Commit
·
0796d75
1
Parent(s):
b1f7ea4
use official sdk for express mode 0605
Browse files- app/api_helpers.py +1 -1
- app/direct_vertex_client.py +0 -423
- app/project_id_discovery.py +73 -0
- app/routes/chat_api.py +10 -19
app/api_helpers.py
CHANGED
@@ -254,7 +254,7 @@ def is_gemini_response_valid(response: Any) -> bool:
|
|
254 |
if hasattr(response, 'text') and isinstance(response.text, str) and response.text.strip():
|
255 |
return True
|
256 |
|
257 |
-
# Check for candidates
|
258 |
if hasattr(response, 'candidates') and response.candidates:
|
259 |
for candidate in response.candidates:
|
260 |
# Check for direct text on candidate
|
|
|
254 |
if hasattr(response, 'text') and isinstance(response.text, str) and response.text.strip():
|
255 |
return True
|
256 |
|
257 |
+
# Check for candidates in the response
|
258 |
if hasattr(response, 'candidates') and response.candidates:
|
259 |
for candidate in response.candidates:
|
260 |
# Check for direct text on candidate
|
app/direct_vertex_client.py
DELETED
@@ -1,423 +0,0 @@
|
|
1 |
-
import aiohttp
|
2 |
-
import asyncio
|
3 |
-
import json
|
4 |
-
import re
|
5 |
-
from typing import Dict, Any, List, Union, Optional, AsyncGenerator
|
6 |
-
import time
|
7 |
-
|
8 |
-
# Global cache for project IDs: {api_key: project_id}
|
9 |
-
PROJECT_ID_CACHE: Dict[str, str] = {}
|
10 |
-
|
11 |
-
|
12 |
-
class DirectVertexClient:
|
13 |
-
"""
|
14 |
-
A client that connects to Vertex AI using direct URLs instead of the SDK.
|
15 |
-
Mimics the interface of genai.Client for seamless integration.
|
16 |
-
"""
|
17 |
-
|
18 |
-
def __init__(self, api_key: str):
|
19 |
-
self.api_key = api_key
|
20 |
-
self.project_id: Optional[str] = None
|
21 |
-
self.base_url = "https://aiplatform.googleapis.com/v1"
|
22 |
-
self.session: Optional[aiohttp.ClientSession] = None
|
23 |
-
# Mimic the model_name attribute that might be accessed
|
24 |
-
self.model_name = "direct_vertex_client"
|
25 |
-
|
26 |
-
# Create nested structure to mimic genai.Client interface
|
27 |
-
self.aio = self._AioNamespace(self)
|
28 |
-
|
29 |
-
class _AioNamespace:
|
30 |
-
def __init__(self, parent):
|
31 |
-
self.parent = parent
|
32 |
-
self.models = self._ModelsNamespace(parent)
|
33 |
-
|
34 |
-
class _ModelsNamespace:
|
35 |
-
def __init__(self, parent):
|
36 |
-
self.parent = parent
|
37 |
-
|
38 |
-
async def generate_content(self, model: str, contents: Any, config: Dict[str, Any]) -> Any:
|
39 |
-
"""Non-streaming content generation"""
|
40 |
-
return await self.parent._generate_content(model, contents, config, stream=False)
|
41 |
-
|
42 |
-
async def generate_content_stream(self, model: str, contents: Any, config: Dict[str, Any]):
|
43 |
-
"""Streaming content generation - returns an async generator"""
|
44 |
-
# This needs to be an async method that returns the generator
|
45 |
-
# to match the SDK's interface where you await the method call
|
46 |
-
return self.parent._generate_content_stream(model, contents, config)
|
47 |
-
|
48 |
-
async def _ensure_session(self):
|
49 |
-
"""Ensure aiohttp session is created"""
|
50 |
-
if self.session is None:
|
51 |
-
self.session = aiohttp.ClientSession()
|
52 |
-
|
53 |
-
async def close(self):
|
54 |
-
"""Clean up resources"""
|
55 |
-
if self.session:
|
56 |
-
await self.session.close()
|
57 |
-
self.session = None
|
58 |
-
|
59 |
-
async def discover_project_id(self) -> None:
|
60 |
-
"""Discover project ID by triggering an intentional error"""
