Commit
·
40acccd
1
Parent(s):
f16ae1a
added project files
Browse files- .DS_Store +0 -0
- Dockerfile +20 -0
- app/config.py +24 -0
- app/main.py +780 -0
- app/requirements.txt +6 -0
- credentials/Placeholder Place credential json files here +0 -0
- docker-compose.yml +20 -0
.DS_Store
ADDED
|
Binary file (6.15 kB). View file
|
|
|
Dockerfile
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM python:3.11-slim
|
| 2 |
+
|
| 3 |
+
WORKDIR /app
|
| 4 |
+
|
| 5 |
+
# Install dependencies
|
| 6 |
+
COPY app/requirements.txt .
|
| 7 |
+
RUN pip install --no-cache-dir -r requirements.txt
|
| 8 |
+
|
| 9 |
+
# Copy application code
|
| 10 |
+
COPY app/ .
|
| 11 |
+
|
| 12 |
+
# Create a directory for the credentials
|
| 13 |
+
RUN mkdir -p /app/credentials
|
| 14 |
+
|
| 15 |
+
# Expose the port
|
| 16 |
+
EXPOSE 8050
|
| 17 |
+
|
| 18 |
+
# Command to run the application
|
| 19 |
+
# Use the default Hugging Face port 7860
|
| 20 |
+
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]
|
app/config.py
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
|
| 3 |
+
# Default password if not set in environment
|
| 4 |
+
DEFAULT_PASSWORD = "123456"
|
| 5 |
+
|
| 6 |
+
# Get password from environment variable or use default
|
| 7 |
+
API_KEY = os.environ.get("API_KEY", DEFAULT_PASSWORD)
|
| 8 |
+
|
| 9 |
+
# Function to validate API key
|
| 10 |
+
def validate_api_key(api_key: str) -> bool:
|
| 11 |
+
"""
|
| 12 |
+
Validate the provided API key against the configured key
|
| 13 |
+
|
| 14 |
+
Args:
|
| 15 |
+
api_key: The API key to validate
|
| 16 |
+
|
| 17 |
+
Returns:
|
| 18 |
+
bool: True if the key is valid, False otherwise
|
| 19 |
+
"""
|
| 20 |
+
if not API_KEY:
|
| 21 |
+
# If no API key is configured, authentication is disabled
|
| 22 |
+
return True
|
| 23 |
+
|
| 24 |
+
return api_key == API_KEY
|
app/main.py
ADDED
|
@@ -0,0 +1,780 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from fastapi import FastAPI, HTTPException, Depends, Header, Request
|
| 2 |
+
from fastapi.responses import JSONResponse, StreamingResponse
|
| 3 |
+
from fastapi.security import APIKeyHeader
|
| 4 |
+
from pydantic import BaseModel, ConfigDict, Field
|
| 5 |
+
from typing import List, Dict, Any, Optional, Union, Literal
|
| 6 |
+
import base64
|
| 7 |
+
import re
|
| 8 |
+
import json
|
| 9 |
+
import time
|
| 10 |
+
import os
|
| 11 |
+
import glob
|
| 12 |
+
import random
|
| 13 |
+
from google.oauth2 import service_account
|
| 14 |
+
import config
|
| 15 |
+
|
| 16 |
+
from google.genai import types
|
| 17 |
+
|
| 18 |
+
from google import genai
|
| 19 |
+
|
| 20 |
+
client = None
|
| 21 |
+
|
| 22 |
+
app = FastAPI(title="OpenAI to Gemini Adapter")
|
| 23 |
+
|
| 24 |
+
# API Key security scheme
|
| 25 |
+
api_key_header = APIKeyHeader(name="Authorization", auto_error=False)
|
| 26 |
+
|
| 27 |
+
# Dependency for API key validation
|
| 28 |
+
async def get_api_key(authorization: Optional[str] = Header(None)):
|
| 29 |
+
if authorization is None:
|
| 30 |
+
raise HTTPException(
|
| 31 |
+
status_code=401,
|
| 32 |
+
detail="Missing API key. Please include 'Authorization: Bearer YOUR_API_KEY' header."
