import os os.environ["CUDA_VISIBLE_DEVICES"] = "" os.environ["TOKENIZERS_PARALLELISM"] = "false" import torch torch.set_default_device('cpu') from sentence_transformers import SentenceTransformer from sklearn.linear_model import LogisticRegression from sklearn.pipeline import make_pipeline from sklearn.preprocessing import StandardScaler from sklearn.metrics import accuracy_score, f1_score, classification_report def train_intent_model(X_train, y_train, X_val, y_val, cfg): print("\n--- Training Intent Model on CPU ---") embedder = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2', device='cpu') print("Encoding training data...") X_train_emb = embedder.encode( list(X_train), show_progress_bar=True, batch_size=16, convert_to_numpy=True, device='cpu' ) print("Encoding validation data...") X_val_emb = embedder.encode( list(X_val), show_progress_bar=True, batch_size=16, convert_to_numpy=True, device='cpu' ) clf = make_pipeline( StandardScaler(), LogisticRegression( max_iter=2000, class_weight='balanced', C=2.0, solver='lbfgs', random_state=cfg.random_state, n_jobs=-1 ) ) print("Fitting classifier...") clf.fit(X_train_emb, y_train) print("Evaluating classifier...") pred = clf.predict(X_val_emb) metrics = { 'val_accuracy': float(accuracy_score(y_val, pred)), 'val_f1_macro': float(f1_score(y_val, pred, average='macro')), 'val_f1_weighted': float(f1_score(y_val, pred, average='weighted')), 'classification_report': classification_report(y_val, pred, zero_division=0, output_dict=True) } return embedder, clf, metrics