|
61 |
-
# Check cache first
|
62 |
-
if self.api_key in PROJECT_ID_CACHE:
|
63 |
-
self.project_id = PROJECT_ID_CACHE[self.api_key]
|
64 |
-
print(f"INFO: Using cached project ID: {self.project_id}")
|
65 |
-
return
|
66 |
-
|
67 |
-
await self._ensure_session()
|
68 |
-
|
69 |
-
# Use a non-existent model to trigger error
|
70 |
-
error_url = f"{self.base_url}/publishers/google/models/gemini-2.7-pro-preview-05-06:streamGenerateContent?key={self.api_key}"
|
71 |
-
|
72 |
-
try:
|
73 |
-
# Send minimal request to trigger error
|
74 |
-
payload = {
|
75 |
-
"contents": [{"role": "user", "parts": [{"text": "test"}]}]
|
76 |
-
}
|
77 |
-
|
78 |
-
async with self.session.post(error_url, json=payload) as response:
|
79 |
-
response_text = await response.text()
|
80 |
-
|
81 |
-
try:
|
82 |
-
# Try to parse as JSON first
|
83 |
-
error_data = json.loads(response_text)
|
84 |
-
|
85 |
-
# Handle array response format
|
86 |
-
if isinstance(error_data, list) and len(error_data) > 0:
|
87 |
-
error_data = error_data[0]
|
88 |
-
|
89 |
-
if "error" in error_data:
|
90 |
-
error_message = error_data["error"].get("message", "")
|
91 |
-
# Extract project ID from error message
|
92 |
-
# Pattern: "projects/39982734461/locations/..."
|
93 |
-
match = re.search(r'projects/(\d+)/locations/', error_message)
|
94 |
-
if match:
|
95 |
-
self.project_id = match.group(1)
|
96 |
-
PROJECT_ID_CACHE[self.api_key] = self.project_id
|
97 |
-
print(f"INFO: Discovered project ID: {self.project_id}")
|
98 |
-
return
|
99 |
-
except json.JSONDecodeError:
|
100 |
-
# If not JSON, try to find project ID in raw text
|
101 |
-
match = re.search(r'projects/(\d+)/locations/', response_text)
|
102 |
-
if match:
|
103 |
-
self.project_id = match.group(1)
|
104 |
-
PROJECT_ID_CACHE[self.api_key] = self.project_id
|
105 |
-
print(f"INFO: Discovered project ID from raw response: {self.project_id}")
|
106 |
-
return
|
107 |
-
|
108 |
-
raise Exception(f"Failed to discover project ID. Status: {response.status}, Response: {response_text[:500]}")
|
109 |
-
|
110 |
-
except Exception as e:
|
111 |
-
print(f"ERROR: Failed to discover project ID: {e}")
|
112 |
-
raise
|
113 |
-
|
114 |
-
def _convert_contents(self, contents: Any) -> List[Dict[str, Any]]:
|
115 |
-
"""Convert SDK Content objects to REST API format"""
|
116 |
-
if isinstance(contents, list):
|
117 |
-
return [self._convert_content_item(item) for item in contents]
|
118 |
-
else:
|
119 |
-
return [self._convert_content_item(contents)]
|
120 |
-
|
121 |
-
def _convert_content_item(self, content: Any) -> Dict[str, Any]:
|
122 |
-
"""Convert a single content item to REST API format"""
|
123 |
-
if isinstance(content, dict):
|
124 |
-
return content
|
125 |
-
|
126 |
-
# Handle SDK Content objects
|
127 |
-
result = {}
|
128 |
-
if hasattr(content, 'role'):
|
129 |
-
result['role'] = content.role
|
130 |
-
if hasattr(content, 'parts'):
|
131 |
-
result['parts'] = []
|
132 |
-
for part in content.parts:
|
133 |
-
if isinstance(part, dict):
|
134 |
-
result['parts'].append(part)
|
135 |
-
elif hasattr(part, 'text'):
|
136 |
-
result['parts'].append({'text': part.text})
|
137 |
-
elif hasattr(part, 'inline_data'):
|
138 |
-
result['parts'].append({
|
139 |
-
'inline_data': {
|
140 |
-
'mime_type': part.inline_data.mime_type,
|
141 |
-
'data': part.inline_data.data
|
142 |
-
}
|
143 |
-
})
|
144 |
-
return result
|
145 |
-
|
146 |
-
