_key)
-        genai.configure(api_key=current_api_key)
-        generation_config = {
-            "temperature": temperature,
-            "max_output_tokens": max_tokens
         }
-        gen_model = genai.GenerativeModel(
-            model_name=model,
-            generation_config=generation_config,
-            # safety_settings=safety_settings_g2 if model == 'gemini-2.0-flash-exp' else safety_settings,
-            safety_settings=safety_settings,
-            system_instruction=system_instruction
-        )
         try:
-            if gemini_history:
-                chat_session = gen_model.start_chat(history=gemini_history)
-                response = chat_session.send_message(user_message, stream=stream)
             else:
-                response = gen_model.generate_content(user_message, stream=stream)
-            return 1, response
-        except Exception as e:
             return handle_api_error(e, attempt)
-    def generate(response):
         try:
-            logger.info(f"流式开始...")
-            for chunk in response:
-                if chunk.text:
-                    data = {
-                        'choices': [
-                            {
-                                'delta': {
-                                    'content': chunk.text
-                                },
-                                'finish_reason': None,
-                                'index': 0
-                            }
-                        ],
-                        'object': 'chat.completion.chunk'
-                    }
-                    yield f"data: {json.dumps(data)}\n\n"
-            data = {
-                        'choices': [
-                            {
-                                'delta': {},
-                                'finish_reason': 'stop',
-                                'index': 0
-                            }
-                        ],
-                        'object': 'chat.completion.chunk'
-                    }
-            logger.info(f"流式结束")
-            yield f"data: {json.dumps(data)}\n\n"
-            logger.info(f"200!")
-        except Exception:
-            logger.error(f"流式输出中途被截断，请关闭流式输出或修改你的输入")
-            logger.info(f"流式结束")
-            error_data = {
-                'error': {
-                    'message': '流式输出时截断，请关闭流式输出或修改你的输入',
-                    'type': 'internal_server_error'
-                }
-            }
-            yield f"data: {json.dumps(error_data)}\n\n"
-            data = {
-                        'choices': [
-                            {
-                                'delta': {},
-                                'finish_reason': 'stop',
-                                'index': 0
-                            }
-                        ],
-                        'object': 'chat.completion.chunk'
-                    }
-            yield f"data: {json.dumps(data)}\n\n"
     attempt = 0
     success = 0
     response = None
     for attempt in range(1, MAX_RETRIES + 1):
         logger.info(f"第 {attempt}/{MAX_RETRIES} 次尝试 ...")
         success, response = do_request(current_api_key, attempt)
@@ -438,7 +479,10 @@ def chat_completions():
         return jsonify(response), 500 if response is not None else 503
     if stream:
-        return Response(stream_with_context(generate(response)), mimetype='text/event-stream')
     else:
         try:
             text_content = response.text
@@ -500,7 +544,7 @@ if __name__ == '__main__':
     scheduler.add_job(keep_alive, 'interval', hours=12)
     scheduler.start()
-    logger.info(f"Reminiproxy v2.2.1 启动")
     logger.info(f"最大尝试次数/MaxRetries: {MAX_RETRIES}")
     logger.info(f"最大请求次数/MaxRequests: {MAX_REQUESTS}")
     logger.info(f"请求限额窗口/LimitWindow: {LIMIT_WINDOW} 秒")

