gpt4free

公开

关注 0 Fork 0 Star 1

XFEstudio/gpt4free

feat: improve file handling and streamline provider implementations

- Added file upload usage example with bucket_id in docs/file.md - Fixed ARTA provider by refactoring error handling with a new raise_error function - Simplified aspect_ratio handling in ARTA with proper default - Improved AllenAI provider by cleaning up image handling logic - Fixed Blackbox provider's media handling to properly process images - Updated file tools to handle URL downloads correctly - Fixed bucket_id pattern matching in ToolHandler.process_bucket_tool - Cleaned up imports in typing.py by removing unnecessary sys import - Fixed inconsistent function parameters in g4f/tools/files.py - Fixed return value of upload_and_process function to return bucket_id

c083f852

hlohaus <983577+hlohaus@users.noreply.github.com>

提交于 2025-04-08 19:00

父提交 fa36dcc

代码差异

7 个文件 +82 -104

Modified docs/file.md +30 -1

@@ -75,16 +75,45 @@ def upload_and_process(files_or_urls, bucket_id=None):

75 75 else:

76 76 print(f"Unhandled SSE event: {line}")

77 77 response.close()

78 return bucket_id5

78 79

79 80 # Example with URLs

80 81 urls = [{"url": "https://github.com/xtekky/gpt4free/issues"}]

81 82 bucket_id = upload_and_process(urls)

82 83

83 84 #Example with files

84 files = {'files': open('document.pdf', 'rb'), 'files': open('data.json', 'rb')}

85 files = {'files': ('document.pdf', open('document.pdf', 'rb'))}

85 86 bucket_id = upload_and_process(files)

86 87 ```

87 88

89 **Usage of Uploaded Files:**

90 ```python

91 from g4f.client import Client

93 # Enable debug mode

94 import g4f.debug

95 g4f.debug.logging = True

97 client = Client()

99 # Upload example file

100 files = {'files': ('demo.docx', open('demo.docx', 'rb'))}

101 bucket_id = upload_and_process(files)

102

103 # Send request with file:

104 response = client.chat.completions.create(

105 [{"role": "user", "content": [

106 {"type": "text", "text": "Discribe this file."},

107 {"bucket_id": bucket_id}

108 ]}],

109 )

110 print(response.choices[0].message.content)

111 ```

112

113 **Example Output:**

114 ```

115 This document is a demonstration of the DOCX Input plugin capabilities in the software ...

116 ```

88 117

89 118 **Example Usage (JavaScript):**

90 119

Modified g4f/Provider/ARTA.py +20 -50

@@ -5,7 +5,7 @@ import time

5 5 import json

6 6 import random

7 7 from pathlib import Path

8 from aiohttp import ClientSession

8 from aiohttp import ClientSession, ClientResponse

9 9 import asyncio

10 10

11 11 from ..typing import AsyncResult, Messages

@@ -92,17 +92,8 @@ class ARTA(AsyncGeneratorProvider, ProviderModelMixin):

92 92 # Step 1: Generate Authentication Token

93 93 auth_payload = {"clientType": "CLIENT_TYPE_ANDROID"}

94 94 async with session.post(cls.auth_url, json=auth_payload, proxy=proxy) as auth_response:

95 if auth_response.status >= 400:

96 error_text = await auth_response.text()

                    raise ResponseError(f"Failed to obtain authentication token. Status: {auth_response.status}, Response: {error_text}")

99 try:

100 auth_data = await auth_response.json()

101 except Exception as e:

102 error_text = await auth_response.text()

103 content_type = auth_response.headers.get('Content-Type', 'unknown')

104

                    raise ResponseError(f"Failed to parse auth response as JSON. Content-Type: {content_type}, Error: {str(e)}, Response: {error_text}")

105

95 await raise_error(f"Failed to obtain authentication token", auth_response)

96 auth_data = await auth_response.json()

106 97 auth_token = auth_data.get("idToken")

107 98 #refresh_token = auth_data.get("refreshToken")

108 99 if not auth_token:

@@ -118,17 +109,8 @@ class ARTA(AsyncGeneratorProvider, ProviderModelMixin):

118 109 "refresh_token": refresh_token,

119 110 }

120 111 async with session.post(cls.token_refresh_url, data=payload, proxy=proxy) as response:

121 if response.status >= 400:

122 error_text = await response.text()

123 raise ResponseError(f"Failed to refresh token. Status: {response.status}, Response: {error_text}")

124

125 try:

126 response_data = await response.json()

127 except Exception as e:

128 error_text = await response.text()

129 content_type = response.headers.get('Content-Type', 'unknown')

130

                    raise ResponseError(f"Failed to parse token refresh response as JSON. Content-Type: {content_type}, Error: {str(e)}, Response: {error_text}")

131

112 await raise_error(f"Failed to refresh token", response)

113 response_data = await response.json()

132 114 return response_data.get("id_token"), response_data.get("refresh_token")

133 115

134 116 @classmethod

@@ -156,7 +138,7 @@ class ARTA(AsyncGeneratorProvider, ProviderModelMixin):

156 138 n: int = 1,

157 139 guidance_scale: int = 7,

158 140 num_inference_steps: int = 30,

159 aspect_ratio: str = "1:1",

141 aspect_ratio: str = None,

160 142 seed: int = None,

161 143 **kwargs

162 144 ) -> AsyncResult:

@@ -179,7 +161,7 @@ class ARTA(AsyncGeneratorProvider, ProviderModelMixin):

179 161 "images_num": str(n),

180 162 "cfg_scale": str(guidance_scale),

181 163 "steps": str(num_inference_steps),

182 "aspect_ratio": aspect_ratio,

164 "aspect_ratio": "1:1" if aspect_ratio is None else aspect_ratio,

183 165 "seed": str(seed),

184 166 }

185 167

@@ -188,45 +170,26 @@ class ARTA(AsyncGeneratorProvider, ProviderModelMixin):

188 170 }

189 171

190 172

            async with session.post(cls.image_generation_url, data=image_payload, headers=headers, proxy=proxy) as image_response:

191 if image_response.status >= 400:

192 error_text = await image_response.text()

193

                    raise ResponseError(f"Failed to initiate image generation. Status: {image_response.status}, Response: {error_text}")

194

195 try:

196 image_data = await image_response.json()

197 except Exception as e:

198 error_text = await image_response.text()

199 content_type = image_response.headers.get('Content-Type', 'unknown')

200

                    raise ResponseError(f"Failed to parse response as JSON. Content-Type: {content_type}, Error: {str(e)}, Response: {error_text}")

201

173 await raise_error(f"Failed to initiate image generation", image_response)

174 image_data = await image_response.json()

202 175 record_id = image_data.get("record_id")

203 176 if not record_id:

204 177 raise ResponseError(f"Failed to initiate image generation: {image_data}")

205 178

206 179 # Step 3: Check Generation Status

207 180 status_url = cls.status_check_url.format(record_id=record_id)

208 counter = 4

209 181 start_time = time.time()

210 182 last_status = None

211 183 while True:

212 184 async with session.get(status_url, headers=headers, proxy=proxy) as status_response:

213 if status_response.status >= 400:

214 error_text = await status_response.text()

215

                        raise ResponseError(f"Failed to check image generation status. Status: {status_response.status}, Response: {error_text}")

216

217 try:

218 status_data = await status_response.json()

219 except Exception as e:

220 error_text = await status_response.text()

221 content_type = status_response.headers.get('Content-Type', 'unknown')

222

                        raise ResponseError(f"Failed to parse status response as JSON. Content-Type: {content_type}, Error: {str(e)}, Response: {error_text}")

223

185 await raise_error(f"Failed to check image generation status", status_response)

186 status_data = await status_response.json()

224 187 status = status_data.get("status")

225 188

226 189 if status == "DONE":

227 190 image_urls = [image["url"] for image in status_data.get("response", [])]

228 191 duration = time.time() - start_time

229 yield Reasoning(label="Generated", status=f"{n} image(s) in {duration:.2f}s")

192

                        yield Reasoning(label="Generated", status=f"{n} image in {duration:.2f}s" if n == 1 else f"{n} images in {duration:.2f}s")

230 193 yield ImageResponse(urls=image_urls, alt=prompt)

231 194 return

232 195 elif status in ("IN_QUEUE", "IN_PROGRESS"):