def _convert_safety_settings(self, safety_settings: Any) -> List[Dict[str, str]]:
|
147 |
-
"""Convert SDK SafetySetting objects to REST API format"""
|
148 |
-
if not safety_settings:
|
149 |
-
return []
|
150 |
-
|
151 |
-
result = []
|
152 |
-
for setting in safety_settings:
|
153 |
-
if isinstance(setting, dict):
|
154 |
-
result.append(setting)
|
155 |
-
elif hasattr(setting, 'category') and hasattr(setting, 'threshold'):
|
156 |
-
# Convert SDK SafetySetting to dict
|
157 |
-
result.append({
|
158 |
-
'category': setting.category,
|
159 |
-
'threshold': setting.threshold
|
160 |
-
})
|
161 |
-
return result
|
162 |
-
|
163 |
-
def _convert_tools(self, tools: Any) -> List[Dict[str, Any]]:
|
164 |
-
"""Convert SDK Tool objects to REST API format"""
|
165 |
-
if not tools:
|
166 |
-
return []
|
167 |
-
|
168 |
-
result = []
|
169 |
-
for tool in tools:
|
170 |
-
if isinstance(tool, dict):
|
171 |
-
result.append(tool)
|
172 |
-
else:
|
173 |
-
# Convert SDK Tool object to dict
|
174 |
-
result.append(self._convert_tool_item(tool))
|
175 |
-
return result
|
176 |
-
|
177 |
-
def _convert_tool_item(self, tool: Any) -> Dict[str, Any]:
|
178 |
-
"""Convert a single tool item to REST API format"""
|
179 |
-
if isinstance(tool, dict):
|
180 |
-
return tool
|
181 |
-
|
182 |
-
tool_dict = {}
|
183 |
-
|
184 |
-
# Convert all non-private attributes
|
185 |
-
if hasattr(tool, '__dict__'):
|
186 |
-
for attr_name, attr_value in tool.__dict__.items():
|
187 |
-
if not attr_name.startswith('_'):
|
188 |
-
# Convert attribute names from snake_case to camelCase for REST API
|
189 |
-
rest_api_name = self._to_camel_case(attr_name)
|
190 |
-
|
191 |
-
# Special handling for known types
|
192 |
-
if attr_name == 'google_search' and attr_value is not None:
|
193 |
-
tool_dict[rest_api_name] = {} # GoogleSearch is empty object in REST
|
194 |
-
elif attr_name == 'function_declarations' and attr_value is not None:
|
195 |
-
tool_dict[rest_api_name] = attr_value
|
196 |
-
elif attr_value is not None:
|
197 |
-
# Recursively convert any other SDK objects
|
198 |
-
tool_dict[rest_api_name] = self._convert_sdk_object(attr_value)
|
199 |
-
|
200 |
-
return tool_dict
|
201 |
-
|
202 |
-
def _to_camel_case(self, snake_str: str) -> str:
|
203 |
-
"""Convert snake_case to camelCase"""
|
204 |
-
components = snake_str.split('_')
|
205 |
-
return components[0] + ''.join(x.title() for x in components[1:])
|
206 |
-
|
207 |
-
def _convert_sdk_object(self, obj: Any) -> Any:
|
208 |
-
"""Generic SDK object converter"""
|
209 |
-
if isinstance(obj, (str, int, float, bool, type(None))):
|
210 |
-
return obj
|
211 |
-
elif isinstance(obj, dict):
|
212 |
-
return {k: self._convert_sdk_object(v) for k, v in obj.items()}
|
213 |
-
elif isinstance(obj, list):
|
214 |
-
return [self._convert_sdk_object(item) for item in obj]
|
215 |
-
elif hasattr(obj, '__dict__'):
|
216 |
-
# Convert SDK object to dict
|
217 |
-
result = {}
|
218 |
-
for key, value in obj.__dict__.items():
|
219 |
-
if not key.startswith('_'):
|
220 |
-
result[self._to_camel_case(key)] = self._convert_sdk_object(value)
|
221 |
-
return result
|
222 |
-
else:
|
223 |
-
return obj
|
224 |
-
|
225 |
-
async def _generate_content(self, model: str, contents: Any, config: Dict[str, Any], stream: bool = False) -> Any:
|
226 |
-
"""Internal method for content generation"""
|
227 |
-
if not self.project_id:
|
228 |
-
raise ValueError("Project ID not discovered. Call discover_project_id() first.")
