Neu BAckground model for HXMT (Spec-BERT v1.0.0)

This repository contains the trained model weights and test dataset used for the publication and open-source project NeuBach, a transformer-based background modeling framework for the Insight-HXMT mission. The model best_model.pt is a PyTorch checkpoint trained to predict NaI detector background spectra in the 25–200 keV range based on satellite orbital and instrumental parameters. The included test dataset (dataset_P*_HE_v3.0.csv) contains representative samples used to evaluate the model’s performance, including parameters such as ELV, COR, SAT_X, SAT_Y, SAT_Z, anti-coincidence detector counts, and particle monitor counts. Key features: Pretrained background spectrum predictor for HXMT NaI detector Developed using PyTorch and transformer architecture (BERT variant) Compatible with scripts in the GitHub repository https://github.com/tuoyl/NeuBacH Model input: second-level telemetry parameters Output: predicted photon count spectrum (25–200 keV)