@@ -238,4 +201,11 @@ class ARTA(AsyncGeneratorProvider, ProviderModelMixin):

238 201 yield Reasoning(label="Generating")

239 202 await asyncio.sleep(2) # Poll every 2 seconds

240 203 else:

241 raise ResponseError(f"Image generation failed with status: {status}")

204 raise ResponseError(f"Image generation failed with status: {status}")

205

206 async def raise_error(response: ClientResponse, message: str):

207 if response.ok:

208 return

209 error_text = await response.text()

210 content_type = response.headers.get('Content-Type', 'unknown')

211 raise ResponseError(f"{message}. Content-Type: {content_type}, Response: {error_text}")

Modified g4f/Provider/AllenAI.py +12 -28

@@ -83,17 +83,7 @@ class AllenAI(AsyncGeneratorProvider, ProviderModelMixin):

83 83 ) -> AsyncResult:

84 84 actual_model = cls.get_model(model)

85 85

86 # Use format_image_prompt for vision models when media is provided

87 if media is not None and len(media) > 0:

88 # For vision models, use format_image_prompt

89 if actual_model in cls.vision_models:

90 prompt = format_image_prompt(messages)

91 else:

92 # For non-vision models with images, still use the last user message

93 prompt = get_last_user_message(messages)

94 else:

95 # For text-only messages, use the standard format

96 prompt = format_prompt(messages) if conversation is None else get_last_user_message(messages)

86 prompt = format_prompt(messages) if conversation is None else get_last_user_message(messages)

97 87

98 88 # Determine the correct host for the model

99 89 if host is None:

@@ -157,18 +147,16 @@ class AllenAI(AsyncGeneratorProvider, ProviderModelMixin):

157 147 if media is not None and len(media) > 0:

158 148 conversation = Conversation(actual_model)

159 149

160 # Add image if provided

161 if media is not None and len(media) > 0:

162 # For each image in the media list (using merge_media to handle different formats)

163 for image, image_name in merge_media(media, messages):

164 image_bytes = to_bytes(image)

165 form_data.extend([

166 f'--{boundary}\r\n'

167 f'Content-Disposition: form-data; name="files"; filename="{image_name}"\r\n'

168 f'Content-Type: {is_accepted_format(image_bytes)}\r\n\r\n'

169 ])

170 form_data.append(image_bytes.decode('latin1'))

171 form_data.append('\r\n')

150 # For each image in the media list (using merge_media to handle different formats)

151 for image, image_name in merge_media(media, messages):

152 image_bytes = to_bytes(image)

153 form_data.extend([

154 f'--{boundary}\r\n'

155 f'Content-Disposition: form-data; name="files"; filename="{image_name}"\r\n'

156 f'Content-Type: {is_accepted_format(image_bytes)}\r\n\r\n'

157 ])

158 form_data.append(image_bytes.decode('latin1'))

159 form_data.append('\r\n')

172 160

173 161 form_data.append(f'--{boundary}--\r\n')

174 162 data = "".join(form_data).encode('latin1')

@@ -182,11 +170,7 @@ class AllenAI(AsyncGeneratorProvider, ProviderModelMixin):

182 170 await raise_for_status(response)

183 171 current_parent = None

184 172

185 async for chunk in response.content:

186 if not chunk:

187 continue

188 decoded = chunk.decode(errors="ignore")

189 for line in decoded.splitlines():

173 async for line in response.content:

190 174 line = line.strip()

191 175 if not line:

192 176 continue

Modified g4f/Provider/Blackbox.py +3 -2

@@ -581,14 +581,15 @@ class Blackbox(AsyncGeneratorProvider, ProviderModelMixin):

581 581 }

582 582 current_messages.append(current_msg)

583 583

584 if media is not None:

584 media = list(merge_media(media, messages))

585 if media:

585 586 current_messages[-1]['data'] = {

586 587 "imagesData": [

587 588 {

588 589 "filePath": f"/{image_name}",

589 590 "contents": to_data_uri(image)

590 591 }

591 for image, image_name in merge_media(media, messages)

592 for image, image_name in media

592 593 ],

593 594 "fileText": "",

594 595 "title": ""

