mirrors/gpt4free
1
Fork 0
mirror of https://github.com/xtekky/gpt4free.git synced 2025-12-05 18:20:35 -08:00
Code Issues Projects Releases Packages Wiki Activity

Improve Yupp provider account handling , request timeout and get byte from url (#3249)

Browse source 
* Add image caching to Yupp provider

Introduces an image cache to avoid redundant uploads in the Yupp provider. Refactors media attachment handling into a new prepare_files method, improving efficiency and code organization. Updates .gitignore to exclude .idea directory.

* Refactor Yupp stream handling and chunk processing

Improves stream segmentation in the Yupp provider by introducing buffers for target, variant, quick, thinking, and extra streams. Refactors chunk processing to better handle image-gen, quick responses, and variant outputs, and adds more robust stream ID extraction and routing logic. Yields a consolidated JsonResponse with all stream segments for downstream use.

* Handle ClientResponseError in Yupp provider

Adds specific handling for aiohttp ClientResponseError in the Yupp provider. Marks account as invalid on 500 Internal Server Error, otherwise increments error count and raises ProviderException for other errors.

* Update Yupp.py

fix 429  'Too Many Requests'

* Update Yupp.py

* Improve Yupp provider account handling and request timeout

Refactored account loading to preserve account history and error counts when updating tokens. Enhanced request logic to support custom timeouts using aiohttp's ClientTimeout, allowing for more flexible timeout configuration.

* Update __init__.py

* Handle multi-line <think> and <yapp> blocks in Yupp

Added logic to capture and process multi-line <think> and <yapp class="image-gen"> blocks referenced by special IDs. Introduced block storage and extraction functions, enabling reasoning and image-gen content to be handled via references in the response stream.

* Update LMArena.py

Not Found Model error

* Refactor to use StreamSession in Qwen and Yupp providers

Replaced aiohttp.ClientSession with StreamSession in Qwen.py and Yupp.py for improved session handling. Updated exception and timeout references in Yupp.py to use aiohttp types. Improved default argument handling in StreamSession initialization.

* Update Yupp.py

* Add status parameter to get_generated_image method

Introduces a 'status' parameter to the get_generated_image method to allow passing image generation status. Updates method calls and response objects to include status in their metadata for improved tracking of image generation progress.

* Update OpenaiChat.py

* Refactor Qwen image upload and caching logic and token

Reworked the image upload flow in Qwen provider to use direct file uploads with OSS headers, added caching for uploaded images, and improved file type detection. Updated prepare_files to handle uploads via session and cache results, and added utility for generating OSS headers. Minor imports and typing adjustments included and token support.

* Refactor Qwen and Yupp providers for improved async handling

Updated Qwen provider to handle timeout via kwargs and improved type annotations. Refactored Yupp provider for better code organization, formatting, and async account rotation logic. Enhanced readability and maintainability by reordering imports, adding whitespace, and clarifying function implementations.

* Add image caching to OpenaiChat provider

Introduces an image cache mechanism to OpenaiChat for uploaded images, reducing redundant uploads and improving efficiency. Also refactors code for clarity, updates type hints, and makes minor formatting improvements throughout the file.
This commit is contained in:
Ammar 2025-11-26 15:02:51 +02:00 • committed by GitHub
parent 32215bb7bb
commit 7771cf3d43
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 623 additions and 289 deletions
Download patch file Download diff file Expand all files Collapse all files

