Update Vercel.py

@@ -42,120 +42,19 @@ vercel_models = {'anthropic:claude-instant-v1': {'id': 'anthropic:claude-instant

42 42

    'id': 'huggingface:bigcode/santacoder', 'provider': 'huggingface', 'providerHumanName': 'HuggingFace', 'makerHumanName': 'BigCode', 'instructions': 'The model was trained on GitHub code. As such it is not an instruction model and commands like "Write a function that computes the square root." do not work well. You should phrase commands like they occur in source code such as comments (e.g. # the following function computes the sqrt) or write a function signature and docstring and let the model complete the function body.', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 0.95, 'range': [0.01, 0.99]}, 'topK': {'value': 4, 'range': [1, 500]}, 'repetitionPenalty': {'value': 1.03, 'range': [0.1, 2]}}, 'name': 'santacoder'}, 'cohere:command-medium-nightly': {'id': 'cohere:command-medium-nightly', 'provider': 'cohere', 'providerHumanName': 'Cohere', 'makerHumanName': 'Cohere', 'name': 'command-medium-nightly', 'parameters': {'temperature': {'value': 0.9, 'range': [0, 2]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0, 1]}, 'topK': {'value': 0, 'range': [0, 500]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'cohere:command-xlarge-nightly': {'id': 'cohere:command-xlarge-nightly', 'provider': 'cohere', 'providerHumanName': 'Cohere', 'makerHumanName': 'Cohere', 'name': 'command-xlarge-nightly', 'parameters': {'temperature': {'value': 0.9, 'range': [0, 2]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0, 1]}, 'topK': {'value': 0, 'range': [0, 500]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'openai:gpt-4': {'id': 'openai:gpt-4', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'name': 'gpt-4', 'minBillingTier': 'pro', 'parameters': {'temperature': {'value': 0.7, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'openai:code-cushman-001': {'id': 'openai:code-cushman-001', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}, 'name': 'code-cushman-001'}, 'openai:code-davinci-002': {'id': 'openai:code-davinci-002', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}, 'name': 'code-davinci-002'}, 'openai:gpt-3.5-turbo': {'id': 'openai:gpt-3.5-turbo', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'parameters': {'temperature': {'value': 0.7, 'range': [0, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'topK': {'value': 1, 'range': [1, 500]}, 'presencePenalty': {'value': 1, 'range': [0, 1]}, 'frequencyPenalty': {'value': 1, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}, 'name': 'gpt-3.5-turbo'}, 'openai:text-ada-001': {'id': 'openai:text-ada-001', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'name': 'text-ada-001', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'openai:text-babbage-001': {'id': 'openai:text-babbage-001', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'name': 'text-babbage-001', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'openai:text-curie-001': {'id': 'openai:text-curie-001', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'name': 'text-curie-001', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'openai:text-davinci-002': {'id': 'openai:text-davinci-002', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'name': 'text-davinci-002', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}, 'openai:text-davinci-003': {'id': 'openai:text-davinci-003', 'provider': 'openai', 'providerHumanName': 'OpenAI', 'makerHumanName': 'OpenAI', 'name': 'text-davinci-003', 'parameters': {'temperature': {'value': 0.5, 'range': [0.1, 1]}, 'maximumLength': {'value': 200, 'range': [50, 1024]}, 'topP': {'value': 1, 'range': [0.1, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'stopSequences': {'value': [], 'range': []}}}}

43 43

44 44

45 # based on https://github.com/ading2210/vercel-llm-api // modified

46 class Client:

47 def __init__(self):

48 self.session = requests.Session()

49 self.headers = {

50

            'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/110 Safari/537.36',

51 'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,*/*;q=0.8',

52 'Accept-Encoding': 'gzip, deflate, br',

53 'Accept-Language': 'en-US,en;q=0.5',

54 'Te': 'trailers',

55 'Upgrade-Insecure-Requests': '1'

56 }

57 self.session.headers.update(self.headers)

58

59 def get_token(self):

60 b64 = self.session.get('https://sdk.vercel.ai/openai.jpeg').text

61 data = json.loads(base64.b64decode(b64))

62

63 code = 'const globalThis = {data: `sentinel`}; function token() {return (%s)(%s)}' % (

64 data['c'], data['a'])

65

66 token_string = json.dumps(separators=(',', ':'),

67 obj={'r': execjs.compile(code).call('token'), 't': data['t']})

68

69 return base64.b64encode(token_string.encode()).decode()

70

71 def get_default_params(self, model_id):

72 return {key: param['value'] for key, param in vercel_models[model_id]['parameters'].items()}

73

74 def generate(self, model_id: str, prompt: str, params: dict = {}):

75 if not ':' in model_id:

76 model_id = models[model_id]

77

78 defaults = self.get_default_params(model_id)

79

80 payload = defaults | params | {

81 'prompt': prompt,

82 'model': model_id,

83 }

84

85 headers = self.headers | {

86 'Accept-Encoding': 'gzip, deflate, br',

87 'Custom-Encoding': self.get_token(),

88 'Host': 'sdk.vercel.ai',

89 'Origin': 'https://sdk.vercel.ai',

90 'Referrer': 'https://sdk.vercel.ai',

91 'Sec-Fetch-Dest': 'empty',

92 'Sec-Fetch-Mode': 'cors',

93 'Sec-Fetch-Site': 'same-origin',

94 }

95

96 chunks_queue = queue.Queue()

97 error = None

98 response = None

99

100 def callback(data):

101 chunks_queue.put(data.decode())

102

103 def request_thread():

104 nonlocal response, error

105 for _ in range(3):

106 try:

107 response = self.session.post('https://sdk.vercel.ai/api/generate',

108 json=payload, headers=headers, content_callback=callback)

109 response.raise_for_status()

110

111 except Exception as e:

112 if _ == 2:

113 error = e

114

115 else:

116 continue

117

118 thread = threading.Thread(target=request_thread, daemon=True)

119 thread.start()

120

121 text = ''

122 index = 0

123 while True:

124 try:

125 chunk = chunks_queue.get(block=True, timeout=0.1)

126

127 except queue.Empty:

128 if error:

129 raise error

130

131 elif response:

132 break

133

134 else:

135 continue

136

137 text += chunk

138 lines = text.split('\n')

139

140 if len(lines) - 1 > index:

141 new = lines[index:-1]

142 for word in new:

143 yield json.loads(word)

144 index = len(lines) - 1

145

146 45 def _create_completion(model: str, messages: list, stream: bool, **kwargs):

46 return

47

    # conversation = 'This is a conversation between a human and a language model, respond to the last message accordingly, referring to the past history of messages if needed.\n'

147 48

148

    conversation = 'This is a conversation between a human and a language model, respond to the last message accordingly, referring to the past history of messages if needed.\n'

149

150 for message in messages:

151 conversation += '%s: %s\n' % (message['role'], message['content'])

49 # for message in messages:

50 # conversation += '%s: %s\n' % (message['role'], message['content'])

152 51

153 conversation += 'assistant: '

52 # conversation += 'assistant: '

154 53

155 completion = Client().generate(model, conversation)

54 # completion = Client().generate(model, conversation)

156 55

157 for token in completion:

158 yield token

56 # for token in completion:

57 # yield token

159 58

160 59 params = f'g4f.Providers.{os.path.basename(__file__)[:-3]} supports: ' + \

161 60

    '(%s)' % ', '.join([f"{name}: {get_type_hints(_create_completion)[name].__name__}" for name in _create_completion.__code__.co_varnames[:_create_completion.__code__.co_argcount]])

gpt4free

代码差异