|
229 |
-
|
230 |
-
await self._ensure_session()
|
231 |
-
|
232 |
-
# Build URL
|
233 |
-
endpoint = "streamGenerateContent" if stream else "generateContent"
|
234 |
-
url = f"{self.base_url}/projects/{self.project_id}/locations/global/publishers/google/models/{model}:{endpoint}?key={self.api_key}"
|
235 |
-
|
236 |
-
# Convert contents to REST API format
|
237 |
-
payload = {
|
238 |
-
"contents": self._convert_contents(contents)
|
239 |
-
}
|
240 |
-
|
241 |
-
# Extract specific config sections
|
242 |
-
if "system_instruction" in config:
|
243 |
-
# System instruction should be a content object
|
244 |
-
if isinstance(config["system_instruction"], dict):
|
245 |
-
payload["systemInstruction"] = config["system_instruction"]
|
246 |
-
else:
|
247 |
-
payload["systemInstruction"] = self._convert_content_item(config["system_instruction"])
|
248 |
-
|
249 |
-
if "safety_settings" in config:
|
250 |
-
payload["safetySettings"] = self._convert_safety_settings(config["safety_settings"])
|
251 |
-
|
252 |
-
if "tools" in config:
|
253 |
-
payload["tools"] = self._convert_tools(config["tools"])
|
254 |
-
|
255 |
-
# All other config goes under generationConfig
|
256 |
-
generation_config = {}
|
257 |
-
for key, value in config.items():
|
258 |
-
if key not in ["system_instruction", "safety_settings", "tools"]:
|
259 |
-
generation_config[key] = value
|
260 |
-
|
261 |
-
if generation_config:
|
262 |
-
payload["generationConfig"] = generation_config
|
263 |
-
|
264 |
-
try:
|
265 |
-
async with self.session.post(url, json=payload) as response:
|
266 |
-
if response.status != 200:
|
267 |
-
error_data = await response.json()
|
268 |
-
error_msg = error_data.get("error", {}).get("message", f"HTTP {response.status}")
|
269 |
-
raise Exception(f"Vertex AI API error: {error_msg}")
|
270 |
-
|
271 |
-
# Get the JSON response
|
272 |
-
response_data = await response.json()
|
273 |
-
|
274 |
-
# Convert dict to object with attributes for compatibility
|
275 |
-
return self._dict_to_obj(response_data)
|
276 |
-
|
277 |
-
except Exception as e:
|
278 |
-
print(f"ERROR: Direct Vertex API call failed: {e}")
|
279 |
-
raise
|
280 |
-
|
281 |
-
def _dict_to_obj(self, data):
|
282 |
-
"""Convert a dict to an object with attributes"""
|
283 |
-
if isinstance(data, dict):
|
284 |
-
# Create a simple object that allows attribute access
|
285 |
-
class AttrDict:
|
286 |
-
def __init__(self, d):
|
287 |
-
for key, value in d.items():
|
288 |
-
setattr(self, key, self._convert_value(value))
|
289 |
-
|
290 |
-
def _convert_value(self, value):
|
291 |
-
if isinstance(value, dict):
|
292 |
-
return AttrDict(value)
|
293 |
-
elif isinstance(value, list):
|
294 |
-
return [self._convert_value(item) for item in value]
|
295 |
-
else:
|
296 |
-
return value
|
297 |
-
|
298 |
-
return AttrDict(data)
|
299 |
-
elif isinstance(data, list):
|
300 |
-
return [self._dict_to_obj(item) for item in data]
|
301 |
-
else:
|
302 |
-
return data
|
303 |
-
|
304 |
-
async def _generate_content_stream(self, model: str, contents: Any, config: Dict[str, Any]) -> AsyncGenerator:
|
305 |
-
"""Internal method for streaming content generation"""
|
306 |
-
if not self.project_id:
|
307 |
-
raise ValueError("Project ID not discovered. Call discover_project_id() first.")