|
| 33 |
+
)
|
| 34 |
+
|
| 35 |
+
# Check if the header starts with "Bearer "
|
| 36 |
+
if not authorization.startswith("Bearer "):
|
| 37 |
+
raise HTTPException(
|
| 38 |
+
status_code=401,
|
| 39 |
+
detail="Invalid API key format. Use 'Authorization: Bearer YOUR_API_KEY'"
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
+
# Extract the API key
|
| 43 |
+
api_key = authorization.replace("Bearer ", "")
|
| 44 |
+
|
| 45 |
+
# Validate the API key
|
| 46 |
+
if not config.validate_api_key(api_key):
|
| 47 |
+
raise HTTPException(
|
| 48 |
+
status_code=401,
|
| 49 |
+
detail="Invalid API key"
|
| 50 |
+
)
|
| 51 |
+
|
| 52 |
+
return api_key
|
| 53 |
+
|
| 54 |
+
# Credential Manager for handling multiple service accounts
|
| 55 |
+
class CredentialManager:
|
| 56 |
+
def __init__(self, default_credentials_dir="/app/credentials"):
|
| 57 |
+
# Use environment variable if set, otherwise use default
|
| 58 |
+
self.credentials_dir = os.environ.get("CREDENTIALS_DIR", default_credentials_dir)
|
| 59 |
+
self.credentials_files = []
|
| 60 |
+
self.current_index = 0
|
| 61 |
+
self.credentials = None
|
| 62 |
+
self.project_id = None
|
| 63 |
+
self.load_credentials_list()
|
| 64 |
+
|
| 65 |
+
def load_credentials_list(self):
|
| 66 |
+
"""Load the list of available credential files"""
|
| 67 |
+
# Look for all .json files in the credentials directory
|
| 68 |
+
pattern = os.path.join(self.credentials_dir, "*.json")
|
| 69 |
+
self.credentials_files = glob.glob(pattern)
|
| 70 |
+
|
| 71 |
+
if not self.credentials_files:
|
| 72 |
+
print(f"No credential files found in {self.credentials_dir}")
|
| 73 |
+
return False
|
| 74 |
+
|
| 75 |
+
print(f"Found {len(self.credentials_files)} credential files: {[os.path.basename(f) for f in self.credentials_files]}")
|
| 76 |
+
return True
|
| 77 |
+
|
| 78 |
+
def refresh_credentials_list(self):
|
| 79 |
+
"""Refresh the list of credential files (useful if files are added/removed)"""
|
| 80 |
+
old_count = len(self.credentials_files)
|
| 81 |
+
self.load_credentials_list()
|
| 82 |
+
new_count = len(self.credentials_files)
|
| 83 |
+
|
| 84 |
+
if old_count != new_count:
|
| 85 |
+
print(f"Credential files updated: {old_count} -> {new_count}")
|
| 86 |
+
|
| 87 |
+
return len(self.credentials_files) > 0
|
| 88 |
+
|
| 89 |
+
def get_next_credentials(self):
|
| 90 |
+
"""Rotate to the next credential file and load it"""
|
| 91 |
+
if not self.credentials_files:
|
| 92 |
+
return None, None
|
| 93 |
+
|
| 94 |
+
# Get the next credential file in rotation
|
| 95 |
+
file_path = self.credentials_files[self.current_index]
|
| 96 |
+
self.current_index = (self.current_index + 1) % len(self.credentials_files)
|
| 97 |
+
|
| 98 |
+
try:
|
| 99 |
+
credentials = service_account.Credentials.from_service_account_file(file_path,scopes=['https://www.googleapis.com/auth/cloud-platform'])
|
| 100 |
+
project_id = credentials.project_id
|
| 101 |
+
print(f"Loaded credentials from {file_path} for project: {project_id}")
|
| 102 |
+
self.credentials = credentials
|
| 103 |
+
self.project_id = project_id
|
| 104 |
+
return credentials, project_id
|
| 105 |
+
except Exception as e:
|
| 106 |
+
print(f"Error loading credentials from {file_path}: {e}")
|
| 107 |
+
# Try the next file if this one fails
|
| 108 |
+
if len(self.credentials_files) > 1:
|
| 109 |
+
print("Trying next credential file...")
|
| 110 |
+
return self.get_next_credentials()
|
| 111 |
+
return None, None
|
| 112 |
+
|
| 113 |
+
def get_random_credentials(self):
|
| 114 |
+
"""Get a random credential file and load it"""
|
| 115 |
+
if not self.credentials_files:
|
| 116 |
+
return None, None
|
| 117 |
+
|
| 118 |
+
# Choose a random credential file
|
| 119 |
+
file_path = random.choice(self.credentials_files)
|
| 120 |
+
|
| 121 |
+
try:
|
| 122 |
+
credentials = service_account.Credentials.from_service_account_file(file_path,scopes=['https://www.googleapis.com/auth/cloud-platform'])
|
| 123 |
+
project_id = credentials.project_id
|
| 124 |
+
print(f"Loaded credentials from {file_path} for project: {project_id}")
|
| 125 |
+
self.credentials = credentials
|
| 126 |
+
self.project_id = project_id
|
| 127 |
+
return credentials, project_id
|
| 128 |
+
except Exception as e:
|
| 129 |
+
print(f"Error loading credentials from {file_path}: {e}")
|
| 130 |
+
# Try another random file if this one fails
|
| 131 |
+
if len(self.credentials_files) > 1:
|
| 132 |
+
print("Trying another credential file...")
|
| 133 |
+
return self.get_random_credentials()
|
| 134 |
+
return None, None
|
| 135 |
+
|
| 136 |
+
# Initialize the credential manager
|
| 137 |
+
credential_manager = CredentialManager()
|
| 138 |
+
|
| 139 |
+
# Define data models
|
| 140 |
+
class ImageUrl(BaseModel):
|
| 141 |
+
url: str
|
| 142 |
+
|
| 143 |
+
class ContentPartImage(BaseModel):
|
| 144 |
+
type: Literal["image_url"]
|
| 145 |
+
image_url: ImageUrl
|
| 146 |
+
|
| 147 |
+
class ContentPartText(BaseModel):
|
| 148 |
+
type: Literal["text"]
|
| 149 |
+
text: str
|
| 150 |
+
|
| 151 |
+
class OpenAIMessage(BaseModel):
|
| 152 |
+
role: str
|
| 153 |
+
content: Union[str, List[Union[ContentPartText, ContentPartImage, Dict[str, Any]]]]
|
| 154 |
+
|
| 155 |
+
class OpenAIRequest(BaseModel):
|
| 156 |
+
model: str
|
| 157 |
+