220
g4f/Provider/Qwen.py
View file

@ -1,22 +1,29 @@
from __future__ import annotations
import asyncio
import datetime
import hashlib
import hmac
import json
import mimetypes
import re
import uuid
from time import time
from typing import Literal, Optional
from typing import Literal, Optional, Dict
from urllib.parse import quote
import aiohttp
from ..errors import RateLimitError, ResponseError
from ..typing import AsyncResult, Messages, MediaListType
from ..providers.response import JsonConversation, Reasoning, Usage, ImageResponse, FinishReason
from ..requests import sse_stream
from ..tools.media import merge_media
from g4f.image import to_bytes, detect_file_type
from g4f.requests import raise_for_status
from .base_provider import AsyncGeneratorProvider, ProviderModelMixin
from .helper import get_last_user_message
from .. import debug
from ..errors import RateLimitError, ResponseError
from ..providers.response import JsonConversation, Reasoning, Usage, ImageResponse, FinishReason
from ..requests import sse_stream
from ..requests.aiohttp import StreamSession
from ..tools.media import merge_media
from ..typing import AsyncResult, Messages, MediaListType
try:
import curl_cffi
@ -25,6 +32,56 @@ try:
except ImportError:
has_curl_cffi = False
# Global variables to manage Qwen Image Cache
ImagesCache: Dict[str, dict] = {}
def get_oss_headers(method: str, date_str: str, sts_data: dict, content_type: str) -> dict[str, str]:
bucket_name = sts_data.get('bucketname', 'qwen-webui-prod')
file_path = sts_data.get('file_path', '')
access_key_id = sts_data.get('access_key_id')
access_key_secret = sts_data.get('access_key_secret')
security_token = sts_data.get('security_token')
headers = {
'Content-Type': content_type,
'x-oss-content-sha256': 'UNSIGNED-PAYLOAD',
'x-oss-date': date_str,
'x-oss-security-token': security_token,
'x-oss-user-agent': 'aliyun-sdk-js/6.23.0 Chrome 132.0.0.0 on Windows 10 64-bit'
}
headers_lower = {k.lower(): v for k, v in headers.items()}
canonical_headers_list = []
signed_headers_list = []
required_headers = ['content-md5', 'content-type', 'x-oss-content-sha256', 'x-oss-date', 'x-oss-security-token',
'x-oss-user-agent']
for header_name in sorted(required_headers):
if header_name in headers_lower:
canonical_headers_list.append(f"{header_name}:{headers_lower[header_name]}")
signed_headers_list.append(header_name)
canonical_headers = '\n'.join(canonical_headers_list) + '\n'
canonical_uri = f"/{bucket_name}/{quote(file_path, safe='/')}"
canonical_request = f"{method}\n{canonical_uri}\n\n{canonical_headers}\n\nUNSIGNED-PAYLOAD"
date_parts = date_str.split('T')
date_scope = f"{date_parts[0]}/ap-southeast-1/oss/aliyun_v4_request"
string_to_sign = f"OSS4-HMAC-SHA256\n{date_str}\n{date_scope}\n{hashlib.sha256(canonical_request.encode()).hexdigest()}"
def sign(key, msg):
return hmac.new(key, msg.encode() if isinstance(msg, str) else msg, hashlib.sha256).digest()
date_key = sign(f"aliyun_v4{access_key_secret}".encode(), date_parts[0])
region_key = sign(date_key, "ap-southeast-1")
service_key = sign(region_key, "oss")
signing_key = sign(service_key, "aliyun_v4_request")
signature = hmac.new(signing_key, string_to_sign.encode(), hashlib.sha256).hexdigest()
headers['authorization'] = f"OSS4-HMAC-SHA256 Credential={access_key_id}/{date_scope},Signature={signature}"
return headers
text_models = [
'qwen3-max-preview', 'qwen-plus-2025-09-11', 'qwen3-235b-a22b', 'qwen3-coder-plus', 'qwen3-30b-a3b',
'qwen3-coder-30b-a3b-instruct', 'qwen-max-latest', 'qwen-plus-2025-01-25', 'qwq-32b', 'qwen-turbo-2025-02-11',
@ -60,19 +117,19 @@ class Qwen(AsyncGeneratorProvider, ProviderModelMixin):
active_by_default = True
supports_stream = True
supports_message_history = False
image_cache = True
_models_loaded = True
image_models = image_models
text_models = text_models
vision_models = vision_models
models = models
models: list[str] = models
default_model = "qwen3-235b-a22b"
_midtoken: str = None
_midtoken_uses: int = 0
@classmethod
def get_models(cls) -> list[str]:
def get_models(cls, **kwargs) -> list[str]:
if not cls._models_loaded and has_curl_cffi:
response = curl_cffi.get(f"{cls.url}/api/models")
if response.ok:
@ -97,34 +154,106 @@ class Qwen(AsyncGeneratorProvider, ProviderModelMixin):
return cls.models
@classmethod
async def prepare_files(cls, media, chat_type="")->list:
async def prepare_files(cls, media, session: aiohttp.ClientSession, headers=None) -> list:
if headers is None:
headers = {}
files = []
for _file, file_name in media:
file_type, _ = mimetypes.guess_type(file_name)
file_class: Literal["default", "vision", "video", "audio", "document"] = "default"
_type: Literal["file", "image", "video", "audio"] = "file"
showType: Literal["file", "image", "video", "audio"] = "file"
for index, (_file, file_name) in enumerate(media):
if isinstance(_file, str) and _file.startswith('http'):
if chat_type == "image_edit" or (file_type and file_type.startswith("image")):
file_class = "vision"
_type = "image"
if not file_type:
# Try to infer from file extension, fallback to generic
ext = file_name.split('.')[-1].lower() if '.' in file_name else ''
file_type = mimetypes.types_map.get(f'.{ext}', 'application/octet-stream')
showType = "image"
data_bytes = to_bytes(_file)
# Check Cache
hasher = hashlib.md5()
hasher.update(data_bytes)
image_hash = hasher.hexdigest()
file = ImagesCache.get(image_hash)
if cls.image_cache and file:
debug.log("Using cached image")
files.append(file)
continue
files.append(
{
"type": _type,
extension, file_type = detect_file_type(data_bytes)
file_name = file_name or f"file-{len(data_bytes)}{extension}"
file_size = len(data_bytes)
# Get File Url
async with session.post(
f'{cls.url}/api/v2/files/getstsToken',
json={"filename": file_name,
"filesize": file_size, "filetype": file_type},
headers=headers
) as r:
await raise_for_status(r, "Create file failed")
res_data = await r.json()
data = res_data.get("data")
if res_data["success"] is False:
raise RateLimitError(f"{data['code']}:{data['details']}")
file_url = data.get("file_url")
file_id = data.get("file_id")
# Put File into Url
str_date = datetime.datetime.now(datetime.UTC).strftime('%Y%m%dT%H%M%SZ')
headers = get_oss_headers('PUT', str_date, data, file_type)
async with session.put(
file_url.split("?")[0],
data=data_bytes,
headers=headers
) as response:
await raise_for_status(response)
file_class: Literal["default", "vision", "video", "audio", "document"]
_type: Literal["file", "image", "video", "audio"]
show_type: Literal["file", "image", "video", "audio"]
if "image" in file_type:
_type = "image"
show_type = "image"
file_class = "vision"
elif "video" in file_type:
_type = "video"
show_type = "video"
file_class = "video"
elif "audio" in file_type:
_type = "audio"
show_type = "audio"
file_class = "audio"
else:
_type = "file"
show_type = "file"
file_class = "document"
file = {
"type": _type,
"file": {
"created_at": int(time() * 1000),
"data": {},
"filename": file_name,
"hash": None,
"id": file_id,
"meta": {
"name": file_name,
"file_type": file_type,
"showType": showType,
"file_class": file_class,
"url": _file
}
)
"size": file_size,
"content_type": file_type
},
"update_at": int(time() * 1000),
},
"id": file_id,
"url": file_url,
"name": file_name,
"collection_name": "",
"progress": 0,
"status": "uploaded",
"greenNet": "success",
"size": file_size,
"error": "",
"itemId": str(uuid.uuid4()),
"file_type": file_type,
"showType": show_type,
"file_class": file_class,
"uploadTaskId": str(uuid.uuid4())
}
ImagesCache[image_hash] = file
files.append(file)
return files
@classmethod
@ -135,7 +264,6 @@ class Qwen(AsyncGeneratorProvider, ProviderModelMixin):
media: MediaListType = None,
conversation: JsonConversation = None,
proxy: str = None,
timeout: int = 120,
stream: bool = True,
enable_thinking: bool = True,
chat_type: Literal[
@ -157,7 +285,7 @@ class Qwen(AsyncGeneratorProvider, ProviderModelMixin):
"""
model_name = cls.get_model(model)
token = kwargs.get("token")
headers = {
'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/138.0.0.0 Safari/537.36',
'Accept': '*/*',
@ -169,13 +297,24 @@ class Qwen(AsyncGeneratorProvider, ProviderModelMixin):
'Sec-Fetch-Mode': 'cors',
'Sec-Fetch-Site': 'same-origin',
'Connection': 'keep-alive',
'Authorization': 'Bearer',
'Authorization': f'Bearer {token}' if token else "Bearer",
'Source': 'web'
}
prompt = get_last_user_message(messages)
async with aiohttp.ClientSession(headers=headers) as session:
_timeout = kwargs.get("timeout")
if isinstance(_timeout, aiohttp.ClientTimeout):
timeout = _timeout
else:
total = float(_timeout) if isinstance(_timeout, (int, float)) else 5 * 60
timeout = aiohttp.ClientTimeout(total=total)
async with StreamSession(headers=headers) as session:
try:
async with session.get('https://chat.qwen.ai/api/v1/auths/', proxy=proxy) as user_info_res:
user_info_res.raise_for_status()
debug.log(await user_info_res.json())
except:
...
for attempt in range(5):
try:
if not cls._midtoken:
@ -221,7 +360,8 @@ class Qwen(AsyncGeneratorProvider, ProviderModelMixin):
files = []
media = list(merge_media(media, messages))
if media:
files = await cls.prepare_files(media, chat_type=chat_type)
files = await cls.prepare_files(media, session=session,
headers=req_headers)
msg_payload = {
"stream": stream,

