Model Card for Model ID

This model is a zeroth-generation, downsampled training of the CyberSolve LinAlg model. See the model card for the most updated full training of CyberSolve LinAlg here.

Simulating the larger, full training and evaluation process, we trained and evaluated CyberSolve on a 10% split of the 2M total records available in the 1D Linear Algebra split of the Google DeepMind Mathematics dataset. The results found in this smaller training convinced us that the FLAN-T5 model would indeed learn to effectively solve linear equations. That is, this preliminary training green lighted the full model training for us.