messages: List[OpenAIMessage]
|
| 158 |
+
temperature: Optional[float] = 1.0
|
| 159 |
+
max_tokens: Optional[int] = None
|
| 160 |
+
top_p: Optional[float] = 1.0
|
| 161 |
+
top_k: Optional[int] = None
|
| 162 |
+
stream: Optional[bool] = False
|
| 163 |
+
stop: Optional[List[str]] = None
|
| 164 |
+
presence_penalty: Optional[float] = None
|
| 165 |
+
frequency_penalty: Optional[float] = None
|
| 166 |
+
seed: Optional[int] = None
|
| 167 |
+
logprobs: Optional[int] = None
|
| 168 |
+
response_logprobs: Optional[bool] = None
|
| 169 |
+
n: Optional[int] = None # Maps to candidate_count in Vertex AI
|
| 170 |
+
|
| 171 |
+
# Allow extra fields to pass through without causing validation errors
|
| 172 |
+
model_config = ConfigDict(extra='allow')
|
| 173 |
+
|
| 174 |
+
# Configure authentication
|
| 175 |
+
def init_vertex_ai():
|
| 176 |
+
global client # Ensure we modify the global client variable
|
| 177 |
+
try:
|
| 178 |
+
# Priority 1: Check for credentials JSON content in environment variable (Hugging Face)
|
| 179 |
+
credentials_json_str = os.environ.get("GOOGLE_CREDENTIALS_JSON")
|
| 180 |
+
if credentials_json_str:
|
| 181 |
+
try:
|
| 182 |
+
credentials_info = json.loads(credentials_json_str)
|
| 183 |
+
credentials = service_account.Credentials.from_service_account_info(credentials_info, scopes=['https://www.googleapis.com/auth/cloud-platform'])
|
| 184 |
+
project_id = credentials.project_id
|
| 185 |
+
client = genai.Client(vertexai=True, credentials=credentials, project=project_id, location="us-central1")
|
| 186 |
+
print(f"Initialized Vertex AI using GOOGLE_CREDENTIALS_JSON env var for project: {project_id}")
|
| 187 |
+
return True
|
| 188 |
+
except Exception as e:
|
| 189 |
+
print(f"Error loading credentials from GOOGLE_CREDENTIALS_JSON: {e}")
|
| 190 |
+
# Fall through to other methods if this fails
|
| 191 |
+
|
| 192 |
+
# Priority 2: Try to use the credential manager to get credentials from files
|
| 193 |
+
credentials, project_id = credential_manager.get_next_credentials()
|
| 194 |
+
|
| 195 |
+
if credentials and project_id:
|
| 196 |
+
client = genai.Client(vertexai=True, credentials=credentials, project=project_id, location="us-central1")
|
| 197 |
+
print(f"Initialized Vertex AI using Credential Manager for project: {project_id}")
|
| 198 |
+
return True
|
| 199 |
+
|
| 200 |
+
# Priority 3: Fall back to GOOGLE_APPLICATION_CREDENTIALS environment variable (file path)
|
| 201 |
+
file_path = os.environ.get("GOOGLE_APPLICATION_CREDENTIALS")
|
| 202 |
+
if file_path and os.path.exists(file_path):
|
| 203 |
+
try:
|
| 204 |
+
credentials = service_account.Credentials.from_service_account_file(file_path, scopes=['https://www.googleapis.com/auth/cloud-platform'])
|
| 205 |
+
project_id = credentials.project_id
|
| 206 |
+
client = genai.Client(vertexai=True, credentials=credentials, project=project_id, location="us-central1")
|
| 207 |
+
print(f"Initialized Vertex AI using GOOGLE_APPLICATION_CREDENTIALS file path for project: {project_id}")
|
| 208 |
+
return True
|
| 209 |
+
except Exception as e:
|
| 210 |
+
print(f"Error loading credentials from GOOGLE_APPLICATION_CREDENTIALS path {file_path}: {e}")
|
| 211 |
+
|
| 212 |
+
# If none of the methods worked
|
| 213 |
+
print(f"Error: No valid credentials found. Tried GOOGLE_CREDENTIALS_JSON, Credential Manager ({credential_manager.credentials_dir}), and GOOGLE_APPLICATION_CREDENTIALS.")
|
| 214 |
+
return False
|
| 215 |
+
except Exception as e:
|
| 216 |
+
print(f"Error initializing authentication: {e}")
|
| 217 |
+
return False
|
| 218 |
+
|
| 219 |
+
# Initialize Vertex AI at startup
|
| 220 |
+
@app.on_event("startup")
|
| 221 |
+
async def startup_event():
|
| 222 |
+
if not init_vertex_ai():
|
| 223 |
+
print("WARNING: Failed to initialize Vertex AI authentication")
|
| 224 |
+
|
| 225 |
+
# Conversion functions
|
| 226 |
+
def create_gemini_prompt(messages: List[OpenAIMessage]) -> Union[str, List[Any]]:
|
| 227 |
+
"""
|
| 228 |
+
Convert OpenAI messages to Gemini format.
|
| 229 |
+
Returns either a string prompt or a list of content parts if images are present.
|
| 230 |
+
"""
|
| 231 |
+
# Check if any message contains image content
|
| 232 |
+
has_images = False
|
| 233 |
+
for message in messages:
|
| 234 |
+
if isinstance(message.content, list):
|
| 235 |
+
for part in message.content:
|
| 236 |
+
if isinstance(part, dict) and part.get('type') == 'image_url':
|
| 237 |
+
has_images = True
|
| 238 |
+
break
|
| 239 |
+
elif isinstance(part, ContentPartImage):
|
| 240 |
+
has_images = True
|
| 241 |
+
break
|
| 242 |
+
if has_images:
|
| 243 |
+
break
|
| 244 |
+
|
| 245 |
+
# If no images, use the text-only format
|
| 246 |
+
if not has_images:
|
| 247 |
+
prompt = ""
|
| 248 |
+
|
| 249 |
+
# Extract system message if present
|
| 250 |
+
system_message = None
|
| 251 |
+
for message in messages:
|
| 252 |
+
if message.role == "system":
|
| 253 |
+
# Handle both string and list[dict] content types
|
| 254 |
+
if isinstance(message.content, str):
|
| 255 |
+
system_message = message.content
|
| 256 |
+
elif isinstance(message.content, list) and message.content and isinstance(message.content[0], dict) and 'text' in message.content[0]:
|
| 257 |
+
system_message = message.content[0]['text']
|
| 258 |
+
else:
