Spaces:
Running
Running
| import os | |
| import json | |
| import base64 | |
| import quickjs | |
| import queue | |
| import threading | |
| from curl_cffi import requests | |
| from ...typing import sha256, Dict, get_type_hints | |
| url = 'https://play.vercel.ai' | |
| model = None | |
| supports_stream = True | |
| models = { | |
| 'claude-instant-v1': 'anthropic:claude-instant-v1', | |
| 'claude-v1': 'anthropic:claude-v1', | |
| 'alpaca-7b': 'replicate:replicate/alpaca-7b', | |
| 'stablelm-tuned-alpha-7b': 'replicate:stability-ai/stablelm-tuned-alpha-7b', | |
| 'bloom': 'huggingface:bigscience/bloom', | |
| 'bloomz': 'huggingface:bigscience/bloomz', | |
| 'flan-t5-xxl': 'huggingface:google/flan-t5-xxl', | |
| 'flan-ul2': 'huggingface:google/flan-ul2', | |
| 'gpt-neox-20b': 'huggingface:EleutherAI/gpt-neox-20b', | |
| 'oasst-sft-4-pythia-12b-epoch-3.5': 'huggingface:OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5', | |
| 'santacoder': 'huggingface:bigcode/santacoder', | |
| 'command-medium-nightly': 'cohere:command-medium-nightly', | |
| 'command-xlarge-nightly': 'cohere:command-xlarge-nightly', | |
| 'gpt-4': 'openai:gpt-4', | |
| 'code-cushman-001': 'openai:code-cushman-001', | |
| 'code-davinci-002': 'openai:code-davinci-002', | |
| 'gpt-3.5-turbo': 'openai:gpt-3.5-turbo', | |
| 'text-ada-001': 'openai:text-ada-001', | |
| 'text-babbage-001': 'openai:text-babbage-001', | |
| 'text-curie-001': 'openai:text-curie-001', | |
| 'text-davinci-002': 'openai:text-davinci-002', | |
| 'text-davinci-003': 'openai:text-davinci-003' | |
| } | |
| vercel_models = {'anthropic:claude-instant-v1': {'id': 'anthropic:claude-instant-v1', 'provider': 'anthropic', 'providerHumanName': 'Anthropic', 'makerHumanName': 'Anthropic', 'minBillingTier': 'hobby', 'parameters': {'temperature': {'value': 1, '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': ['\n\nHuman:'], 'range': []}}, 'name': 'claude-instant-v1'}, 'anthropic:claude-v1': {'id': 'anthropic:claude-v1', 'provider': 'anthropic', 'providerHumanName': 'Anthropic', 'makerHumanName': 'Anthropic', 'minBillingTier': 'hobby', 'parameters': {'temperature': {'value': 1, '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': ['\n\nHuman:'], 'range': []}}, 'name': 'claude-v1'}, 'replicate:replicate/alpaca-7b': {'id': 'replicate:replicate/alpaca-7b', 'provider': 'replicate', 'providerHumanName': 'Replicate', 'makerHumanName': 'Stanford', 'parameters': {'temperature': {'value': 0.75, 'range': [0.01, 5]}, 'maximumLength': {'value': 200, 'range': [50, 512]}, 'topP': {'value': 0.95, 'range': [0.01, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'repetitionPenalty': {'value': 1.1765, 'range': [0.01, 5]}, 'stopSequences': {'value': [], 'range': []}}, 'version': '2014ee1247354f2e81c0b3650d71ca715bc1e610189855f134c30ecb841fae21', 'name': 'alpaca-7b'}, 'replicate:stability-ai/stablelm-tuned-alpha-7b': {'id': 'replicate:stability-ai/stablelm-tuned-alpha-7b', 'provider': 'replicate', 'makerHumanName': 'StabilityAI', 'providerHumanName': 'Replicate', 'parameters': {'temperature': {'value': 0.75, 'range': [0.01, 5]}, 'maximumLength': {'value': 200, 'range': [50, 512]}, 'topP': {'value': 0.95, 'range': [0.01, 1]}, 'presencePenalty': {'value': 0, 'range': [0, 1]}, 'frequencyPenalty': {'value': 0, 'range': [0, 1]}, 'repetitionPenalty': {'value': 1.1765, 'range': [0.01, 5]}, 'stopSequences': {'value': [], 'range': []}}, 'version': '4a9a32b4fd86c2d047f1d271fa93972683ec6ef1cf82f402bd021f267330b50b', 'name': 'stablelm-tuned-alpha-7b'}, 'huggingface:bigscience/bloom': {'id': 'huggingface:bigscience/bloom', 'provider': 'huggingface', 'providerHumanName': 'HuggingFace', 'makerHumanName': 'BigScience', 'instructions': "Do NOT talk to Bloom as an entity, it's not a chatbot but a webpage/blog/article completion model. For the best results: mimic a few words of a webpage similar to the content you want to generate. Start a sentence as if YOU were writing a blog, webpage, math post, coding article and Bloom will generate a coherent follow-up.", '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': 'bloom'}, 'huggingface:bigscience/bloomz': {'id': 'huggingface:bigscience/bloomz', 'provider': 'huggingface', 'providerHumanName': 'HuggingFace', 'makerHumanName': 'BigScience', 'instructions': 'We recommend using the model to perform tasks expressed in natural language. For example, given the prompt "Translate to English: Je t\'aime.", the model will most likely answer "I love you.".', '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': 'bloomz'}, 'huggingface:google/flan-t5-xxl': {'id': 'huggingface:google/flan-t5-xxl', 'provider': 'huggingface', 'makerHumanName': 'Google', 'providerHumanName': 'HuggingFace', 'name': 'flan-t5-xxl', '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]}}}, 'huggingface:google/flan-ul2': {'id': 'huggingface:google/flan-ul2', 'provider': 'huggingface', 'providerHumanName': 'HuggingFace', 'makerHumanName': 'Google', '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': 'flan-ul2'}, 'huggingface:EleutherAI/gpt-neox-20b': {'id': 'huggingface:EleutherAI/gpt-neox-20b', 'provider': 'huggingface', 'providerHumanName': 'HuggingFace', 'makerHumanName': 'EleutherAI', '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]}, 'stopSequences': {'value': [], 'range': []}}, 'name': 'gpt-neox-20b'}, 'huggingface:OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5': {'id': 'huggingface:OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5', 'provider': 'huggingface', 'providerHumanName': 'HuggingFace', 'makerHumanName': 'OpenAssistant', 'parameters': {'maximumLength': {'value': 200, 'range': [50, 1024]}, 'typicalP': {'value': 0.2, 'range': [0.1, 0.99]}, 'repetitionPenalty': {'value': 1, 'range': [0.1, 2]}}, 'name': 'oasst-sft-4-pythia-12b-epoch-3.5'}, 'huggingface:bigcode/santacoder': { | |
| '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(impersonate='chrome110') | |
| 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://play.vercel.ai/openai.jpeg').text | |
| data = json.loads(base64.b64decode(b64)) | |
| script = 'const globalThis = {data: `sentinel`}; (%s)(%s)' % ( | |
| data['c'], data['a']) | |
| token_data = json.loads(quickjs.Context().eval(script).json()) | |
| token_string = json.dumps(separators=(',', ':'), | |
| obj={'r': token_data, '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': 'play.vercel.ai', | |
| 'Origin': 'https://play.vercel.ai', | |
| 'Referrer': 'https://play.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://play.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): | |
| 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']) | |
| conversation += 'assistant: ' | |
| completion = Client().generate(model, conversation) | |
| 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]]) |