Modified g4f/tools/files.py +6 -5

@@ -518,11 +518,12 @@ async def async_read_and_download_urls(bucket_dir: Path, delete_files: bool = Fa

518 518 if urls:

519 519 count = 0

520 520 with open(os.path.join(bucket_dir, FILE_LIST), 'a') as f:

521 async for filename in download_urls(bucket_dir, urls):

522 f.write(f"{filename}\n")

523 if event_stream:

524 count += 1

525 yield f'data: {json.dumps({"action": "download", "count": count})}\n\n'

521 for url in urls:

522 async for filename in download_urls(bucket_dir, **url):

523 f.write(f"{filename}\n")

524 if event_stream:

525 count += 1

526 yield f'data: {json.dumps({"action": "download", "count": count})}\n\n'

526 527

527 528

def stream_chunks(bucket_dir: Path, delete_files: bool = False, refine_chunks_with_spacy: bool = False, event_stream: bool = False) -> Iterator[str]:

528 529 size = 0

Modified g4f/tools/run_tools.py +10 -11

@@ -80,7 +80,7 @@ class ToolHandler:

80 80 has_bucket = False

81 81 for message in messages:

82 82 if "content" in message and isinstance(message["content"], str):

83 new_message_content = re.sub(r'{"bucket_id":"([^"]*)"}', on_bucket, message["content"])

83 new_message_content = re.sub(r'{"bucket_id":\s*"([^"]*)"}', on_bucket, message["content"])

84 84 if new_message_content != message["content"]:

85 85 has_bucket = True

86 86 message["content"] = new_message_content

@@ -97,29 +97,28 @@ class ToolHandler:

97 97 """Process all tool calls and return updated messages and kwargs"""

98 98 if not tool_calls:

99 99 return messages, {}

100

101 101 extra_kwargs = {}

102 102 messages = messages.copy()

103 103 sources = None

104

105 105 for tool in tool_calls:

106 106 if tool.get("type") != "function":

107 107 continue

108

109 109 function_name = tool.get("function", {}).get("name")

110

111 111 if function_name == TOOL_NAMES["SEARCH"]:

112 112 messages, sources = await ToolHandler.process_search_tool(messages, tool)

113

114 114 elif function_name == TOOL_NAMES["CONTINUE"]:

115 115 messages, kwargs = ToolHandler.process_continue_tool(messages, tool, provider)

116 116 extra_kwargs.update(kwargs)

117

118 118 elif function_name == TOOL_NAMES["BUCKET"]:

119 119 messages = ToolHandler.process_bucket_tool(messages, tool)

120

121 return messages, sources, extra_kwargs

122 120

121 return messages, sources, extra_kwargs

123 122

124 123 class AuthManager:

125 124 """Handles API key management"""

@@ -128,13 +127,13 @@ class AuthManager:

128 127 def get_api_key_file(cls) -> Path:

129 128 """Get the path to the API key file for a provider"""

130 129 return Path(get_cookies_dir()) / f"api_key_{cls.parent if hasattr(cls, 'parent') else cls.__name__}.json"

131

130

132 131 @staticmethod

133 132 def load_api_key(provider: Any) -> Optional[str]:

134 133 """Load API key from config file if needed"""

135 134 if not getattr(provider, "needs_auth", False):

136 135 return None

137

136

138 137 auth_file = AuthManager.get_api_key_file(provider)

139 138 try:

140 139 if auth_file.exists():

Modified g4f/typing.py +1 -7

@@ -1,6 +1,5 @@

1 import sys

2 1 import os

from typing import Any, AsyncGenerator, Generator, AsyncIterator, Iterator, NewType, Tuple, Union, List, Dict, Type, IO, Optional

from typing import Any, AsyncGenerator, Generator, AsyncIterator, Iterator, NewType, Tuple, Union, List, Dict, Type, IO, Optional, TypedDict

4 3

5 4 try:

6 5 from PIL.Image import Image

@@ -8,11 +7,6 @@ except ImportError:

8 7 class Image:

9 8 pass

10 9

11 if sys.version_info >= (3, 8):

12 from typing import TypedDict

13 else:

14 from typing_extensions import TypedDict

16 10 from .providers.response import ResponseType

17 11

18 12 SHA256 = NewType('sha_256_hash', str)