|
| 259 |
+
# Handle unexpected format or raise error? For now, assume it's usable or skip.
|
| 260 |
+
system_message = str(message.content) # Fallback, might need refinement
|
| 261 |
+
break
|
| 262 |
+
|
| 263 |
+
# If system message exists, prepend it
|
| 264 |
+
if system_message:
|
| 265 |
+
prompt += f"System: {system_message}\n\n"
|
| 266 |
+
|
| 267 |
+
# Add other messages
|
| 268 |
+
for message in messages:
|
| 269 |
+
if message.role == "system":
|
| 270 |
+
continue # Already handled
|
| 271 |
+
|
| 272 |
+
# Handle both string and list[dict] content types
|
| 273 |
+
content_text = ""
|
| 274 |
+
if isinstance(message.content, str):
|
| 275 |
+
content_text = message.content
|
| 276 |
+
elif isinstance(message.content, list) and message.content and isinstance(message.content[0], dict) and 'text' in message.content[0]:
|
| 277 |
+
content_text = message.content[0]['text']
|
| 278 |
+
else:
|
| 279 |
+
# Fallback for unexpected format
|
| 280 |
+
content_text = str(message.content)
|
| 281 |
+
|
| 282 |
+
if message.role == "user":
|
| 283 |
+
prompt += f"Human: {content_text}\n"
|
| 284 |
+
elif message.role == "assistant":
|
| 285 |
+
prompt += f"AI: {content_text}\n"
|
| 286 |
+
|
| 287 |
+
# Add final AI prompt if last message was from user
|
| 288 |
+
if messages[-1].role == "user":
|
| 289 |
+
prompt += "AI: "
|
| 290 |
+
|
| 291 |
+
return prompt
|
| 292 |
+
|
| 293 |
+
# If images are present, create a list of content parts
|
| 294 |
+
gemini_contents = []
|
| 295 |
+
|
| 296 |
+
# Extract system message if present and add it first
|
| 297 |
+
for message in messages:
|
| 298 |
+
if message.role == "system":
|
| 299 |
+
if isinstance(message.content, str):
|
| 300 |
+
gemini_contents.append(f"System: {message.content}")
|
| 301 |
+
elif isinstance(message.content, list):
|
| 302 |
+
# Extract text from system message
|
| 303 |
+
system_text = ""
|
| 304 |
+
for part in message.content:
|
| 305 |
+
if isinstance(part, dict) and part.get('type') == 'text':
|
| 306 |
+
system_text += part.get('text', '')
|
| 307 |
+
elif isinstance(part, ContentPartText):
|
| 308 |
+
system_text += part.text
|
| 309 |
+
if system_text:
|
| 310 |
+
gemini_contents.append(f"System: {system_text}")
|
| 311 |
+
break
|
| 312 |
+
|
| 313 |
+
# Process user and assistant messages
|
| 314 |
+
for message in messages:
|
| 315 |
+
if message.role == "system":
|
| 316 |
+
continue # Already handled
|
| 317 |
+
|
| 318 |
+
# For string content, add as text
|
| 319 |
+
if isinstance(message.content, str):
|
| 320 |
+
prefix = "Human: " if message.role == "user" else "AI: "
|
| 321 |
+
gemini_contents.append(f"{prefix}{message.content}")
|
| 322 |
+
|
| 323 |
+
# For list content, process each part
|
| 324 |
+
elif isinstance(message.content, list):
|
| 325 |
+
# First collect all text parts
|
| 326 |
+
text_content = ""
|
| 327 |
+
|
| 328 |
+
for part in message.content:
|
| 329 |
+
# Handle text parts
|
| 330 |
+
if isinstance(part, dict) and part.get('type') == 'text':
|
| 331 |
+
text_content += part.get('text', '')
|
| 332 |
+
elif isinstance(part, ContentPartText):
|
| 333 |
+
text_content += part.text
|
| 334 |
+
|
| 335 |
+
# Add the combined text content if any
|
| 336 |
+
if text_content:
|
| 337 |
+
prefix = "Human: " if message.role == "user" else "AI: "
|
| 338 |
+
gemini_contents.append(f"{prefix}{text_content}")
|
| 339 |
+
|
| 340 |
+
# Then process image parts
|
| 341 |
+
for part in message.content:
|
| 342 |
+
# Handle image parts
|
| 343 |
+
if isinstance(part, dict) and part.get('type') == 'image_url':
|
| 344 |
+
image_url = part.get('image_url', {}).get('url', '')
|
| 345 |
+
if image_url.startswith('data:'):
|
| 346 |
+
# Extract mime type and base64 data
|
| 347 |
+
mime_match = re.match(r'data:([^;]+);base64,(.+)', image_url)
|
| 348 |
+
if mime_match:
|
| 349 |
+
mime_type, b64_data = mime_match.groups()
|
| 350 |
+
image_bytes = base64.b64decode(b64_data)
|
| 351 |
+
gemini_contents.append(types.Part.from_bytes(data=image_bytes, mime_type=mime_type))
|
| 352 |
+
elif isinstance(part, ContentPartImage):
|
| 353 |
+
image_url = part.image_url.url
|
| 354 |
+
if image_url.startswith('data:'):
|
| 355 |
+
# Extract mime type and base64 data
|
| 356 |
+
mime_match = re.match(r'data:([^;]+);base64,(.+)', image_url)
|
| 357 |
+
if mime_match:
|
| 358 |
+