264
g4f/Provider/Yupp.py
View file

@ -1,25 +1,26 @@
import asyncio
import hashlib
import json
import os
import re
import time
import uuid
import re
import os
import asyncio
import aiohttp
from aiohttp import ClientResponseError
from ..typing import AsyncResult, Messages, Optional, Dict, Any, List
import aiohttp
from .helper import get_last_user_message
from .yupp.models import YuppModelManager
from ..cookies import get_cookies
from ..debug import log
from ..errors import RateLimitError, ProviderException, MissingAuthError
from ..image import is_accepted_format, to_bytes
from ..providers.base_provider import AsyncGeneratorProvider, ProviderModelMixin
from ..providers.response import Reasoning, PlainTextResponse, PreviewResponse, JsonConversation, ImageResponse, \
ProviderInfo, FinishReason, JsonResponse
from ..errors import RateLimitError, ProviderException, MissingAuthError
from ..cookies import get_cookies
from ..requests.aiohttp import StreamSession
from ..tools.auth import AuthManager
from ..tools.media import merge_media
from ..image import is_accepted_format, to_bytes
from .yupp.models import YuppModelManager
from .helper import get_last_user_message
from ..debug import log
from ..typing import AsyncResult, Messages, Optional, Dict, Any, List
# Global variables to manage Yupp accounts
YUPP_ACCOUNT = Dict[str, Any]
@ -27,22 +28,24 @@ YUPP_ACCOUNTS: List[YUPP_ACCOUNT] = []
account_rotation_lock = asyncio.Lock()
# Global variables to manage Yupp Image Cache
ImagesCache:Dict[str, dict] = {}
ImagesCache: Dict[str, dict] = {}
class YuppAccount:
"""Yupp account representation"""
def __init__(self, token: str, is_valid: bool = True, error_count: int = 0, last_used: float = 0):
self.token = token
self.is_valid = is_valid
self.error_count = error_count
self.last_used = last_used
def load_yupp_accounts(tokens_str: str):
"""Load Yupp accounts from token string"""
global YUPP_ACCOUNTS
if not tokens_str:
return
tokens = [token.strip() for token in tokens_str.split(',') if token.strip()]
YUPP_ACCOUNTS = [
{
@ -54,6 +57,7 @@ def load_yupp_accounts(tokens_str: str):
for token in tokens
]
def create_headers() -> Dict[str, str]:
"""Create headers for requests"""
return {
@ -66,6 +70,7 @@ def create_headers() -> Dict[str, str]:
"Sec-Fetch-Site": "same-origin",
}
async def get_best_yupp_account() -> Optional[YUPP_ACCOUNT]:
"""Get the best available Yupp account using smart selection algorithm"""
max_error_count = int(os.getenv("MAX_ERROR_COUNT", "3"))
@ -77,10 +82,10 @@ async def get_best_yupp_account() -> Optional[YUPP_ACCOUNT]:
acc
for acc in YUPP_ACCOUNTS
if acc["is_valid"]
and (
acc["error_count"] < max_error_count
or now - acc["last_used"] > error_cooldown
)
and (
acc["error_count"] < max_error_count
or now - acc["last_used"] > error_cooldown
)
]
if not valid_accounts:
@ -89,8 +94,8 @@ async def get_best_yupp_account() -> Optional[YUPP_ACCOUNT]:
# Reset error count for accounts in cooldown
for acc in valid_accounts:
if (
acc["error_count"] >= max_error_count
and now - acc["last_used"] > error_cooldown
acc["error_count"] >= max_error_count
and now - acc["last_used"] > error_cooldown
):
acc["error_count"] = 0
@ -100,6 +105,7 @@ async def get_best_yupp_account() -> Optional[YUPP_ACCOUNT]:
account["last_used"] = now
return account
async def claim_yupp_reward(session: aiohttp.ClientSession, account: YUPP_ACCOUNT, reward_id: str):
"""Claim Yupp reward asynchronously"""
try:
@ -109,7 +115,7 @@ async def claim_yupp_reward(session: aiohttp.ClientSession, account: YUPP_ACCOUN
headers = {
"Content-Type": "application/json",
"Cookie": f"__Secure-yupp.session-token={account['token']}",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36 Edg/137.0.0.0",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36 Edg/137.0.0.0",
}
async with session.post(url, json=payload, headers=headers) as response:
@ -122,6 +128,7 @@ async def claim_yupp_reward(session: aiohttp.ClientSession, account: YUPP_ACCOUN
log_debug(f"Failed to claim reward {reward_id}. Error: {e}")
return None
async def make_chat_private(session: aiohttp.ClientSession, account: YUPP_ACCOUNT, chat_id: str) -> bool:
"""Set a Yupp chat's sharing status to PRIVATE"""
try:
@ -138,7 +145,7 @@ async def make_chat_private(session: aiohttp.ClientSession, account: YUPP_ACCOUN
headers = {
"Content-Type": "application/json",
"Cookie": f"__Secure-yupp.session-token={account['token']}",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36 Edg/137.0.0.0",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36 Edg/137.0.0.0",
}
@ -146,8 +153,8 @@ async def make_chat_private(session: aiohttp.ClientSession, account: YUPP_ACCOUN
response.raise_for_status()
data = await response.json()
if (
isinstance(data, list) and len(data) > 0
and "json" in data[0].get("result", {}).get("data", {})
isinstance(data, list) and len(data) > 0
and "json" in data[0].get("result", {}).get("data", {})
):
log_debug(f"Chat {chat_id} is now PRIVATE ✅")
return True
@ -159,6 +166,7 @@ async def make_chat_private(session: aiohttp.ClientSession, account: YUPP_ACCOUN
log_debug(f"Failed to make chat {chat_id} private: {e}")
return False
def log_debug(message: str):
"""Debug logging"""
if os.getenv("DEBUG_MODE", "false").lower() == "true":
@ -166,11 +174,12 @@ def log_debug(message: str):
else:
log(f"[Yupp] {message}")
def format_messages_for_yupp(messages: Messages) -> str:
"""Format multi-turn conversation for Yupp single-turn format"""
if not messages:
return ""
if len(messages) == 1 and isinstance(messages[0].get("content"), str):
return messages[0].get("content", "").strip()
@ -202,6 +211,7 @@ def format_messages_for_yupp(messages: Messages) -> str:
return result
class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
"""
Yupp.ai Provider for g4f
@ -214,7 +224,7 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
active_by_default = True
supports_stream = True
image_cache = True
@classmethod
def get_models(cls, api_key: str = None, **kwargs) -> List[str]:
if not cls.models:
@ -230,11 +240,12 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
cls.models_tags = {model.get("name"): manager.processor.generate_tags(model) for model in models}
cls.models = [model.get("name") for model in models]
cls.image_models = [model.get("name") for model in models if model.get("isImageGeneration")]
cls.vision_models = [model.get("name") for model in models if "image/*" in model.get("supportedAttachmentMimeTypes", [])]
cls.vision_models = [model.get("name") for model in models if
"image/*" in model.get("supportedAttachmentMimeTypes", [])]
return cls.models
@classmethod
async def prepare_files(cls, media, session:aiohttp.ClientSession, account:YUPP_ACCOUNT)->list:
async def prepare_files(cls, media, session: aiohttp.ClientSession, account: YUPP_ACCOUNT) -> list:
files = []
if not media:
return files
@ -291,11 +302,11 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
@classmethod
async def create_async_generator(
cls,
model: str,
messages: Messages,
proxy: str = None,
**kwargs,
cls,
model: str,
messages: Messages,
proxy: str = None,
**kwargs,
) -> AsyncResult:
"""
Create async completion using Yupp.ai API with account rotation
@ -314,15 +325,16 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
conversation = kwargs.get("conversation")
url_uuid = conversation.url_uuid if conversation else None
is_new_conversation = url_uuid is None
prompt = kwargs.get("prompt")
if prompt is None:
if is_new_conversation:
prompt = format_messages_for_yupp(messages)
else:
prompt = get_last_user_message(messages, prompt)
log_debug(f"Use url_uuid: {url_uuid}, Formatted prompt length: {len(prompt)}, Is new conversation: {is_new_conversation}")
log_debug(
f"Use url_uuid: {url_uuid}, Formatted prompt length: {len(prompt)}, Is new conversation: {is_new_conversation}")
# Try all accounts with rotation
max_attempts = len(YUPP_ACCOUNTS)
@ -332,10 +344,9 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
raise ProviderException("No valid Yupp accounts available")
try:
async with aiohttp.ClientSession() as session:
async with StreamSession() as session:
turn_id = str(uuid.uuid4())
# Handle media attachments
media = kwargs.get("media")
if media:
@ -390,16 +401,23 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
log_debug(f"Sending request to: {url}")
log_debug(f"Payload structure: {type(payload)}, length: {len(str(payload))}")
_timeout = kwargs.get("timeout")
if isinstance(_timeout, aiohttp.ClientTimeout):
timeout = _timeout
else:
total = float(_timeout) if isinstance(_timeout, (int, float)) else 5 * 60
