I'm very excited to see you contributing code to this cross-modal fusion model in the open source community. After reading your code and paper, I found that the code only contains model code, and there is no code related to development and training details. The relevant tricks mentioned in the paper, such as using the same architecture of CM3leon, using CM3 targets, introducing multi-modal retrieval enhancements, adopting mixed-modal scaling laws for hyper-parameter settings, and mixed-modal decoding strategies, have rich guiding training significance, which makes me I found it very interesting and aroused great curiosity. I believe that based on the relevant introduction of the paper and code assistance, it can further deepen the understanding of the paper and the model, and can also enrich your warehouse to increase its attention. Therefore, I would like to ask you again if you can open source the relevant code for the training part. I would be grateful if the relevant code could be open sourced. Thank you again for your outstanding contributions to the open source community!