mime_type, b64_data = mime_match.groups()
|
| 359 |
+
image_bytes = base64.b64decode(b64_data)
|
| 360 |
+
gemini_contents.append(types.Part.from_bytes(data=image_bytes, mime_type=mime_type))
|
| 361 |
+
|
| 362 |
+
return gemini_contents
|
| 363 |
+
|
| 364 |
+
def create_generation_config(request: OpenAIRequest) -> Dict[str, Any]:
|
| 365 |
+
config = {}
|
| 366 |
+
|
| 367 |
+
# Basic parameters that were already supported
|
| 368 |
+
if request.temperature is not None:
|
| 369 |
+
config["temperature"] = request.temperature
|
| 370 |
+
|
| 371 |
+
if request.max_tokens is not None:
|
| 372 |
+
config["max_output_tokens"] = request.max_tokens
|
| 373 |
+
|
| 374 |
+
if request.top_p is not None:
|
| 375 |
+
config["top_p"] = request.top_p
|
| 376 |
+
|
| 377 |
+
if request.top_k is not None:
|
| 378 |
+
config["top_k"] = request.top_k
|
| 379 |
+
|
| 380 |
+
if request.stop is not None:
|
| 381 |
+
config["stop_sequences"] = request.stop
|
| 382 |
+
|
| 383 |
+
# Additional parameters with direct mappings
|
| 384 |
+
if request.presence_penalty is not None:
|
| 385 |
+
config["presence_penalty"] = request.presence_penalty
|
| 386 |
+
|
| 387 |
+
if request.frequency_penalty is not None:
|
| 388 |
+
config["frequency_penalty"] = request.frequency_penalty
|
| 389 |
+
|
| 390 |
+
if request.seed is not None:
|
| 391 |
+
config["seed"] = request.seed
|
| 392 |
+
|
| 393 |
+
if request.logprobs is not None:
|
| 394 |
+
config["logprobs"] = request.logprobs
|
| 395 |
+
|
| 396 |
+
if request.response_logprobs is not None:
|
| 397 |
+
config["response_logprobs"] = request.response_logprobs
|
| 398 |
+
|
| 399 |
+
# Map OpenAI's 'n' parameter to Vertex AI's 'candidate_count'
|
| 400 |
+
if request.n is not None:
|
| 401 |
+
config["candidate_count"] = request.n
|
| 402 |
+
|
| 403 |
+
return config
|
| 404 |
+
|
| 405 |
+
# Response format conversion
|
| 406 |
+
def convert_to_openai_format(gemini_response, model: str) -> Dict[str, Any]:
|
| 407 |
+
# Handle multiple candidates if present
|
| 408 |
+
if hasattr(gemini_response, 'candidates') and len(gemini_response.candidates) > 1:
|
| 409 |
+
choices = []
|
| 410 |
+
for i, candidate in enumerate(gemini_response.candidates):
|
| 411 |
+
choices.append({
|
| 412 |
+
"index": i,
|
| 413 |
+
"message": {
|
| 414 |
+
"role": "assistant",
|
| 415 |
+
"content": candidate.text
|
| 416 |
+
},
|
| 417 |
+
"finish_reason": "stop"
|
| 418 |
+
})
|
| 419 |
+
else:
|
| 420 |
+
# Handle single response (backward compatibility)
|
| 421 |
+
choices = [
|
| 422 |
+
{
|
| 423 |
+
"index": 0,
|
| 424 |
+
"message": {
|
| 425 |
+
"role": "assistant",
|
| 426 |
+
"content": gemini_response.text
|
| 427 |
+
},
|
| 428 |
+
"finish_reason": "stop"
|
| 429 |
+
}
|
| 430 |
+
]
|
| 431 |
+
|
| 432 |
+
# Include logprobs if available
|
| 433 |
+
for i, choice in enumerate(choices):
|
| 434 |
+
if hasattr(gemini_response, 'candidates') and i < len(gemini_response.candidates):
|
| 435 |
+
candidate = gemini_response.candidates[i]
|
| 436 |
+
if hasattr(candidate, 'logprobs'):
|
| 437 |
+
choice["logprobs"] = candidate.logprobs
|
| 438 |
+
|
| 439 |
+
return {
|
| 440 |
+
"id": f"chatcmpl-{int(time.time())}",
|
| 441 |
+
"object": "chat.completion",
|
| 442 |
+
"created": int(time.time()),
|
| 443 |
+
"model": model,
|
| 444 |
+
"choices": choices,
|
| 445 |
+
"usage": {
|
| 446 |
+
"prompt_tokens": 0, # Would need token counting logic
|
| 447 |
+
"completion_tokens": 0,
|
| 448 |
+
"total_tokens": 0
|
| 449 |
+
}
|
| 450 |
+
}
|
| 451 |
+
|
| 452 |
+
def convert_chunk_to_openai(chunk, model: str, response_id: str, candidate_index: int = 0) -> str:
|
| 453 |
+
chunk_content = chunk.text if hasattr(chunk, 'text') else ""
|
| 454 |
+
|
| 455 |
+
chunk_data = {
|
| 456 |
+
"id": response_id,
|
| 457 |
+
"object": "chat.completion.chunk",
|
| 458 |
+
"created": int(time.time()),
|
| 459 |
+
"model": model,
|
| 460 |
+
"choices": [
|
| 461 |
+
{
|
| 462 |
+
"index": candidate_index,
|
| 463 |
+
"delta": {
|
| 464 |
+
"content": chunk_content
|
| 465 |
+
},
|
| 466 |
+
"finish_reason": None
|
| 467 |
+
}
|
| 468 |
+
]
|
| 469 |
+
}
|
| 470 |
+
|
| 471 |
+
# Add logprobs if available
|
| 472 |
+
if hasattr(chunk, 'logprobs'):
|
| 473 |
+
chunk_data["choices"][0]["logprobs"] = chunk.logprobs
|
| 474 |
+
|
| 475 |
+
return f"data: {json.dumps(chunk_data)}\n\n"
|
| 476 |
+
|
| 477 |