timeout = aiohttp.ClientTimeout(total=total)
# Send request
async with session.post(url, json=payload, headers=headers, proxy=proxy) as response:
async with session.post(url, json=payload, headers=headers, proxy=proxy,
timeout=timeout) as response:
response.raise_for_status()
# Make chat private in background
asyncio.create_task(make_chat_private(session, account, url_uuid))
# Process stream
async for chunk in cls._process_stream_response(response.content, account, session, prompt, model):
async for chunk in cls._process_stream_response(response.content, account, session, prompt,
model):
yield chunk
return
@ -417,7 +435,7 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
else:
account["error_count"] += 1
continue
except ClientResponseError as e:
except aiohttp.ClientResponseError as e:
log_debug(f"Account ...{account['token'][-4:]} failed: {str(e)}")
# No Available Yupp credits
if e.status == 500 and 'Internal Server Error' in e.message:
@ -439,12 +457,12 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
@classmethod
async def _process_stream_response(
cls,
response_content,
account: YUPP_ACCOUNT,
session: aiohttp.ClientSession,
prompt: str,
model_id: str
cls,
response_content,
account: YUPP_ACCOUNT,
session: aiohttp.ClientSession,
prompt: str,
model_id: str
) -> AsyncResult:
"""Process Yupp stream response asynchronously"""
@ -461,15 +479,33 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
"target": [],
"variant": [],
"quick": [],
"thinking": [] ,
"thinking": [],
"extra": []
}
# Holds leftStream / rightStream definitions to determine target/variant
select_stream = [None, None]
# State for capturing a multi-line <think> + <yapp> block (fa-style)
capturing_ref_id: Optional[str] = None
capturing_lines: List[bytes] = []
# Storage for special referenced blocks like $fa
think_blocks: Dict[str, str] = {}
image_blocks: Dict[str, str] = {}
def extract_ref_id(ref):
"""Extract ID from reference string, e.g., from '$@123' extract '123'"""
return ref[2:] if ref and isinstance(ref, str) and ref.startswith("$@") else None
def extract_ref_name(ref: str) -> Optional[str]:
"""Extract simple ref name from '$fa' → 'fa'"""
if not isinstance(ref, str):
return None
if ref.startswith("$@"):
return ref[2:]
if ref.startswith("$") and len(ref) > 1:
return ref[1:]
return None
def is_valid_content(content: str) -> bool:
"""Check if content is valid"""
if not content or content in [None, "", "$undefined"]:
@ -515,7 +551,27 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
if for_target:
normal_content += content
yield content
def finalize_capture_block(ref_id: str, lines: List[bytes]):
"""Parse captured <think> + <yapp> block for a given ref ID."""
text = b"".join(lines).decode("utf-8", errors="ignore")
# Extract <think>...</think>
think_start = text.find("<think>")
think_end = text.find("</think>")
if think_start != -1 and think_end != -1 and think_end > think_start:
inner = text[think_start + len("<think>"):think_end].strip()
if inner:
think_blocks[ref_id] = inner
# Extract <yapp class="image-gen">...</yapp>
yapp_start = text.find('<yapp class="image-gen">')
if yapp_start != -1:
yapp_end = text.find("</yapp>", yapp_start)
if yapp_end != -1:
yapp_block = text[yapp_start:yapp_end + len("</yapp>")]
image_blocks[ref_id] = yapp_block
try:
line_count = 0
quick_response_id = None
@ -531,17 +587,55 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
right_message_id = None
nudge_new_chat_id = None
nudge_new_chat = False
async for line in response_content:
line_count += 1
# If we are currently capturing a think/image block for some ref ID
if capturing_ref_id is not None:
capturing_lines.append(line)
# Check if this line closes the <yapp> block; after that, block is complete
if b"</yapp>" in line: # or b':{"curr"' in line:
# We may have trailing "2:{...}" after </yapp> on the same line
# Get id using re
idx = line.find(b"</yapp>")
suffix = line[idx + len(b"</yapp>"):]
# Finalize captured block for this ref ID
finalize_capture_block(capturing_ref_id, capturing_lines)
capturing_ref_id = None
capturing_lines = []
# If there is trailing content (e.g. '2:{"curr":"$fa"...}')
if suffix.strip():
# Process suffix as a new "line" in the same iteration
line = suffix
else:
# Nothing more on this line
continue
else:
# Still inside captured block; skip normal processing
continue
# Detect start of a <think> block assigned to a ref like 'fa:...<think>'
if b"<think>" in line:
m = line_pattern.match(line)
if m:
capturing_ref_id = m.group(1).decode()
capturing_lines = [line]
# Skip normal parsing; the rest of the block will be captured until </yapp>
continue
match = line_pattern.match(line)
if not match:
continue
chunk_id, chunk_data = match.groups()
chunk_id = chunk_id.decode()
if nudge_new_chat_id and chunk_id == nudge_new_chat_id:
nudge_new_chat = chunk_data.decode()
continue
try:
data = json.loads(chunk_data) if chunk_data != b"{}" else {}
except json.JSONDecodeError:
@ -550,7 +644,7 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
if chunk_id == reward_id and isinstance(data, dict) and "unclaimedRewardInfo" in data:
reward_info = data
log_debug(f"Found reward info")
# Process initial setup
elif chunk_id == "1":
yield PlainTextResponse(line.decode(errors="ignore"))
@ -558,7 +652,8 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
left_stream = data.get("leftStream", {})
right_stream = data.get("rightStream", {})
if data.get("quickResponse", {}) != "$undefined":
quick_response_id = extract_ref_id(data.get("quickResponse", {}).get("stream", {}).get("next"))
quick_response_id = extract_ref_id(
data.get("quickResponse", {}).get("stream", {}).get("next"))
if data.get("turnId", {}) != "$undefined":
turn_id = extract_ref_id(data.get("turnId", {}).get("next"))
@ -592,7 +687,7 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
provider_info["variantUrl"] = selection.get("externalUrl")
log_debug(f"Found variant stream ID: {variant_stream_id}")
yield ProviderInfo.from_dict(provider_info)
# Process target stream content
elif target_stream_id and chunk_id == target_stream_id:
yield PlainTextResponse(line.decode(errors="ignore"))
@ -600,15 +695,41 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
target_stream_id = extract_ref_id(data.get("next"))
content = data.get("curr", "")
if content:
async for chunk in process_content_chunk(
content,
chunk_id,
line_count,
for_target=True
):
stream["target"].append(chunk)
is_started = True
yield chunk
# Handle special "$fa" / "$<id>" reference
ref_name = extract_ref_name(content)
if ref_name and (ref_name in think_blocks or ref_name in image_blocks):
# Thinking block
if ref_name in think_blocks:
t_text = think_blocks[ref_name]
if t_text:
reasoning = Reasoning(t_text)
# thinking_content += t_text
stream["thinking"].append(reasoning)
# yield reasoning
# Image-gen block
if ref_name in image_blocks:
img_block_text = image_blocks[ref_name]
async for chunk in process_content_chunk(
img_block_text,
ref_name,
line_count,
for_target=True
):
stream["target"].append(chunk)
is_started = True
yield chunk
else:
# Normal textual chunk
async for chunk in process_content_chunk(
content,
chunk_id,
line_count,
for_target=True
):
stream["target"].append(chunk)
is_started = True
yield chunk
# Variant stream (comparison)
elif variant_stream_id and chunk_id == variant_stream_id:
yield PlainTextResponse("[Variant] " + line.decode(errors="ignore"))
@ -651,8 +772,6 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
...
elif chunk_id == left_message_id:
...
elif chunk_id == nudge_new_chat_id:
nudge_new_chat = data
# Miscellaneous extra content
elif isinstance(data, dict) and "curr" in data:
content = data.get("curr", "")
@ -664,24 +783,23 @@ class Yupp(AsyncGeneratorProvider, ProviderModelMixin):
for_target=False
):
stream["extra"].append(chunk)
if isinstance(chunk,str) and "<streaming stopped unexpectedly" in chunk:
if isinstance(chunk, str) and "<streaming stopped unexpectedly" in chunk:
yield FinishReason(chunk)
yield PlainTextResponse("[Extra] " + line.decode(errors="ignore"))
if variant_image is not None:
yield variant_image
elif variant_text:
yield PreviewResponse(variant_text)
yield JsonResponse(**stream)
log_debug(f"Finished processing {line_count} lines")
except:
raise
finally:
# Claim reward in background
if reward_info and "unclaimedRewardInfo" in reward_info:
reward_id = reward_info["unclaimedRewardInfo"].get("rewardId")
if reward_id:
await claim_yupp_reward(session, account, reward_id)
await claim_yupp_reward(session, account, reward_id)