|
308 |
-
|
309 |
-
await self._ensure_session()
|
310 |
-
|
311 |
-
# Build URL for streaming
|
312 |
-
url = f"{self.base_url}/projects/{self.project_id}/locations/global/publishers/google/models/{model}:streamGenerateContent?key={self.api_key}"
|
313 |
-
|
314 |
-
# Convert contents to REST API format
|
315 |
-
payload = {
|
316 |
-
"contents": self._convert_contents(contents)
|
317 |
-
}
|
318 |
-
|
319 |
-
# Extract specific config sections
|
320 |
-
if "system_instruction" in config:
|
321 |
-
# System instruction should be a content object
|
322 |
-
if isinstance(config["system_instruction"], dict):
|
323 |
-
payload["systemInstruction"] = config["system_instruction"]
|
324 |
-
else:
|
325 |
-
payload["systemInstruction"] = self._convert_content_item(config["system_instruction"])
|
326 |
-
|
327 |
-
if "safety_settings" in config:
|
328 |
-
payload["safetySettings"] = self._convert_safety_settings(config["safety_settings"])
|
329 |
-
|
330 |
-
if "tools" in config:
|
331 |
-
payload["tools"] = self._convert_tools(config["tools"])
|
332 |
-
|
333 |
-
# All other config goes under generationConfig
|
334 |
-
generation_config = {}
|
335 |
-
for key, value in config.items():
|
336 |
-
if key not in ["system_instruction", "safety_settings", "tools"]:
|
337 |
-
generation_config[key] = value
|
338 |
-
|
339 |
-
if generation_config:
|
340 |
-
payload["generationConfig"] = generation_config
|
341 |
-
|
342 |
-
try:
|
343 |
-
async with self.session.post(url, json=payload) as response:
|
344 |
-
if response.status != 200:
|
345 |
-
error_data = await response.json()
|
346 |
-
# Handle array response format
|
347 |
-
if isinstance(error_data, list) and len(error_data) > 0:
|
348 |
-
error_data = error_data[0]
|
349 |
-
error_msg = error_data.get("error", {}).get("message", f"HTTP {response.status}") if isinstance(error_data, dict) else str(error_data)
|
350 |
-
raise Exception(f"Vertex AI API error: {error_msg}")
|
351 |
-
|
352 |
-
# The Vertex AI streaming endpoint returns JSON array elements
|
353 |
-
# We need to parse these as they arrive
|
354 |
-
buffer = ""
|
355 |
-
|
356 |
-
async for chunk in response.content.iter_any():
|
357 |
-
decoded_chunk = chunk.decode('utf-8')
|
358 |
-
buffer += decoded_chunk
|
359 |
-
|
360 |
-
# Try to extract complete JSON objects from the buffer
|
361 |
-
while True:
|
362 |
-
# Skip whitespace and array brackets
|
363 |
-
buffer = buffer.lstrip()
|
364 |
-
if buffer.startswith('['):
|
365 |
-
buffer = buffer[1:].lstrip()
|
366 |
-
continue
|
367 |
-
if buffer.startswith(']'):
|
368 |
-
# End of array
|
369 |
-
return
|
370 |
-
|
371 |
-
# Skip comma and whitespace between objects
|
372 |
-
if buffer.startswith(','):
|
373 |
-
buffer = buffer[1:].lstrip()
|
374 |
-
continue
|
375 |
-
|
376 |
-
# Look for a complete JSON object
|
377 |
-
if buffer.startswith('{'):
|
378 |
-
# Find the matching closing brace
|
379 |
-
brace_count = 0
|
380 |
-
in_string = False
|
381 |
-
escape_next = False
|
382 |
-
|
383 |
-
for i, char in enumerate(buffer):
|
384 |
-
if escape_next:
|
385 |
-
escape_next = False
|
386 |
-
continue