+
def create_final_chunk(model: str, response_id: str, candidate_count: int = 1) -> str:
|
| 478 |
+
choices = []
|
| 479 |
+
for i in range(candidate_count):
|
| 480 |
+
choices.append({
|
| 481 |
+
"index": i,
|
| 482 |
+
"delta": {},
|
| 483 |
+
"finish_reason": "stop"
|
| 484 |
+
})
|
| 485 |
+
|
| 486 |
+
final_chunk = {
|
| 487 |
+
"id": response_id,
|
| 488 |
+
"object": "chat.completion.chunk",
|
| 489 |
+
"created": int(time.time()),
|
| 490 |
+
"model": model,
|
| 491 |
+
"choices": choices
|
| 492 |
+
}
|
| 493 |
+
|
| 494 |
+
return f"data: {json.dumps(final_chunk)}\n\n"
|
| 495 |
+
|
| 496 |
+
# /v1/models endpoint
|
| 497 |
+
@app.get("/v1/models")
|
| 498 |
+
async def list_models(api_key: str = Depends(get_api_key)):
|
| 499 |
+
# Based on current information for Vertex AI models
|
| 500 |
+
models = [
|
| 501 |
+
{
|
| 502 |
+
"id": "gemini-2.5-pro-exp-03-25",
|
| 503 |
+
"object": "model",
|
| 504 |
+
"created": int(time.time()),
|
| 505 |
+
"owned_by": "google",
|
| 506 |
+
"permission": [],
|
| 507 |
+
"root": "gemini-2.5-pro-exp-03-25",
|
| 508 |
+
"parent": None,
|
| 509 |
+
},
|
| 510 |
+
{
|
| 511 |
+
"id": "gemini-2.5-pro-exp-03-25-search",
|
| 512 |
+
"object": "model",
|
| 513 |
+
"created": int(time.time()),
|
| 514 |
+
"owned_by": "google",
|
| 515 |
+
"permission": [],
|
| 516 |
+
"root": "gemini-2.5-pro-exp-03-25",
|
| 517 |
+
"parent": None,
|
| 518 |
+
},
|
| 519 |
+
{
|
| 520 |
+
"id": "gemini-2.0-flash",
|
| 521 |
+
"object": "model",
|
| 522 |
+
"created": int(time.time()),
|
| 523 |
+
"owned_by": "google",
|
| 524 |
+
"permission": [],
|
| 525 |
+
"root": "gemini-2.0-flash",
|
| 526 |
+
"parent": None,
|
| 527 |
+
},
|
| 528 |
+
{
|
| 529 |
+
"id": "gemini-2.0-flash-search",
|
| 530 |
+
"object": "model",
|
| 531 |
+
"created": int(time.time()),
|
| 532 |
+
"owned_by": "google",
|
| 533 |
+
"permission": [],
|
| 534 |
+
"root": "gemini-2.0-flash",
|
| 535 |
+
"parent": None,
|
| 536 |
+
},
|
| 537 |
+
{
|
| 538 |
+
"id": "gemini-2.0-flash-lite",
|
| 539 |
+
"object": "model",
|
| 540 |
+
"created": int(time.time()),
|
| 541 |
+
"owned_by": "google",
|
| 542 |
+
"permission": [],
|
| 543 |
+
"root": "gemini-2.0-flash-lite",
|
| 544 |
+
"parent": None,
|
| 545 |
+
},
|
| 546 |
+
{
|
| 547 |
+
"id": "gemini-2.0-flash-lite-search",
|
| 548 |
+
"object": "model",
|
| 549 |
+
"created": int(time.time()),
|
| 550 |
+
"owned_by": "google",
|
| 551 |
+
"permission": [],
|
| 552 |
+
"root": "gemini-2.0-flash-lite",
|
| 553 |
+
"parent": None,
|
| 554 |
+
},
|
| 555 |
+
{
|
| 556 |
+
"id": "gemini-2.0-pro-exp-02-05",
|
| 557 |
+
"object": "model",
|
| 558 |
+
"created": int(time.time()),
|
| 559 |
+
"owned_by": "google",
|
| 560 |
+
"permission": [],
|
| 561 |
+
"root": "gemini-2.0-pro-exp-02-05",
|
| 562 |
+
"parent": None,
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"id": "gemini-1.5-flash",
|
| 566 |
+
"object": "model",
|
| 567 |
+
"created": int(time.time()),
|
| 568 |
+
"owned_by": "google",
|
| 569 |
+
"permission": [],
|
| 570 |
+
"root": "gemini-1.5-flash",
|
| 571 |
+
"parent": None,
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"id": "gemini-1.5-flash-8b",
|
| 575 |
+
"object": "model",
|
| 576 |
+
"created": int(time.time()),
|
| 577 |
+
"owned_by": "google",
|
| 578 |
+
"permission": [],
|
| 579 |
+
"root": "gemini-1.5-flash-8b",
|
| 580 |
+
"parent": None,
|
| 581 |
+
},
|
| 582 |
+
{
|
| 583 |
+
"id": "gemini-1.5-pro",
|
| 584 |
+
"object": "model",
|
| 585 |
+
"created": int(time.time()),
|
| 586 |
+
"owned_by": "google",
|
| 587 |
+
"permission": [],
|
| 588 |
+
"root": "gemini-1.5-pro",
|
| 589 |
+
"parent": None,
|
| 590 |
+
},
|
| 591 |
+
{
|
| 592 |
+
"id": "gemini-1.0-pro-002",
|
| 593 |
+
"object": "model",
|
| 594 |
+
"created": int(time.time()),
|
| 595 |
+
"owned_by": "google",
|
| 596 |
+
"permission": [],
|
| 597 |
+
"root": "gemini-1.0-pro-002",
|
| 598 |
+
"parent": None,
|
| 599 |
+
},
|
| 600 |
+
{
|
| 601 |
+
"id": "gemini-1.0-pro-vision-001",
|
| 602 |
+
"object": "model",
|
| 603 |
+
"created": int(time.time()),
|
| 604 |
+
"owned_by": "google",
|
| 605 |
+
"permission": [],
|
| 606 |
+
"root": "gemini-1.0-pro-vision-001",
|
| 607 |
+
"parent": None,
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"id": "gemini-embedding-exp",
|
| 611 |
+
"object": "model",
|
| 612 |
+
"created": int(time.time()),