2
g4f/Provider/needs_auth/LMArena.py
View file

@ -642,7 +642,7 @@ class LMArena(AsyncGeneratorProvider, ProviderModelMixin, AuthFileMixin):
if not cls._models_loaded:
cls.get_models()
is_image_model = model in image_models
is_image_model = model in cls.image_models
if not model:
model = cls.default_model
if model in cls.model_aliases:

405
g4f/Provider/needs_auth/OpenaiChat.py
View file

@ -1,18 +1,20 @@
from __future__ import annotations
import os
import re
import asyncio
import uuid
import json
import base64
import time
import hashlib
import json
import os
import random
from typing import AsyncIterator, Iterator, Optional, Generator, Dict, Union, List, Any
import re
import time
import uuid
from copy import copy
from typing import AsyncIterator, Iterator, Optional, Generator, Dict, Union, List, Any
try:
import nodriver
has_nodriver = True
except ImportError:
has_nodriver = False
@ -22,15 +24,17 @@ from ...typing import AsyncResult, Messages, Cookies, MediaListType
from ...requests.raise_for_status import raise_for_status
from ...requests import StreamSession
from ...requests import get_nodriver_session
from ...image import ImageRequest, to_image, to_bytes, is_accepted_format, detect_file_type
from ...image import ImageRequest, to_image, to_bytes, detect_file_type
from ...errors import MissingAuthError, NoValidHarFileError, ModelNotFoundError
from ...providers.response import JsonConversation, FinishReason, SynthesizeData, AuthResult, ImageResponse, ImagePreview, ResponseType, JsonRequest, format_link
from ...providers.response import JsonConversation, FinishReason, SynthesizeData, AuthResult, ImageResponse, \
ImagePreview, ResponseType, JsonRequest, format_link
from ...providers.response import TitleGeneration, RequestLogin, Reasoning
from ...tools.media import merge_media
from ..helper import format_cookies, format_media_prompt, to_string
from ..openai.models import default_model, default_image_model, models, image_models, text_models, model_aliases
from ..openai.har_file import get_request_config
from ..openai.har_file import RequestConfig, arkReq, arkose_url, start_url, conversation_url, backend_url, prepare_url, backend_anon_url
from ..openai.har_file import RequestConfig, arkReq, arkose_url, start_url, conversation_url, backend_url, prepare_url, \
backend_anon_url
from ..openai.proofofwork import generate_proof_token
from ..openai.new import get_requirements_token, get_config
from ... import debug
@ -87,6 +91,9 @@ UPLOAD_HEADERS = {
"user-agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36"
}
ImagesCache: Dict[str, dict] = {}
class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
"""A class for creating and managing conversations with OpenAI chat service"""
@ -95,6 +102,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
working = True
active_by_default = True
use_nodriver = True
image_cache = True
supports_gpt_4 = True
supports_message_history = True
supports_system_message = True
@ -127,11 +135,11 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
@classmethod
async def upload_files(
cls,
session: StreamSession,
auth_result: AuthResult,
media: MediaListType,
) -> list[ImageRequest]:
cls,
session: StreamSession,
auth_result: AuthResult,
media: MediaListType,
) -> List[ImageRequest]:
"""
Upload an image to the service and get the download URL
@ -143,11 +151,20 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
Returns:
An ImageRequest object that contains the download URL, file name, and other data
"""
async def upload_file(file, image_name=None):
async def upload_file(file, image_name=None) -> ImageRequest:
debug.log(f"Uploading file: {image_name}")
file_data = {}
data_bytes = to_bytes(file)
# Check Cache
hasher = hashlib.md5()
hasher.update(data_bytes)
image_hash = hasher.hexdigest()
cache_file = ImagesCache.get(image_hash)
if cls.image_cache and file:
debug.log("Using cached image")
return ImageRequest(cache_file)
extension, mime_type = detect_file_type(data_bytes)
if "image" in mime_type:
# Convert the image to a PIL Image object
@ -181,30 +198,31 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
# Put the image bytes to the upload URL and check the status
await asyncio.sleep(1)
async with session.put(
file_data["upload_url"],
data=data_bytes,
headers={
**UPLOAD_HEADERS,
"Content-Type": file_data["mime_type"],
"x-ms-blob-type": "BlockBlob",
"x-ms-version": "2020-04-08",
"Origin": "https://chatgpt.com",
}
file_data["upload_url"],
data=data_bytes,
headers={
**UPLOAD_HEADERS,
"Content-Type": file_data["mime_type"],
"x-ms-blob-type": "BlockBlob",
"x-ms-version": "2020-04-08",
"Origin": "https://chatgpt.com",
}
) as response:
await raise_for_status(response)
# Post the file ID to the service and get the download URL
async with session.post(
f"{cls.url}/backend-api/files/{file_data['file_id']}/uploaded",
json={},
headers=auth_result.headers
f"{cls.url}/backend-api/files/{file_data['file_id']}/uploaded",
json={},
headers=auth_result.headers
) as response:
cls._update_request_args(auth_result, session)
await raise_for_status(response, "Get download url failed")
uploaded_data = await response.json()
file_data["download_url"] = uploaded_data["download_url"]
ImagesCache[image_hash] = file_data.copy()
return ImageRequest(file_data)
medias = []
medias: List["ImageRequest"] = []
for item in media:
item = item if isinstance(item, tuple) else (item,)
__uploaded_media = await upload_file(*item)
@ -242,7 +260,8 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
"id": str(uuid.uuid4()),
"author": {"role": message["role"]},
"content": {"content_type": "text", "parts": [to_string(message["content"])]},
"metadata": {"serialization_metadata": {"custom_symbol_offsets": []}, **({"system_hints": system_hints} if system_hints else {})},
"metadata": {"serialization_metadata": {"custom_symbol_offsets": []},
**({"system_hints": system_hints} if system_hints else {})},
"create_time": time.time(),
} for message in messages]
# Check if there is an image response
@ -256,11 +275,11 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
"size_bytes": image_request.get("file_size"),
"width": image_request.get("width"),
}
for image_request in image_requests
for image_request in image_requests
# Add For Images Only
if image_request.get("use_case") == "multimodal"
],
messages[-1]["content"]["parts"][0]]
messages[-1]["content"]["parts"][0]]
}
# Add the metadata object with the attachments
messages[-1]["metadata"] = {
@ -278,12 +297,14 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
else {}
),
}
for image_request in image_requests]
for image_request in image_requests]
}
return messages
@classmethod
async def get_generated_image(cls, session: StreamSession, auth_result: AuthResult, element: Union[dict, str], prompt: str = None, conversation_id: str = None) -> ImagePreview|ImageResponse|None:
async def get_generated_image(cls, session: StreamSession, auth_result: AuthResult, element: Union[dict, str],
prompt: str = None, conversation_id: str = None,
status: Optional[str] = None) -> ImagePreview | ImageResponse | None:
download_urls = []
is_sediment = False
if prompt is None:
@ -292,7 +313,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
except KeyError:
pass
if "asset_pointer" in element:
element = element["asset_pointer"]
element = element["asset_pointer"]
if isinstance(element, str) and element.startswith("file-service://"):
element = element.split("file-service://", 1)[-1]
elif isinstance(element, str) and element.startswith("sediment://"):
@ -303,7 +324,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if is_sediment:
url = f"{cls.url}/backend-api/conversation/{conversation_id}/attachment/{element}/download"
else:
url =f"{cls.url}/backend-api/files/{element}/download"
url = f"{cls.url}/backend-api/files/{element}/download"
try:
async with session.get(url, headers=auth_result.headers) as response:
cls._update_request_args(auth_result, session)
@ -319,27 +340,31 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
debug.error("OpenaiChat: Download image failed")
debug.error(e)
if download_urls:
return ImagePreview(download_urls, prompt, {"headers": auth_result.headers}) if is_sediment else ImageResponse(download_urls, prompt, {"headers": auth_result.headers})
# status = None, finished_successfully
if is_sediment and status is None:
return ImagePreview(download_urls, prompt, {"status": status, "headers": auth_result.headers})
else:
return ImageResponse(download_urls, prompt, {"status": status, "headers": auth_result.headers})
@classmethod
async def create_authed(
cls,
model: str,
messages: Messages,
auth_result: AuthResult,
proxy: str = None,
timeout: int = 360,
auto_continue: bool = False,
action: Optional[str] = None,
conversation: Conversation = None,
media: MediaListType = None,
return_conversation: bool = True,
web_search: bool = False,
prompt: str = None,
conversation_mode: Optional[dict] = None,
temporary: Optional[bool] = None,
conversation_id: Optional[str] = None,
**kwargs
cls,
model: str,
messages: Messages,
auth_result: AuthResult,
proxy: str = None,
timeout: int = 360,
auto_continue: bool = False,
action: Optional[str] = None,
conversation: Conversation = None,
media: MediaListType = None,
return_conversation: bool = True,
web_search: bool = False,
prompt: str = None,
conversation_mode: Optional[dict] = None,
temporary: Optional[bool] = None,
conversation_id: Optional[str] = None,
**kwargs
) -> AsyncResult:
"""
Create an asynchronous generator for the conversation.
@ -367,12 +392,12 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if action is None:
action = "next"
async with StreamSession(
proxy=proxy,
impersonate="chrome",
timeout=timeout
proxy=proxy,
impersonate="chrome",
timeout=timeout
) as session:
image_requests = None
media = merge_media(media, messages)
media = merge_media(media, messages)
if not cls.needs_auth and not media:
if cls._headers is None:
cls._create_request_args(cls._cookies)
@ -436,18 +461,19 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if temporary:
data["history_and_training_disabled"] = True
async with session.post(
prepare_url,
json=data,
headers=cls._headers
prepare_url,
json=data,
headers=cls._headers
) as response:
await raise_for_status(response)
conduit_token = (await response.json())["conduit_token"]
async with session.post(
f"{cls.url}/backend-anon/sentinel/chat-requirements"
if cls._api_key is None else
f"{cls.url}/backend-api/sentinel/chat-requirements",
json={"p": None if not getattr(auth_result, "proof_token", None) else get_requirements_token(getattr(auth_result, "proof_token", None))},
headers=cls._headers
f"{cls.url}/backend-anon/sentinel/chat-requirements"
if cls._api_key is None else
f"{cls.url}/backend-api/sentinel/chat-requirements",
json={"p": None if not getattr(auth_result, "proof_token", None) else get_requirements_token(
getattr(auth_result, "proof_token", None))},
headers=cls._headers
) as response:
if response.status in (401, 403):
raise MissingAuthError(f"Response status: {response.status}")
