Update (g4f/Provider/airforce/AirforceChat.py)

@@ -1,14 +1,15 @@

1 1 from __future__ import annotations

2 2 import re

3 from aiohttp import ClientSession

4 3 import json

5 from typing import List

4 from aiohttp import ClientSession

6 5 import requests

6 from typing import List

7 7

8 8 from ...typing import AsyncResult, Messages

9 9 from ..base_provider import AsyncGeneratorProvider, ProviderModelMixin

10 10 from ..helper import format_prompt

11 11

12 # Helper function to clean the response

12 13 def clean_response(text: str) -> str:

13 14 """Clean response from unwanted patterns."""

14 15 patterns = [

@@ -16,35 +17,27 @@ def clean_response(text: str) -> str:

16 17 r"Rate limit \(\d+\/minute\) exceeded\. Join our discord for more: .+https:\/\/discord\.com\/invite\/\S+",

17 18 r"Rate limit \(\d+\/hour\) exceeded\. Join our discord for more: https:\/\/discord\.com\/invite\/\S+",

18 19 r"</s>", # zephyr-7b-beta

20 r"\[ERROR\] '\w{8}-\w{4}-\w{4}-\w{4}-\w{12}'", # Matches [ERROR] 'UUID'

19 21 ]

20

21 22 for pattern in patterns:

22 23 text = re.sub(pattern, '', text)

23 24 return text.strip()

24 25

25 def split_message(message: dict, chunk_size: int = 995) -> List[dict]:

26 """Split a message into chunks of specified size."""

27 content = message.get('content', '')

28 if len(content) <= chunk_size:

29 return [message]

30

26 def split_message(message: str, max_length: int = 1000) -> List[str]:

27 """Splits the message into chunks of a given length (max_length)"""

28 # Split the message into smaller chunks to avoid exceeding the limit

31 29 chunks = []

32 while content:

33 chunk = content[:chunk_size]

34 content = content[chunk_size:]

35 chunks.append({

36 'role': message['role'],

37 'content': chunk

38 })

30 while len(message) > max_length:

31 # Find the last space or punctuation before max_length to avoid cutting words

32 split_point = message.rfind(' ', 0, max_length)

33 if split_point == -1: # No space found, split at max_length

34 split_point = max_length

35 chunks.append(message[:split_point])

36 message = message[split_point:].strip()

37 if message:

38 chunks.append(message) # Append the remaining part of the message

39 39 return chunks

40 40

41 def split_messages(messages: Messages, chunk_size: int = 995) -> Messages:

42 """Split all messages that exceed chunk_size into smaller messages."""

43 result = []

44 for message in messages:

45 result.extend(split_message(message, chunk_size))

46 return result

47

48 41 class AirforceChat(AsyncGeneratorProvider, ProviderModelMixin):

49 42 label = "AirForce Chat"

50 43 api_endpoint = "https://api.airforce/chat/completions"

@@ -57,45 +50,44 @@ class AirforceChat(AsyncGeneratorProvider, ProviderModelMixin):

57 50 data = response.json()

58 51

59 52 text_models = [model['id'] for model in data['data']]

60

61 53 models = [*text_models]