|
| 613 |
+
"owned_by": "google",
|
| 614 |
+
"permission": [],
|
| 615 |
+
"root": "gemini-embedding-exp",
|
| 616 |
+
"parent": None,
|
| 617 |
+
}
|
| 618 |
+
]
|
| 619 |
+
|
| 620 |
+
return {"object": "list", "data": models}
|
| 621 |
+
|
| 622 |
+
# Main chat completion endpoint
|
| 623 |
+
# OpenAI-compatible error response
|
| 624 |
+
def create_openai_error_response(status_code: int, message: str, error_type: str) -> Dict[str, Any]:
|
| 625 |
+
return {
|
| 626 |
+
"error": {
|
| 627 |
+
"message": message,
|
| 628 |
+
"type": error_type,
|
| 629 |
+
"code": status_code,
|
| 630 |
+
"param": None,
|
| 631 |
+
}
|
| 632 |
+
}
|
| 633 |
+
|
| 634 |
+
@app.post("/v1/chat/completions")
|
| 635 |
+
async def chat_completions(request: OpenAIRequest, api_key: str = Depends(get_api_key)):
|
| 636 |
+
try:
|
| 637 |
+
# Validate model availability
|
| 638 |
+
models_response = await list_models()
|
| 639 |
+
if not request.model or not any(model["id"] == request.model for model in models_response.get("data", [])):
|
| 640 |
+
error_response = create_openai_error_response(
|
| 641 |
+
400, f"Model '{request.model}' not found", "invalid_request_error"
|
| 642 |
+
)
|
| 643 |
+
return JSONResponse(status_code=400, content=error_response)
|
| 644 |
+
|
| 645 |
+
# Check if this is a grounded search model
|
| 646 |
+
is_grounded_search = request.model.endswith("-search")
|
| 647 |
+
|
| 648 |
+
# Extract the base model name (remove -search suffix if present)
|
| 649 |
+
gemini_model = request.model.replace("-search", "") if is_grounded_search else request.model
|
| 650 |
+
|
| 651 |
+
# Create generation config
|
| 652 |
+
generation_config = create_generation_config(request)
|
| 653 |
+
|
| 654 |
+
# Get fresh credentials for this request
|
| 655 |
+
credentials, project_id = credential_manager.get_next_credentials()
|
| 656 |
+
|
| 657 |
+
if not credentials or not project_id:
|
| 658 |
+
error_response = create_openai_error_response(
|
| 659 |
+
500, "Failed to obtain valid credentials", "server_error"
|
| 660 |
+
)
|
| 661 |
+
return JSONResponse(status_code=500, content=error_response)
|
| 662 |
+
|
| 663 |
+
# Initialize Vertex AI with the rotated credentials
|
| 664 |
+
try:
|
| 665 |
+
# Re-initialize client for this request - credentials might have rotated
|
| 666 |
+
client = genai.Client(vertexai=True, credentials=credentials, project=project_id, location="us-central1")
|
| 667 |
+
print(f"Using credentials for project: {project_id} for this request")
|
| 668 |
+
except Exception as auth_error:
|
| 669 |
+
error_response = create_openai_error_response(
|
| 670 |
+
500, f"Failed to initialize authentication: {str(auth_error)}", "server_error"
|
| 671 |
+
)
|
| 672 |
+
return JSONResponse(status_code=500, content=error_response)
|
| 673 |
+
|
| 674 |
+
# Initialize Gemini model
|
| 675 |
+
search_tool = types.Tool(google_search=types.GoogleSearch())
|
| 676 |
+
|
| 677 |
+
safety_settings = [
|
| 678 |
+
types.SafetySetting(
|
| 679 |
+
category="HARM_CATEGORY_HATE_SPEECH",
|
| 680 |
+
threshold="OFF"
|
| 681 |
+
),types.SafetySetting(
|
| 682 |
+
category="HARM_CATEGORY_DANGEROUS_CONTENT",
|
| 683 |
+
threshold="OFF"
|
| 684 |
+
),types.SafetySetting(
|
| 685 |
+
category="HARM_CATEGORY_SEXUALLY_EXPLICIT",
|
| 686 |
+
threshold="OFF"
|
| 687 |
+
),types.SafetySetting(
|
| 688 |
+
category="HARM_CATEGORY_HARASSMENT",
|
| 689 |
+
threshold="OFF"
|
| 690 |
+
)]
|
| 691 |
+
|
| 692 |
+
generation_config["safety_settings"] = safety_settings
|
| 693 |
+
if is_grounded_search:
|
| 694 |
+
generation_config["tools"] = [search_tool]
|
| 695 |
+
|
| 696 |
+
# Create prompt from messages
|
| 697 |
+
prompt = create_gemini_prompt(request.messages)
|
| 698 |
+
|
| 699 |
+
if request.stream:
|
| 700 |
+
# Handle streaming response
|
| 701 |
+
async def stream_generator():
|
| 702 |
+
response_id = f"chatcmpl-{int(time.time())}"
|
| 703 |
+
candidate_count = request.n or 1
|
| 704 |
+
|
| 705 |
+
try:
|
| 706 |
+
# For streaming, we can only handle one candidate at a time
|
| 707 |
+
# If multiple candidates are requested, we'll generate them sequentially
|
| 708 |
+
for candidate_index in range(candidate_count):
|
| 709 |
+
# Generate content with streaming
|
| 710 |
+
# Handle both string and list content formats (for images)
|
| 711 |
+
responses = client.models.generate_content_stream(