@ -456,10 +482,10 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
await raise_for_status(response)
chat_requirements = await response.json()
need_turnstile = chat_requirements.get("turnstile", {}).get("required", False)
need_arkose = chat_requirements.get("arkose", {}).get("required", False)
chat_token = chat_requirements.get("token")
need_arkose = chat_requirements.get("arkose", {}).get("required", False)
chat_token = chat_requirements.get("token")
# if need_arkose and cls.request_config.arkose_token is None:
# if need_arkose and cls.request_config.arkose_token is None:
# await get_request_config(proxy)
# cls._create_request_args(auth_result.cookies, auth_result.headers)
# cls._set_api_key(auth_result.access_token)
@ -476,23 +502,25 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
proof_token=proof_token
)
# [debug.log(text) for text in (
#f"Arkose: {'False' if not need_arkose else auth_result.arkose_token[:12]+'...'}",
#f"Proofofwork: {'False' if proofofwork is None else proofofwork[:12]+'...'}",
#f"AccessToken: {'False' if cls._api_key is None else cls._api_key[:12]+'...'}",
# f"Arkose: {'False' if not need_arkose else auth_result.arkose_token[:12]+'...'}",
# f"Proofofwork: {'False' if proofofwork is None else proofofwork[:12]+'...'}",
# f"AccessToken: {'False' if cls._api_key is None else cls._api_key[:12]+'...'}",
# )]
data = {
"action": "next",
"parent_message_id": conversation.message_id,
"model": model,
"timezone_offset_min":-120,
"timezone":"Europe/Berlin",
"conversation_mode":{"kind":"primary_assistant"},
"enable_message_followups":True,
"timezone_offset_min": -120,
"timezone": "Europe/Berlin",
"conversation_mode": {"kind": "primary_assistant"},
"enable_message_followups": True,
"system_hints": ["search"] if web_search else None,
"supports_buffering":True,
"supported_encodings":["v1"],
"client_contextual_info":{"is_dark_mode":False,"time_since_loaded":random.randint(20, 500),"page_height":578,"page_width":1850,"pixel_ratio":1,"screen_height":1080,"screen_width":1920},
"paragen_cot_summary_display_override":"allow"
"supports_buffering": True,
"supported_encodings": ["v1"],
"client_contextual_info": {"is_dark_mode": False, "time_since_loaded": random.randint(20, 500),
"page_height": 578, "page_width": 1850, "pixel_ratio": 1,
"screen_height": 1080, "screen_width": 1920},
"paragen_cot_summary_display_override": "allow"
}
if temporary:
data["history_and_training_disabled"] = True
@ -512,7 +540,8 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
new_messages = []
else:
new_messages.append(message)
data["messages"] = cls.create_messages(new_messages, image_requests, ["search"] if web_search else None)
data["messages"] = cls.create_messages(new_messages, image_requests,
["search"] if web_search else None)
yield JsonRequest.from_dict(data)
headers = {
**cls._headers,
@ -521,18 +550,18 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
"openai-sentinel-chat-requirements-token": chat_token,
**({} if conduit_token is None else {"x-conduit-token": conduit_token})
}
#if cls.request_config.arkose_token:
# if cls.request_config.arkose_token:
# headers["openai-sentinel-arkose-token"] = cls.request_config.arkose_token
if proofofwork is not None:
headers["openai-sentinel-proof-token"] = proofofwork
if need_turnstile and getattr(auth_result, "turnstile_token", None) is not None:
headers['openai-sentinel-turnstile-token'] = auth_result.turnstile_token
async with session.post(
backend_anon_url
if cls._api_key is None else
backend_url,
json=data,
headers=headers
backend_anon_url
if cls._api_key is None else
backend_url,
json=data,
headers=headers
) as response:
cls._update_request_args(auth_result, session)
if response.status in (401, 403, 429, 500):
@ -548,10 +577,12 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if match.group(0) in matches:
continue
matches.append(match.group(0))
generated_image = await cls.get_generated_image(session, auth_result, match.group(0), prompt)
generated_image = await cls.get_generated_image(session, auth_result, match.group(0),
prompt)
if generated_image is not None:
yield generated_image
async for chunk in cls.iter_messages_line(session, auth_result, line, conversation, sources, references):
async for chunk in cls.iter_messages_line(session, auth_result, line, conversation, sources,
references):
if isinstance(chunk, str):
chunk = chunk.replace("\ue203", "").replace("\ue204", "").replace("\ue206", "")
buffer += chunk
@ -561,9 +592,9 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
def citation_replacer(match: re.Match[str]):
ref_type = match.group(1)
ref_index = int(match.group(2))
if ((ref_type == "image" and is_image_embedding) or
is_video_embedding or
ref_type == "forecast"):
if ((ref_type == "image" and is_image_embedding) or
is_video_embedding or
ref_type == "forecast"):
reference = references.get_reference({
"ref_index": ref_index,
@ -571,7 +602,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
})
if not reference:
return ""
if ref_type == "forecast":
if reference.get("alt"):
return reference.get("alt")
@ -580,11 +611,13 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if is_image_embedding and reference.get("content_url", ""):
return f"![{reference.get('title', '')}]({reference.get('content_url')})"
if is_video_embedding:
if reference.get("url", "") and reference.get("thumbnail_url", ""):
if reference.get("url", "") and reference.get("thumbnail_url",
""):
return f"[![{reference.get('title', '')}]({reference['thumbnail_url']})]({reference['url']})"
video_match = re.match(r"video\n(.*?)\nturn[0-9]+", match.group(0))
video_match = re.match(r"video\n(.*?)\nturn[0-9]+",
match.group(0))
if video_match:
return video_match.group(1)
return ""
@ -595,9 +628,9 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
})
if source_index is not None and len(sources.list) > source_index:
link = sources.list[source_index]["url"]
return f"[[{source_index+1}]]({link})"
return f"[[{source_index + 1}]]({link})"
return f""
def products_replacer(match: re.Match[str]):
try:
products_data = json.loads(match.group(1))
@ -612,25 +645,30 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
return ""
sequence_content = match.group(1)
sequence_content = sequence_content.replace("\ue200", "").replace("\ue202", "\n").replace("\ue201", "")
sequence_content = sequence_content.replace("\ue200", "").replace("\ue202",
"\n").replace(
"\ue201", "")
sequence_content = sequence_content.replace("navlist\n", "#### ")
# Handle search, news, view and image citations
is_image_embedding = sequence_content.startswith("i\nturn")
is_video_embedding = sequence_content.startswith("video\n")
sequence_content = re.sub(
r'(?:cite\nturn[0-9]+|forecast\nturn[0-9]+|video\n.*?\nturn[0-9]+|i?\n?turn[0-9]+)(search|news|view|image|forecast)(\d+)',
citation_replacer,
r'(?:cite\nturn[0-9]+|forecast\nturn[0-9]+|video\n.*?\nturn[0-9]+|i?\n?turn[0-9]+)(search|news|view|image|forecast)(\d+)',
citation_replacer,
sequence_content
)
sequence_content = re.sub(r'products\n(.*)', products_replacer, sequence_content)
sequence_content = re.sub(r'product_entity\n\[".*","(.*)"\]', lambda x: x.group(1), sequence_content)
sequence_content = re.sub(r'products\n(.*)', products_replacer,
sequence_content)
sequence_content = re.sub(r'product_entity\n\[".*","(.*)"\]',
lambda x: x.group(1), sequence_content)
return sequence_content
# process only completed sequences and do not touch start of next not completed sequence
buffer = re.sub(r'\ue200(.*?)\ue201', sequence_replacer, buffer, flags=re.DOTALL)
if buffer.find(u"\ue200") != -1: # still have uncompleted sequence
buffer = re.sub(r'\ue200(.*?)\ue201', sequence_replacer, buffer,
flags=re.DOTALL)
if buffer.find(u"\ue200") != -1: # still have uncompleted sequence
continue
else:
# do not yield to consume rest part of special sequence
@ -647,7 +685,8 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if sources.list:
yield sources
if conversation.generated_images:
yield ImageResponse(conversation.generated_images.urls, conversation.prompt, {"headers": auth_result.headers})
yield ImageResponse(conversation.generated_images.urls, conversation.prompt,
{"headers": auth_result.headers})
conversation.generated_images = None
conversation.prompt = None
if return_conversation:
@ -667,7 +706,9 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
yield FinishReason(conversation.finish_reason)
@classmethod
async def iter_messages_line(cls, session: StreamSession, auth_result: AuthResult, line: bytes, fields: Conversation, sources: OpenAISources, references: ContentReferences) -> AsyncIterator:
async def iter_messages_line(cls, session: StreamSession, auth_result: AuthResult, line: bytes,
fields: Conversation, sources: OpenAISources,
references: ContentReferences) -> AsyncIterator:
if not line.startswith(b"data: "):
return
elif line.startswith(b"data: [DONE]"):
@ -706,9 +747,14 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
elif m.get("p") == "/message/metadata/image_gen_title":
fields.prompt = m.get("v")
elif m.get("p") == "/message/content/parts/0/asset_pointer":
generated_images = fields.generated_images = await cls.get_generated_image(session, auth_result, m.get("v"), fields.prompt, fields.conversation_id)
status = next(filter(lambda x: x.get("p") == '/message/status', v), {}).get('v', None)
generated_images = fields.generated_images = await cls.get_generated_image(session, auth_result,
m.get("v"),
fields.prompt,
fields.conversation_id,
status)
if generated_images is not None:
if buffer:
if buffer:
yield buffer
yield generated_images
elif m.get("p") == "/message/metadata/search_result_groups":
@ -735,41 +781,48 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if match and m.get("o") == "append" and isinstance(m.get("v"), dict):
idx = int(match.group(1))
references.merge_reference(idx, m.get("v"))
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/fallback_items$", m.get("p")) and isinstance(m.get("v"), list):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/fallback_items$",
m.get("p")) and isinstance(m.get("v"), list):
for link in m.get("v", []) or []:
sources.add_source(link)
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/items$", m.get("p")) and isinstance(m.get("v"), list):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/items$",
m.get("p")) and isinstance(m.get("v"), list):
for link in m.get("v", []) or []:
sources.add_source(link)
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/refs$", m.get("p")) and isinstance(m.get("v"), list):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/refs$",
m.get("p")) and isinstance(m.get("v"), list):
match = re.match(r"^/message/metadata/content_references/(\d+)/refs$", m.get("p"))
if match:
idx = int(match.group(1))
references.update_reference(idx, m.get("o"), "refs", m.get("v"))
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/alt$", m.get("p")) and isinstance(m.get("v"), list):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/alt$",
m.get("p")) and isinstance(m.get("v"), list):
match = re.match(r"^/message/metadata/content_references/(\d+)/alt$", m.get("p"))