62

54

63 55 model_aliases = {

64 # openchat

65 "openchat-3.5": "openchat-3.5-0106",

66

67 # deepseek-ai

68 "deepseek-coder": "deepseek-coder-6.7b-instruct",

69

70 # NousResearch

71 "hermes-2-dpo": "Nous-Hermes-2-Mixtral-8x7B-DPO",

72 "hermes-2-pro": "hermes-2-pro-mistral-7b",

73

74 # teknium

75 "openhermes-2.5": "openhermes-2.5-mistral-7b",

76

77 # liquid

78 "lfm-40b": "lfm-40b-moe",

79

80 # DiscoResearch

81 "german-7b": "discolm-german-7b-v1",

82

83 # meta-llama

84 "llama-2-7b": "llama-2-7b-chat-int8",

85 "llama-2-7b": "llama-2-7b-chat-fp16",

86 "llama-3.1-70b": "llama-3.1-70b-chat",

87 "llama-3.1-8b": "llama-3.1-8b-chat",

88 "llama-3.1-70b": "llama-3.1-70b-turbo",

89 "llama-3.1-8b": "llama-3.1-8b-turbo",

90

91 # inferless

92 "neural-7b": "neural-chat-7b-v3-1",

93

94 # HuggingFaceH4

95 "zephyr-7b": "zephyr-7b-beta",

96

97 # llmplayground.net

98 #"any-uncensored": "any-uncensored",

56 # openchat

57 "openchat-3.5": "openchat-3.5-0106",

58

59 # deepseek-ai

60 "deepseek-coder": "deepseek-coder-6.7b-instruct",

61

62 # NousResearch

63 "hermes-2-dpo": "Nous-Hermes-2-Mixtral-8x7B-DPO",

64 "hermes-2-pro": "hermes-2-pro-mistral-7b",

65

66 # teknium

67 "openhermes-2.5": "openhermes-2.5-mistral-7b",

68

69 # liquid

70 "lfm-40b": "lfm-40b-moe",

71

72 # DiscoResearch

73 "german-7b": "discolm-german-7b-v1",

74

75 # meta-llama

76 "llama-2-7b": "llama-2-7b-chat-int8",

77 "llama-2-7b": "llama-2-7b-chat-fp16",

78 "llama-3.1-70b": "llama-3.1-70b-chat",

79 "llama-3.1-8b": "llama-3.1-8b-chat",

80 "llama-3.1-70b": "llama-3.1-70b-turbo",

81 "llama-3.1-8b": "llama-3.1-8b-turbo",

82

83 # inferless

84 "neural-7b": "neural-chat-7b-v3-1",

85

86 # HuggingFaceH4

87 "zephyr-7b": "zephyr-7b-beta",

88

89 # llmplayground.net

90 #"any-uncensored": "any-uncensored",

99 91 }

100 92

101 93 @classmethod

@@ -112,8 +104,6 @@ class AirforceChat(AsyncGeneratorProvider, ProviderModelMixin):

112 104 ) -> AsyncResult:

113 105 model = cls.get_model(model)

114 106

115 chunked_messages = split_messages(messages)

116

117 107 headers = {

118 108 'accept': '*/*',

119 109 'accept-language': 'en-US,en;q=0.9',

@@ -133,36 +123,46 @@ class AirforceChat(AsyncGeneratorProvider, ProviderModelMixin):

133 123

            'user-agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/130.0.0.0 Safari/537.36'

134 124 }

135 125

136 data = {

137 "messages": chunked_messages,

138 "model": model,

139 "max_tokens": max_tokens,

140 "temperature": temperature,

141 "top_p": top_p,

142 "stream": stream

143 }

126 # Format the messages for the API

127 formatted_messages = format_prompt(messages)

128 message_chunks = split_message(formatted_messages)

129

130 full_response = ""

131 for chunk in message_chunks:

132 data = {

133 "messages": [{"role": "user", "content": chunk}],

134 "model": model,

135 "max_tokens": max_tokens,

136 "temperature": temperature,

137 "top_p": top_p,

138 "stream": stream

139 }

140

141 async with ClientSession(headers=headers) as session:

142 async with session.post(cls.api_endpoint, json=data, proxy=proxy) as response:

143 response.raise_for_status()

144 144

145 async with ClientSession(headers=headers) as session:

146 async with session.post(cls.api_endpoint, json=data, proxy=proxy) as response:

147 response.raise_for_status()

148 text = ""

149 if stream:

150 async for line in response.content:

151 line = line.decode('utf-8')

152 if line.startswith('data: '):

153 json_str = line[6:]

154 try:

155 chunk = json.loads(json_str)

156 if 'choices' in chunk and chunk['choices']:

157 content = chunk['choices'][0].get('delta', {}).get('content', '')

158 text += content

159 except json.JSONDecodeError as e:

160 print(f"Error decoding JSON: {json_str}, Error: {e}")

161 elif line.strip() == "[DONE]":

162 break

163 yield clean_response(text)

164 else:

165 response_json = await response.json()

166 text = response_json["choices"][0]["message"]["content"]

167 yield clean_response(text)

145 text = ""

146 if stream:

147 async for line in response.content:

148 line = line.decode('utf-8').strip()

149 if line.startswith('data: '):

150 json_str = line[6:]

151 try:

152 if json_str and json_str != "[DONE]":

153 chunk = json.loads(json_str)

154 if 'choices' in chunk and chunk['choices']:

155 content = chunk['choices'][0].get('delta', {}).get('content', '')

156 text += content

157 except json.JSONDecodeError as e:

158 print(f"Error decoding JSON: {json_str}, Error: {e}")

159 elif line == "[DONE]":

160 break

161 full_response += clean_response(text)

162 else:

163 response_json = await response.json()

164 text = response_json["choices"][0]["message"]["content"]

165 full_response += clean_response(text)

168 166

167 # Return the complete response after all chunks

168 yield full_response

gpt4free

Update (g4f/Provider/airforce/AirforceChat.py)

代码差异