Spaces:
Sleeping
Sleeping
| import os | |
| import configparser | |
| import google.generativeai as genai | |
| import nltk | |
| from datasets import load_metric | |
| from langchain.chat_models import ChatOpenAI | |
| from transformers import AutoTokenizer | |
| from texts.bart_score import BARTScorer | |
| # Constants | |
| # TODO: move to .env | |
| env = configparser.ConfigParser() | |
| env.read(".env") # An example environment: .sample-env | |
| # Get API key | |
| OPENAI_API_KEY = env["API_KEY"]["OPENAI_API_KEY"] | |
| GEMINI_API_KEY = env["API_KEY"]["GEMINI_API_KEY"] | |
| TOGETHER_API_KEY = env["API_KEY"]["TOGETHER_API_KEY"] | |
| # Environment setup | |
| os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY | |
| os.environ["GEMINI_API_KEY"] = GEMINI_API_KEY | |
| os.environ["TOGETHER_API_KEY"] = TOGETHER_API_KEY | |
| os.environ["CURL_CA_BUNDLE"] = "" | |
| os.environ["REQUESTS_CA_BUNDLE"] = "" | |
| # File Path | |
| LOG_FILE = "data/99_log.txt" | |
| OUTPUT_FILE = "data/result.txt" | |
| METRIC_NAME = "roc_auc" | |
| # Training and Model Parameters | |
| TRAIN_RATIO = 0.8 | |
| VAL_RATIO = 0.1 | |
| NUMBER_OF_MAX_EPOCH_WITH_EARLY_STOPPING = 10 | |
| PATIENCE = 3 | |
| BATCH_SIZE = 64 | |
| OPTIMIZED_METRIC = "roc_auc" | |
| SEED = 0 | |
| TEMPERATURE = 0.0 | |
| IS_OUTPUT_NORMALIZATION = False | |
| RATIO = 0.9 | |
| HUMAN_LABEL = 0 | |
| MACHINE_LABEL = 1 | |
| BART = "bart" | |
| # Model Options | |
| MULTIMODEL = "multimodel" | |
| SINGLE_FROM_MULTIMODEL = "single_from_multimodel" | |
| # Downloading the NLTK "punkt" only if it's not already downloaded | |
| nltk.download("punkt", quiet=True) | |
| # API Models | |
| # TODO: consider using an enum | |
| API_ERROR = "API_ERROR" | |
| IGNORE_BY_API_ERROR = "IGNORE_BY_API_ERROR" | |
| CHATGPT = "ChatGPT" | |
| GEMINI = "Gemini" | |
| # LLAMA_2_70_CHAT_TEMP_0 = "LLaMa" | |
| # Initialize BARTScorer | |
| # TODO: consider loading model lazily | |
| bart_scorer = BARTScorer(device="cuda:0", checkpoint="facebook/bart-large-cnn") | |
| # Generative AI configuration | |
| OPENAI_MODEL_NAME = "gpt-3.5-turbo-0125" | |
| GEMINI_MODEL_NAME = "gemini-pro" | |
| genai.configure(api_key=GEMINI_API_KEY, transport="rest") | |
| GEMINI_MODEL = genai.GenerativeModel( | |
| GEMINI_MODEL_NAME, | |
| generation_config={"temperature": TEMPERATURE}, | |
| ) | |
| OPENAI_MODEL = ChatOpenAI( | |
| temperature=TEMPERATURE, | |
| model_name=OPENAI_MODEL_NAME, | |
| ) | |
| # Model paths | |
| MODEL_PATHS = { | |
| "LLaMa": "meta-llama/Llama-2-70b-chat-hf", | |
| "QWEN": "Qwen/Qwen1.5-72B-Chat", | |
| "Yi": "NousResearch/Nous-Hermes-2-Yi-34B", | |
| "Mixtral": "mistralai/Mixtral-8x7B-Instruct-v0.1", | |
| "OLMo": "allenai/OLMo-7B-Instruct", | |
| "Phi": "microsoft/phi-2", | |
| "OpenChat": "openchat/openchat-3.5-1210", | |
| "WizardLM": "WizardLM/WizardLM-13B-V1.2", | |
| "Vicuna": "lmsys/vicuna-13b-v1.5", | |
| } | |
| TOGETHER_PATH = "https://api.together.xyz" | |
| # Roberta model configurations | |
| ROBERTA_BASE = "roberta-base" | |
| ROBERTA_LARGE = "roberta-large" | |
| ROBERTA_MODEL_PATHS = { | |
| ROBERTA_BASE: "roberta-base", | |
| ROBERTA_LARGE: "roberta-large", | |
| } | |
| LEARNING_RATES = { | |
| ROBERTA_BASE: 2e-5, | |
| ROBERTA_LARGE: 8e-6, | |
| } | |
| MODEL_NAME = ROBERTA_BASE | |
| # Tokenizer initialization | |
| tokenizer = AutoTokenizer.from_pretrained(ROBERTA_MODEL_PATHS[MODEL_NAME]) | |
| # Metric loading | |
| metric = load_metric(METRIC_NAME) | |