if match:
idx = int(match.group(1))
references.update_reference(idx, m.get("o"), "alt", m.get("v"))
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/prompt_text$", m.get("p")) and isinstance(m.get("v"), list):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/prompt_text$",
m.get("p")) and isinstance(m.get("v"), list):
match = re.match(r"^/message/metadata/content_references/(\d+)/prompt_text$", m.get("p"))
if match:
idx = int(match.group(1))
references.update_reference(idx, m.get("o"), "prompt_text", m.get("v"))
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/refs/\d+$", m.get("p")) and isinstance(m.get("v"), dict):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/refs/\d+$",
m.get("p")) and isinstance(m.get("v"), dict):
match = re.match(r"^/message/metadata/content_references/(\d+)/refs/(\d+)$", m.get("p"))
if match:
reference_idx = int(match.group(1))
ref_idx = int(match.group(2))
references.update_reference(reference_idx, m.get("o"), "refs", m.get("v"), ref_idx)
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/images$", m.get("p")) and isinstance(m.get("v"), list):
elif m.get("p") and re.match(r"^/message/metadata/content_references/\d+/images$",
m.get("p")) and isinstance(m.get("v"), list):
match = re.match(r"^/message/metadata/content_references/(\d+)/images$", m.get("p"))
if match:
idx = int(match.group(1))
references.update_reference(idx, m.get("o"), "images", m.get("v"))
elif m.get("p") == "/message/metadata/finished_text":
fields.is_thinking = False
if buffer:
if buffer:
yield buffer
yield Reasoning(status=m.get("v"))
elif m.get("p") == "/message/metadata" and fields.recipient == "all":
@ -785,10 +838,11 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
fields.recipient = m.get("recipient", fields.recipient)
if fields.recipient == "all":
c = m.get("content", {})
if c.get("content_type") == "text" and m.get("author", {}).get("role") == "tool" and "initial_text" in m.get("metadata", {}):
if c.get("content_type") == "text" and m.get("author", {}).get(
"role") == "tool" and "initial_text" in m.get("metadata", {}):
fields.is_thinking = True
yield Reasoning(status=m.get("metadata", {}).get("initial_text"))
#if c.get("content_type") == "multimodal_text":
# if c.get("content_type") == "multimodal_text":
# for part in c.get("parts"):
# if isinstance(part, dict) and part.get("content_type") == "image_asset_pointer":
# yield await cls.get_generated_image(session, auth_result, part, fields.prompt, fields.conversation_id)
@ -803,13 +857,13 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
@classmethod
async def synthesize(cls, params: dict) -> AsyncIterator[bytes]:
async with StreamSession(
impersonate="chrome",
timeout=0
impersonate="chrome",
timeout=0
) as session:
async with session.get(
f"{cls.url}/backend-api/synthesize",
params=params,
headers=cls._headers
f"{cls.url}/backend-api/synthesize",
params=params,
headers=cls._headers
) as response:
await raise_for_status(response)
async for chunk in response.iter_content():
@ -817,15 +871,15 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
@classmethod
async def login(
cls,
proxy: str = None,
api_key: str = None,
proof_token: str = None,
cookies: Cookies = None,
headers: dict = None,
**kwargs
cls,
proxy: str = None,
api_key: str = None,
proof_token: str = None,
cookies: Cookies = None,
headers: dict = None,
**kwargs
) -> AsyncIterator:
if cls._expires is not None and (cls._expires - 60*10) < time.time():
if cls._expires is not None and (cls._expires - 60 * 10) < time.time():
cls._headers = cls._api_key = None
if cls._headers is None or headers is not None:
cls._headers = {} if headers is None else headers
@ -858,6 +912,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
async def nodriver_auth(cls, proxy: str = None):
async with get_nodriver_session(proxy=proxy) as browser:
page = await browser.get(cls.url)
def on_request(event: nodriver.cdp.network.RequestWillBeSent, page=None):
if event.request.url == start_url or event.request.url.startswith(conversation_url):
if cls.request_config.headers is None:
@ -866,9 +921,10 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
cls.request_config.headers[key.lower()] = value
elif event.request.url in (backend_url, backend_anon_url):
if "OpenAI-Sentinel-Proof-Token" in event.request.headers:
cls.request_config.proof_token = json.loads(base64.b64decode(
event.request.headers["OpenAI-Sentinel-Proof-Token"].split("gAAAAAB", 1)[-1].split("~")[0].encode()
).decode())
cls.request_config.proof_token = json.loads(base64.b64decode(
event.request.headers["OpenAI-Sentinel-Proof-Token"].split("gAAAAAB", 1)[-1].split("~")[
0].encode()
).decode())
if "OpenAI-Sentinel-Turnstile-Token" in event.request.headers:
cls.request_config.turnstile_token = event.request.headers["OpenAI-Sentinel-Turnstile-Token"]
if "Authorization" in event.request.headers:
@ -881,6 +937,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
arkBody=event.request.post_data,
userAgent=event.request.headers.get("User-Agent")
)
await page.send(nodriver.cdp.network.enable())
page.add_handler(nodriver.cdp.network.RequestWillBeSent, on_request)
await page.reload()
@ -912,7 +969,7 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if cls._api_key is not None or not cls.needs_auth:
break
await asyncio.sleep(1)
debug.log(f"OpenaiChat: Access token: {'False' if cls._api_key is None else cls._api_key[:12]+'...'}")
debug.log(f"OpenaiChat: Access token: {'False' if cls._api_key is None else cls._api_key[:12] + '...'}")
while True:
if cls.request_config.proof_token:
break
@ -970,11 +1027,14 @@ class OpenaiChat(AsyncAuthedProvider, ProviderModelMixin):
if cls._cookies:
cls._headers["cookie"] = format_cookies(cls._cookies)
class Conversation(JsonConversation):
"""
Class to encapsulate response fields.
"""
def __init__(self, conversation_id: str = None, message_id: str = None, user_id: str = None, finish_reason: str = None, parent_message_id: str = None, is_thinking: bool = False):
def __init__(self, conversation_id: str = None, message_id: str = None, user_id: str = None,
finish_reason: str = None, parent_message_id: str = None, is_thinking: bool = False):
self.conversation_id = conversation_id
self.message_id = message_id
self.finish_reason = finish_reason
@ -987,8 +1047,9 @@ class Conversation(JsonConversation):
self.prompt = None
self.generated_images: ImagePreview = None
def get_cookies(
urls: Optional[Iterator[str]] = None
urls: Optional[Iterator[str]] = None
) -> Generator[Dict, Dict, Dict[str, str]]:
params = {}
if urls is not None:
@ -1000,6 +1061,7 @@ def get_cookies(
json = yield cmd_dict
return {c["name"]: c["value"] for c in json['cookies']} if 'cookies' in json else {}
class OpenAISources(ResponseType):
list: List[Dict[str, str]]
@ -1025,7 +1087,7 @@ class OpenAISources(ResponseType):
if existing_source and idx is not None:
self.list[idx] = source
return
existing_source, idx = self.find_by_url(source["url"])
if existing_source and idx is not None:
self.list[idx] = source
@ -1038,53 +1100,54 @@ class OpenAISources(ResponseType):
if not self.list:
return ""
return "\n\n\n\n" + ("\n>\n".join([
f"> [{idx+1}] {format_link(link['url'], link.get('title', ''))}"
f"> [{idx + 1}] {format_link(link['url'], link.get('title', ''))}"
for idx, link in enumerate(self.list)
]))
def get_ref_info(self, source: Dict[str, str]) -> dict[str, str|int] | None:
def get_ref_info(self, source: Dict[str, str]) -> dict[str, str | int] | None:
ref_index = source.get("ref_id", {}).get("ref_index", None)
ref_type = source.get("ref_id", {}).get("ref_type", None)
if isinstance(ref_index, int):
return {
"ref_index": ref_index,
"ref_index": ref_index,
"ref_type": ref_type,
}
for ref_info in source.get('refs') or []:
ref_index = ref_info.get("ref_index", None)
ref_type = ref_info.get("ref_type", None)
if isinstance(ref_index, int):
return {
"ref_index": ref_index,
"ref_index": ref_index,
"ref_type": ref_type,
}
return None
def find_by_ref_info(self, ref_info: dict[str, str|int]):
def find_by_ref_info(self, ref_info: dict[str, str | int]):
for idx, source in enumerate(self.list):
source_ref_info = self.get_ref_info(source)
if (source_ref_info and
source_ref_info["ref_index"] == ref_info["ref_index"] and
source_ref_info["ref_type"] == ref_info["ref_type"]):
return source, idx
if (source_ref_info and
source_ref_info["ref_index"] == ref_info["ref_index"] and
source_ref_info["ref_type"] == ref_info["ref_type"]):
return source, idx
return None, None
def find_by_url(self, url: str):
for idx, source in enumerate(self.list):
if source["url"] == url:
return source, idx
return None, None
return None, None
def get_index(self, ref_info: dict[str, str|int]) -> int | None:
def get_index(self, ref_info: dict[str, str | int]) -> int | None:
_, index = self.find_by_ref_info(ref_info)
if index is not None:
return index
return index
return None
class ContentReferences:
def __init__(self) -> None:
self.list: List[Dict[str, Any]] = []
@ -1098,16 +1161,16 @@ class ContentReferences:
self.list[idx] = {**self.list[idx], **reference_part}
def update_reference(self, idx: int, operation: str, field: str, value: Any, ref_idx = None) -> None:
def update_reference(self, idx: int, operation: str, field: str, value: Any, ref_idx=None) -> None:
while len(self.list) <= idx:
self.list.append({})
if operation == "append" or operation == "add":
if not isinstance(self.list[idx].get(field, None), list):
self.list[idx][field] = []
if isinstance(value, list):
self.list[idx][field].extend(value)
else:
else:
self.list[idx][field].append(value)
if operation == "replace" and ref_idx is not None:
@ -1123,10 +1186,10 @@ class ContentReferences:
self.list[idx][field] = value
def get_ref_info(
self,
source: Dict[str, str],
target_ref_info: Dict[str, Union[str, int]]
) -> dict[str, str|int] | None:
self,
source: Dict[str, str],
target_ref_info: Dict[str, Union[str, int]]
) -> dict[str, str | int] | None:
for idx, ref_info in enumerate(source.get("refs", [])) or []:
if not isinstance(ref_info, dict):
continue
@ -1134,11 +1197,11 @@ class ContentReferences:
ref_index = ref_info.get("ref_index", None)
ref_type = ref_info.get("ref_type", None)
if isinstance(ref_index, int) and isinstance(ref_type, str):
if (not target_ref_info or
(target_ref_info["ref_index"] == ref_index and
target_ref_info["ref_type"] == ref_type)):
if (not target_ref_info or
(target_ref_info["ref_index"] == ref_index and
target_ref_info["ref_type"] == ref_type)):
return {
"ref_index": ref_index,
"ref_index": ref_index,
"ref_type": ref_type,
"idx": idx
}
@ -1149,9 +1212,9 @@ class ContentReferences:
for reference in self.list:
reference_ref_info = self.get_ref_info(reference, ref_info)
if (not reference_ref_info or
reference_ref_info["ref_index"] != ref_info["ref_index"] or
reference_ref_info["ref_type"] != ref_info["ref_type"]):
if (not reference_ref_info or
reference_ref_info["ref_index"] != ref_info["ref_index"] or
reference_ref_info["ref_type"] != ref_info["ref_type"]):
continue
if ref_info["ref_type"] != "image":