|
| 712 |
+
model=gemini_model,
|
| 713 |
+
contents=prompt, # This can be either a string or a list of content parts
|
| 714 |
+
config=generation_config,
|
| 715 |
+
)
|
| 716 |
+
|
| 717 |
+
# Convert and yield each chunk
|
| 718 |
+
for response in responses:
|
| 719 |
+
yield convert_chunk_to_openai(response, request.model, response_id, candidate_index)
|
| 720 |
+
|
| 721 |
+
# Send final chunk with all candidates
|
| 722 |
+
yield create_final_chunk(request.model, response_id, candidate_count)
|
| 723 |
+
yield "data: [DONE]\n\n"
|
| 724 |
+
|
| 725 |
+
except Exception as stream_error:
|
| 726 |
+
# Format streaming errors in SSE format
|
| 727 |
+
error_msg = f"Error during streaming: {str(stream_error)}"
|
| 728 |
+
print(error_msg)
|
| 729 |
+
error_response = create_openai_error_response(500, error_msg, "server_error")
|
| 730 |
+
yield f"data: {json.dumps(error_response)}\n\n"
|
| 731 |
+
yield "data: [DONE]\n\n"
|
| 732 |
+
|
| 733 |
+
return StreamingResponse(
|
| 734 |
+
stream_generator(),
|
| 735 |
+
media_type="text/event-stream"
|
| 736 |
+
)
|
| 737 |
+
else:
|
| 738 |
+
# Handle non-streaming response
|
| 739 |
+
try:
|
| 740 |
+
# If multiple candidates are requested, set candidate_count
|
| 741 |
+
if request.n and request.n > 1:
|
| 742 |
+
# Make sure generation_config has candidate_count set
|
| 743 |
+
if "candidate_count" not in generation_config:
|
| 744 |
+
generation_config["candidate_count"] = request.n
|
| 745 |
+
# Handle both string and list content formats (for images)
|
| 746 |
+
response = client.models.generate_content(
|
| 747 |
+
model=gemini_model,
|
| 748 |
+
contents=prompt, # This can be either a string or a list of content parts
|
| 749 |
+
config=generation_config,
|
| 750 |
+
)
|
| 751 |
+
|
| 752 |
+
|
| 753 |
+
openai_response = convert_to_openai_format(response, request.model)
|
| 754 |
+
return JSONResponse(content=openai_response)
|
| 755 |
+
except Exception as generate_error:
|
| 756 |
+
error_msg = f"Error generating content: {str(generate_error)}"
|
| 757 |
+
print(error_msg)
|
| 758 |
+
error_response = create_openai_error_response(500, error_msg, "server_error")
|
| 759 |
+
return JSONResponse(status_code=500, content=error_response)
|
| 760 |
+
|
| 761 |
+
except Exception as e:
|
| 762 |
+
error_msg = f"Error processing request: {str(e)}"
|
| 763 |
+
print(error_msg)
|
| 764 |
+
error_response = create_openai_error_response(500, error_msg, "server_error")
|
| 765 |
+
return JSONResponse(status_code=500, content=error_response)
|
| 766 |
+
|
| 767 |
+
# Health check endpoint
|
| 768 |
+
@app.get("/health")
|
| 769 |
+
def health_check(api_key: str = Depends(get_api_key)):
|
| 770 |
+
# Refresh the credentials list to get the latest status
|
| 771 |
+
credential_manager.refresh_credentials_list()
|
| 772 |
+
|
| 773 |
+
return {
|
| 774 |
+
"status": "ok",
|
| 775 |
+
"credentials": {
|
| 776 |
+
"available": len(credential_manager.credentials_files),
|
| 777 |
+
"files": [os.path.basename(f) for f in credential_manager.credentials_files],
|
| 778 |
+
"current_index": credential_manager.current_index
|
| 779 |
+
}
|
| 780 |
+
}
|
app/requirements.txt
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
fastapi==0.110.0
|
| 2 |
+
uvicorn==0.27.1
|
| 3 |
+
google-auth==2.38.0
|
| 4 |
+
google-cloud-aiplatform==1.86.0
|
| 5 |
+
pydantic==2.6.1
|
| 6 |
+
google-genai==1.8.0
|
credentials/Placeholder Place credential json files here
ADDED
|
File without changes
|
docker-compose.yml
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version: '3.8'
|
| 2 |
+
|
| 3 |
+
services:
|
| 4 |
+
openai-to-gemini:
|
| 5 |
+
build:
|
| 6 |
+
context: .
|
| 7 |
+
dockerfile: Dockerfile
|
| 8 |
+
ports:
|
| 9 |
+
# Map host port 8050 to container port 7860 (for Hugging Face compatibility)
|
| 10 |
+
- "8050:7860"
|
| 11 |
+
volumes:
|
| 12 |
+
- ./credentials:/app/credentials
|
| 13 |
+
environment:
|
| 14 |
+
# This is kept for backward compatibility but our app now primarily uses the credential manager
|
| 15 |
+
- GOOGLE_APPLICATION_CREDENTIALS=/app/credentials/service-account.json
|
| 16 |
+
# Directory where credential files are stored (used by credential manager)
|
| 17 |
+
- CREDENTIALS_DIR=/app/credentials
|
| 18 |
+
# API key for authentication (default: 123456)
|
| 19 |
+
- API_KEY=123456
|
| 20 |
+
restart: unless-stopped
|