News
The model has been trained using a Producer-Consumer architecture, supports scalable multi-resolution processing (256p to 1328p), and is built to run with Megatron-LM and tensor parallelism.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results