|
387 |
-
|
388 |
-
if char == '\\' and in_string:
|
389 |
-
escape_next = True
|
390 |
-
continue
|
391 |
-
|
392 |
-
if char == '"' and not in_string:
|
393 |
-
in_string = True
|
394 |
-
elif char == '"' and in_string:
|
395 |
-
in_string = False
|
396 |
-
elif char == '{' and not in_string:
|
397 |
-
brace_count += 1
|
398 |
-
elif char == '}' and not in_string:
|
399 |
-
brace_count -= 1
|
400 |
-
|
401 |
-
if brace_count == 0:
|
402 |
-
# Found complete object
|
403 |
-
obj_str = buffer[:i+1]
|
404 |
-
buffer = buffer[i+1:]
|
405 |
-
|
406 |
-
try:
|
407 |
-
chunk_data = json.loads(obj_str)
|
408 |
-
converted_obj = self._dict_to_obj(chunk_data)
|
409 |
-
yield converted_obj
|
410 |
-
except json.JSONDecodeError as e:
|
411 |
-
print(f"ERROR: DirectVertexClient - Failed to parse JSON: {e}")
|
412 |
-
|
413 |
-
break
|
414 |
-
else:
|
415 |
-
# No complete object found, need more data
|
416 |
-
break
|
417 |
-
else:
|
418 |
-
# No more objects to process in current buffer
|
419 |
-
break
|
420 |
-
|
421 |
-
except Exception as e:
|
422 |
-
print(f"ERROR: Direct Vertex streaming API call failed: {e}")
|
423 |
-
raise
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
app/project_id_discovery.py
ADDED
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import aiohttp
|
2 |
+
import json
|
3 |
+
import re
|
4 |
+
from typing import Dict, Optional
|
5 |
+
|
6 |
+
# Global cache for project IDs: {api_key: project_id}
|
7 |
+
PROJECT_ID_CACHE: Dict[str, str] = {}
|
8 |
+
|
9 |
+
|
10 |
+
async def discover_project_id(api_key: str) -> str:
|
11 |
+
"""
|
12 |
+
Discover project ID by triggering an intentional error with a non-existent model.
|
13 |
+
The project ID is extracted from the error message and cached for future use.
|
14 |
+
|
15 |
+
Args:
|
16 |
+
api_key: The Vertex AI Express API key
|
17 |
+
|
18 |
+
Returns:
|
19 |
+
The discovered project ID
|
20 |
+
|
21 |
+
Raises:
|
22 |
+
Exception: If project ID discovery fails
|
23 |
+
"""
|
24 |
+
# Check cache first
|
25 |
+
if api_key in PROJECT_ID_CACHE:
|
26 |
+
print(f"INFO: Using cached project ID: {PROJECT_ID_CACHE[api_key]}")
|
27 |
+
return PROJECT_ID_CACHE[api_key]
|
28 |
+
|
29 |
+
# Use a non-existent model to trigger error
|
30 |
+
error_url = f"https://aiplatform.googleapis.com/v1/publishers/google/models/gemini-2.7-pro-preview-05-06:streamGenerateContent?key={api_key}"
|
31 |
+
|
32 |
+
# Create minimal request payload
|
33 |
+
payload = {
|
34 |
+
"contents": [{"role": "user", "parts": [{"text": "test"}]}]
|
35 |
+
}
|
36 |
+
|
37 |
+
async with aiohttp.ClientSession() as session:
|
38 |
+
try:
|
39 |
+
async with session.post(error_url, json=payload) as response:
|
40 |
+
response_text = await response.text()
|
41 |
+
|
42 |
+
try:
|
43 |
+
# Try to parse as JSON first
|
44 |
+
error_data = json.loads(response_text)
|
45 |
+
|
46 |
+
# Handle array response format
|
47 |
+
if isinstance(error_data, list) and len(error_data) > 0:
|
48 |
+
error_data = error_data[0]
|
49 |
+
|
50 |
+
if "error" in error_data:
|
51 |
+
error_message = error_data["error"].get("message", "")
