First release on the entire network: Zhipu "Sora" is now open source
Zhipu's CogVideoX, which is the homologous model of "Qingying", was open sourced when this article was published.
The model has been uploaded to Github and Hugging Face.
The model can be run and adjusted on a single A6000 card.
Generated resolution 720 * 480, 6 seconds, 48 frames.
The training data comes from the Internet, and B Station also provides technical support.
*Note 1: Inference is 21.6G, peak is 36G, and fine-tuning is stable at 46.2G, which is within the range of A6000.
*Note 2: The inference optimization has just been updated, and the peak is now 18G, which can be run on a single 4090 card.
Zhipu's CogVideoX, which is the homologous model of "Qingying", was open sourced when this article was published.
The model has been uploaded to Github and Hugging Face.
The model can be run and adjusted on a single A6000 card.
Generated resolution 720 * 480, 6 seconds, 48 frames.
The training data comes from the Internet, and B Station also provides technical support.
*Note 1: Inference is 21.6G, peak is 36G, and fine-tuning is stable at 46.2G, which is within the range of A6000.
*Note 2: The inference optimization has just been updated, and the peak is now 18G, which can be run on a single 4090 card.