 from flask import Flask, request, jsonify, Response, stream_with_context, render_template_string
+from google.generativeai.types import StopCandidateException, generation_types
 from google.api_core.exceptions import InvalidArgument, ResourceExhausted, Aborted, InternalServerError, ServiceUnavailable, PermissionDenied
 import google.generativeai as genai
 import json
 import requests
 from collections import deque
 import random
+import codecs
+import json
+from dataclasses import dataclass
+from typing import Optional, Iterator, Dict, Any
+app = Flask(__name__)
 os.environ['TZ'] = 'Asia/Shanghai'
 api_key_blacklist = set()
 api_key_blacklist_duration = 60
 safety_settings = [
     {
         "category": "HARM_CATEGORY_HARASSMENT",
         "threshold": "OFF"
     }
 ]
+@dataclass
+class GeneratedText:
+    """用于存储生成的文本片段"""
+    text: str
+    finish_reason: Optional[str] = None
+class ResponseWrapper:
+    """处理非流式响应的包装类"""
+    def __init__(self, data: Dict[Any, Any]):
+        self._data = data
+        self._text = self._extract_text()
+        self._finish_reason = self._extract_finish_reason()
+    def _extract_text(self) -> str:
+        """从响应数据中提取文本"""
+        try:
+            return self._data['candidates'][0]['content']['parts'][0]['text']
+        except (KeyError, IndexError):
+            return ""
+    def _extract_finish_reason(self) -> Optional[str]:
+        """提取完成原因"""
+        try:
+            return self._data['candidates'][0].get('finishReason')
+        except (KeyError, IndexError):
+            return None
+    @property
+    def text(self) -> str:
+        """获取响应文本"""
+        return self._text
+    @property
+    def finish_reason(self) -> Optional[str]:
+        """获取完成原因"""
+        return self._finish_reason
 class APIKeyManager:
     def __init__(self):
         self.api_keys = re.findall(r"AIzaSy[a-zA-Z0-9_-]{33}", os.environ.get('KeyArray'))
 @app.route('/')
 def index():
+    main_content = "Moonfanz Reminiproxy v2.3.0 2025-01-11"
     html_template = """
 <!DOCTYPE html>
 <html>
     max_tokens = request_data.get('max_tokens', 8192)
     stream = request_data.get('stream', False)
     hint = "流式" if stream else "非流"
+    logger.info(f"\n{model} [{hint}] → ...")
+    is_thinking = 'thinking' in model
+    api_version = 'v1alpha' if is_thinking else 'v1beta'
+    response_type = 'streamGenerateContent' if stream else 'generateContent'
+    is_SSE = '&alt=sse' if stream else ''
+    gemini_history, system_instruction, error_response = func.process_messages_for_gemini(messages)
     if error_response:
         logger.error(f"处理输入消息时出错↙\n {error_response}")
         return jsonify(error_response), 400
     def do_request(current_api_key, attempt):
+        isok, time_remaining = is_within_rate_limit(current_api_key)
         if not isok:
+            logger.warning(f"暂时超过限额，该API key将在 {time_remaining} 秒后启用...")
             switch_api_key()
             return 0, None
         increment_request_count(current_api_key)
+        url = f"https://generativelanguage.googleapis.com/{api_version}/models/{model}:{response_type}?key={current_api_key}{is_SSE}"
+        headers = {
+            "Content-Type": "application/json",
         }
+        data = {
+            "contents": gemini_history,
+            "generationConfig": {
+                "temperature": temperature,
+                "maxOutputTokens": max_tokens,
+            },
+            "safetySettings": safety_settings_g2 if model == 'gemini-2.0-flash-exp' else safety_settings,
+        }
+        if system_instruction:
+            data["system_instruction"] = system_instruction
         try:
+            response = requests.post(url, headers=headers, json=data, stream=True)
+            response.raise_for_status()
+            if stream:
+                return 1, response
             else:
+                return 1, ResponseWrapper(response.json())
+        except requests.exceptions.RequestException as e:
             return handle_api_error(e, attempt)
+    def generate_stream(response):
+        buffer = b""
         try:
+            for line in response.iter_lines():
+                if not line:
+                    continue
+                try:
+                    if line.startswith(b'data: '):
+                        line = line[6:]
+                    buffer += line
+                    try:
+                        data = json.loads(buffer.decode('utf-8'))
+                        buffer = b""
+                        if 'candidates' in data and data['candidates']:
+                            candidate = data['candidates'][0]
+                            if 'content' in candidate:
+                                content = candidate['content']
+                                if 'parts' in content and content['parts']:
+                                    text = content['parts'][0].get('text', '')
+                                    finish_reason = candidate.get('finishReason')
+                                    if text:
+                                        data = {
+                                            'choices': [{
+                                                'delta': {
+                                                    'content': text
+                                                },
+                                                'finish_reason': finish_reason,
+                                                'index': 0
+                                            }],
+                                            'object': 'chat.completion.chunk'
+                                        }
+                                        yield f"data: {json.dumps(data)}\n\n"
+                    except json.JSONDecodeError:
+                        logger.debug(f"JSONDecodeError, buffer now: {buffer}")
+                        continue
+                except Exception as e:
+                    logger.error(f"Stream error during processing: {e}, Raw data line: {line}")
+                    yield f"data: {json.dumps({'error': str(e)})}\n\n"
+            yield f"data: {json.dumps({'choices': [{'delta': {}, 'finish_reason': 'stop', 'index': 0}]})}\n\n"
+        except Exception as e:
+            logger.error(f"Stream error: {e}")
+            yield f"data: {json.dumps({'error': str(e)})}\n\n"
     attempt = 0
     success = 0
     response = None
     for attempt in range(1, MAX_RETRIES + 1):
         logger.info(f"第 {attempt}/{MAX_RETRIES} 次尝试 ...")
         success, response = do_request(current_api_key, attempt)
         return jsonify(response), 500 if response is not None else 503
     if stream:
+        return Response(
+                stream_with_context(generate_stream(response)),
+                mimetype='text/event-stream'
+            )
     else:
         try:
             text_content = response.text
     scheduler.add_job(keep_alive, 'interval', hours=12)
     scheduler.start()
+    logger.info(f"Reminiproxy v2.3.0 启动")
     logger.info(f"最大尝试次数/MaxRetries: {MAX_RETRIES}")
     logger.info(f"最大请求次数/MaxRequests: {MAX_REQUESTS}")
     logger.info(f"请求限额窗口/LimitWindow: {LIMIT_WINDOW} 秒")

func.py CHANGED Viewed

@@ -106,13 +106,7 @@ def process_messages_for_gemini(messages):
                 else:
                     errors.append(f"Invalid role: {role}")
-    if gemini_history:
-        user_message = gemini_history[-1]
-        gemini_history = gemini_history[:-1]
-    else:
-        user_message = {"role": "user", "parts": [""]}
     if errors:
-        return gemini_history, user_message, {"parts": [{"text": system_instruction_text}]}, (jsonify({'error': errors}), 400)
     else:
-        return gemini_history, user_message, {"parts": [{"text": system_instruction_text}]}, None

                 else:
                     errors.append(f"Invalid role: {role}")
     if errors:
+        return gemini_history, {"parts": [{"text": system_instruction_text}]}, (jsonify({'error': errors}), 400)
     else:
+        return gemini_history, {"parts": [{"text": system_instruction_text}]}, None