|
52 |
+
# Extract project ID from error message
|
53 |
+
# Pattern: "projects/39982734461/locations/..."
|
54 |
+
match = re.search(r'projects/(\d+)/locations/', error_message)
|
55 |
+
if match:
|
56 |
+
project_id = match.group(1)
|
57 |
+
PROJECT_ID_CACHE[api_key] = project_id
|
58 |
+
print(f"INFO: Discovered project ID: {project_id}")
|
59 |
+
return project_id
|
60 |
+
except json.JSONDecodeError:
|
61 |
+
# If not JSON, try to find project ID in raw text
|
62 |
+
match = re.search(r'projects/(\d+)/locations/', response_text)
|
63 |
+
if match:
|
64 |
+
project_id = match.group(1)
|
65 |
+
PROJECT_ID_CACHE[api_key] = project_id
|
66 |
+
print(f"INFO: Discovered project ID from raw response: {project_id}")
|
67 |
+
return project_id
|
68 |
+
|
69 |
+
raise Exception(f"Failed to discover project ID. Status: {response.status}, Response: {response_text[:500]}")
|
70 |
+
|
71 |
+
except Exception as e:
|
72 |
+
print(f"ERROR: Failed to discover project ID: {e}")
|
73 |
+
raise
|
app/routes/chat_api.py
CHANGED
@@ -24,7 +24,7 @@ from api_helpers import (
|
|
24 |
execute_gemini_call,
|
25 |
)
|
26 |
from openai_handler import OpenAIDirectHandler
|
27 |
-
from
|
28 |
|
29 |
router = APIRouter()
|
30 |
|
@@ -118,9 +118,14 @@ async def chat_completions(fastapi_request: Request, request: OpenAIRequest, api
|
|
118 |
try:
|
119 |
# Check if model contains "gemini-2.5-pro" or "gemini-2.5-flash" for direct URL approach
|
120 |
if "gemini-2.5-pro" in base_model_name or "gemini-2.5-flash" in base_model_name:
|
121 |
-
|
122 |
-
|
123 |
-
|
|
|
|
|
|
|
|
|
|
|
124 |
else:
|
125 |
client_to_use = genai.Client(vertexai=True, api_key=key_val)
|
126 |
print(f"INFO: Attempt {attempt+1}/{total_keys} - Using Vertex Express Mode SDK for model {request.model} (base: {base_model_name}) with API key (original index: {original_idx}).")
|
@@ -185,9 +190,6 @@ async def chat_completions(fastapi_request: Request, request: OpenAIRequest, api
|
|
185 |
try:
|
186 |
# Pass is_auto_attempt=True for auto-mode calls
|
187 |
result = await execute_gemini_call(client_to_use, attempt["model"], attempt["prompt_func"], current_gen_config, request, is_auto_attempt=True)
|
188 |
-
# Clean up DirectVertexClient session if used
|
189 |
-
if isinstance(client_to_use, DirectVertexClient):
|
190 |
-
await client_to_use.close()
|
191 |
return result
|
192 |
except Exception as e_auto:
|
193 |
last_err = e_auto
|
@@ -196,9 +198,6 @@ async def chat_completions(fastapi_request: Request, request: OpenAIRequest, api
|
|
196 |
|
197 |
print(f"All auto attempts failed. Last error: {last_err}")
|
198 |
err_msg = f"All auto-mode attempts failed for model {request.model}. Last error: {str(last_err)}"
|
199 |
-
# Clean up DirectVertexClient session if used
|
200 |
-
if isinstance(client_to_use, DirectVertexClient):
|
201 |
-
await client_to_use.close()
|
202 |
if not request.stream and last_err:
|
203 |
return JSONResponse(status_code=500, content=create_openai_error_response(500, err_msg, "server_error"))
|
204 |
elif request.stream:
|
@@ -245,17 +244,9 @@ async def chat_completions(fastapi_request: Request, request: OpenAIRequest, api
