Reexpress MCP Server

# Copyright Reexpress AI, Inc. All rights reserved. # Update the support set import torch import numpy as np import constants import utils_model import utils_preprocess def batch_support_update(options, main_device): if options.skip_updates_already_in_support: # In this case, we also need to load the calibration set document id's. model = utils_model.load_model_torch(options.model_dir, main_device, load_for_inference=False) else: model = utils_model.load_model_torch(options.model_dir, main_device, load_for_inference=True) print(f"Current support set cardinality: {model.support_index.ntotal}") test_meta_data, _ = \ utils_preprocess.get_metadata_lines(options, options.input_eval_set_file, reduce=False, use_embeddings=options.use_embeddings, concat_embeddings_to_attributes=options.concat_embeddings_to_attributes, calculate_summary_stats=False, is_training=False) test_embeddings = test_meta_data["embeddings"] test_labels = torch.tensor(test_meta_data["labels"]) document_ids = test_meta_data["uuids"] assert test_embeddings.shape[0] == test_labels.shape[0] print(f"test_embeddings.shape: {test_embeddings.shape}") count_already_present_documents = 0 for test_embedding, test_label, document_id in zip(test_embeddings, test_labels, document_ids): if options.skip_updates_already_in_support: if document_id in model.train_uuids or document_id in model.calibration_uuids: count_already_present_documents += 1 continue true_test_label = test_label.item() prediction_meta_data = \ model(test_embedding.unsqueeze(0).to(main_device), forward_type=constants.FORWARD_TYPE_SINGLE_PASS_TEST_WITH_EXEMPLAR) exemplar_vector = prediction_meta_data["exemplar_vector"].detach().cpu().numpy() model.add_to_support(label=true_test_label, predicted_label=prediction_meta_data["prediction"], document_id=document_id, exemplar_vector=exemplar_vector) support_set_cardinality = model.support_index.ntotal assert model.train_labels.shape[0] == support_set_cardinality assert model.train_predicted_labels.shape[0] == support_set_cardinality assert len(model.train_uuids) == support_set_cardinality utils_model.save_support_set_updates(model, options.model_dir) print(f"Updated support set cardinality: {model.support_index.ntotal}") print(f"Note that this does not update the distance quantiles, nor the thresholds on the HR region. " f"This is intended for small, local changes. For more substantive changes, retrain the model.") if options.skip_updates_already_in_support: print(f"Count of skipped document id's already in the support set or calibration set: " f"{count_already_present_documents}")