9
g4f/image/__init__.py
View file

@ -9,6 +9,8 @@ from pathlib import Path
from typing import Optional
from urllib.parse import urlparse
import requests
try:
from PIL import Image, ImageOps
has_requirements = True
@ -383,6 +385,13 @@ def to_bytes(image: ImageType) -> bytes:
return Path(path).read_bytes()
else:
raise FileNotFoundError(f"File not found: {path}")
else:
resp = requests.get(image, headers={
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/137.0.0.0 Safari/537.36 Edg/137.0.0.0",
})
if resp.ok and is_accepted_format(resp.content):
return resp.content
raise ValueError("Invalid image url. Expected bytes, str, or PIL Image.")
else:
raise ValueError("Invalid image format. Expected bytes, str, or PIL Image.")
elif isinstance(image, Image.Image):

12
g4f/requests/aiohttp.py
View file

@ -31,17 +31,21 @@ class StreamResponse(ClientResponse):
except json.JSONDecodeError:
continue
class StreamSession():
class StreamSession:
def __init__(
self,
headers: dict = {},
headers=None,
timeout: int = None,
connector: BaseConnector = None,
proxy: str = None,
proxies: dict = {},
proxies=None,
impersonate = None,
**kwargs
):
if proxies is None:
proxies = {}
if headers is None:
headers = {}
if impersonate:
headers = {
**DEFAULT_HEADERS,
@ -49,7 +53,7 @@ class StreamSession():
}
connect = None
if isinstance(timeout, tuple):
connect, timeout = timeout;
connect, timeout = timeout
if timeout is not None:
timeout = ClientTimeout(timeout, connect)
if proxy is None: