diff options
Diffstat (limited to 'g4f/Provider/Providers/Vercel.py')
| -rw-r--r-- | g4f/Provider/Providers/Vercel.py | 117 |
1 files changed, 8 insertions, 109 deletions
diff --git a/g4f/Provider/Providers/Vercel.py b/g4f/Provider/Providers/Vercel.py index e1e84f9a..03d9be17 100644 --- a/g4f/Provider/Providers/Vercel.py +++ b/g4f/Provider/Providers/Vercel.py @@ -42,120 +42,19 @@ vercel_models = {'anthropic:claude-instant-v1': {'id': 'anthropic:claude-instant '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': []}}}} -# based on https://github.com/ading2210/vercel-llm-api // modified -class Client: - def __init__(self): - self.session = requests.Session() - self.headers = { - 'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/110 Safari/537.36', - 'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,*/*;q=0.8', - 'Accept-Encoding': 'gzip, deflate, br', - 'Accept-Language': 'en-US,en;q=0.5', - 'Te': 'trailers', - 'Upgrade-Insecure-Requests': '1' - } - self.session.headers.update(self.headers) - - def get_token(self): - b64 = self.session.get('https://sdk.vercel.ai/openai.jpeg').text - data = json.loads(base64.b64decode(b64)) - - code = 'const globalThis = {data: `sentinel`}; function token() {return (%s)(%s)}' % ( - data['c'], data['a']) - - token_string = json.dumps(separators=(',', ':'), - obj={'r': execjs.compile(code).call('token'), 't': data['t']}) - - return base64.b64encode(token_string.encode()).decode() - - def get_default_params(self, model_id): - return {key: param['value'] for key, param in vercel_models[model_id]['parameters'].items()} - - def generate(self, model_id: str, prompt: str, params: dict = {}): - if not ':' in model_id: - model_id = models[model_id] - - defaults = self.get_default_params(model_id) - - payload = defaults | params | { - 'prompt': prompt, - 'model': model_id, - } - - headers = self.headers | { - 'Accept-Encoding': 'gzip, deflate, br', - 'Custom-Encoding': self.get_token(), - 'Host': 'sdk.vercel.ai', - 'Origin': 'https://sdk.vercel.ai', - 'Referrer': 'https://sdk.vercel.ai', - 'Sec-Fetch-Dest': 'empty', - 'Sec-Fetch-Mode': 'cors', - 'Sec-Fetch-Site': 'same-origin', - } - - chunks_queue = queue.Queue() - error = None - response = None - - def callback(data): - chunks_queue.put(data.decode()) - - def request_thread(): - nonlocal response, error - for _ in range(3): - try: - response = self.session.post('https://sdk.vercel.ai/api/generate', - json=payload, headers=headers, content_callback=callback) - response.raise_for_status() - - except Exception as e: - if _ == 2: - error = e - - else: - continue - - thread = threading.Thread(target=request_thread, daemon=True) - thread.start() - - text = '' - index = 0 - while True: - try: - chunk = chunks_queue.get(block=True, timeout=0.1) - - except queue.Empty: - if error: - raise error - - elif response: - break - - else: - continue - - text += chunk - lines = text.split('\n') - - if len(lines) - 1 > index: - new = lines[index:-1] - for word in new: - yield json.loads(word) - index = len(lines) - 1 - def _create_completion(model: str, messages: list, stream: bool, **kwargs): + return + # 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' - 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' - - for message in messages: - conversation += '%s: %s\n' % (message['role'], message['content']) + # for message in messages: + # conversation += '%s: %s\n' % (message['role'], message['content']) - conversation += 'assistant: ' + # conversation += 'assistant: ' - completion = Client().generate(model, conversation) + # completion = Client().generate(model, conversation) - for token in completion: - yield token + # for token in completion: + # yield token params = f'g4f.Providers.{os.path.basename(__file__)[:-3]} supports: ' + \ '(%s)' % ', '.join([f"{name}: {get_type_hints(_create_completion)[name].__name__}" for name in _create_completion.__code__.co_varnames[:_create_completion.__code__.co_argcount]])
\ No newline at end of file |