Reexpress MCP Server

# Copyright Reexpress AI, Inc. All rights reserved. from sdm_model import SimilarityDistanceMagnitudeCalibrator import utils_pretraining_initialization import logging import sys logger = logging.getLogger(__name__) logger.setLevel(logging.DEBUG) logger.addHandler(logging.StreamHandler(sys.stdout)) def train(options, train_embeddings=None, calibration_embeddings=None, train_labels=None, calibration_labels=None, model_params=None, main_device=None, model_dir=None, model=None): import baseline_utils_model if model is None: print("Initializing model") model = SimilarityDistanceMagnitudeCalibrator(**model_params).to(main_device) # For the baseline adaptor, we use the 'pretrain' routine model, min_held_out_balanced_loss, min_held_out_balanced_loss_epoch = \ utils_pretraining_initialization.pretrain(options, model=model, model_dir=model_dir, held_out_embeddings=calibration_embeddings, held_out_labels=calibration_labels, train_embeddings=train_embeddings, train_labels=train_labels, pretraining_learning_rate=options.learning_rate, return_min_held_out_balanced_loss=True, main_device=main_device, use_main_device=True ) model = model.to(main_device) baseline_utils_model.save_baseline_model(model, model_dir) logger.info(f"Model saved to iteration model_dir") return min_held_out_balanced_loss, min_held_out_balanced_loss_epoch