|
|
245 |
# but the API call might need the full "gemini-1.5-pro-search".
|
246 |
# Let's use `request.model` for the API call here, and `base_model_name` for checks like Express eligibility.
|
247 |
# For non-auto mode, is_auto_attempt defaults to False in execute_gemini_call
|
248 |
-
|
249 |
-
return await execute_gemini_call(client_to_use, base_model_name, current_prompt_func, generation_config, request)
|
250 |
-
finally:
|
251 |
-
# Clean up DirectVertexClient session if used
|
252 |
-
if isinstance(client_to_use, DirectVertexClient):
|
253 |
-
await client_to_use.close()
|
254 |
|
255 |
except Exception as e:
|
256 |
error_msg = f"Unexpected error in chat_completions endpoint: {str(e)}"
|
257 |
print(error_msg)
|
258 |
-
# Clean up DirectVertexClient session if it exists
|
259 |
-
if 'client_to_use' in locals() and isinstance(client_to_use, DirectVertexClient):
|
260 |
-
await client_to_use.close()
|
261 |
return JSONResponse(status_code=500, content=create_openai_error_response(500, error_msg, "server_error"))
|
|
|
24 |
execute_gemini_call,
|
25 |
)
|
26 |
from openai_handler import OpenAIDirectHandler
|
27 |
+
from project_id_discovery import discover_project_id
|
28 |
|
29 |
router = APIRouter()
|
30 |
|
|
|
118 |
try:
|
119 |
# Check if model contains "gemini-2.5-pro" or "gemini-2.5-flash" for direct URL approach
|
120 |
if "gemini-2.5-pro" in base_model_name or "gemini-2.5-flash" in base_model_name:
|
121 |
+
project_id = await discover_project_id(key_val)
|
122 |
+
base_url = f"https://aiplatform.googleapis.com/v1/projects/{project_id}/locations/global"
|
123 |
+
client_to_use = genai.Client(
|
124 |
+
vertexai=True,
|
125 |
+
api_key=key_val,
|
126 |
+
http_options=types.HttpOptions(base_url=base_url)
|
127 |
+
)
|
128 |
+
print(f"INFO: Attempt {attempt+1}/{total_keys} - Using Vertex Express Mode with custom base URL for model {request.model} (base: {base_model_name}) with API key (original index: {original_idx}).")
|
129 |
else:
|
130 |
client_to_use = genai.Client(vertexai=True, api_key=key_val)
|
131 |
print(f"INFO: Attempt {attempt+1}/{total_keys} - Using Vertex Express Mode SDK for model {request.model} (base: {base_model_name}) with API key (original index: {original_idx}).")
|
|
|
190 |
try:
|
191 |
# Pass is_auto_attempt=True for auto-mode calls
|
192 |
result = await execute_gemini_call(client_to_use, attempt["model"], attempt["prompt_func"], current_gen_config, request, is_auto_attempt=True)
|
|
|
|
|
|
|
193 |
return result
|
194 |
except Exception as e_auto:
|
195 |
last_err = e_auto
|
|
|
198 |
|
199 |
print(f"All auto attempts failed. Last error: {last_err}")
|
200 |
err_msg = f"All auto-mode attempts failed for model {request.model}. Last error: {str(last_err)}"
|
|
|
|
|
|
|
201 |
if not request.stream and last_err:
|
202 |
return JSONResponse(status_code=500, content=create_openai_error_response(500, err_msg, "server_error"))
|
203 |
elif request.stream:
|
|
|
244 |
# but the API call might need the full "gemini-1.5-pro-search".
|
245 |
# Let's use `request.model` for the API call here, and `base_model_name` for checks like Express eligibility.
|
246 |
# For non-auto mode, is_auto_attempt defaults to False in execute_gemini_call
|
247 |
+
return await execute_gemini_call(client_to_use, base_model_name, current_prompt_func, generation_config, request)
|
|
|
|
|
|
|
|
|
|
|
248 |
|
249 |
except Exception as e:
|
250 |
error_msg = f"Unexpected error in chat_completions endpoint: {str(e)}"
|
251 |
print(error_msg)
|
|
|
|
|
|
|
252 |
return JSONResponse(status_code=500, content=create_openai_error_response(500, error_msg, "server_error"))
|