Thanks for sharing your excellent work.
~While attempting to execute inference.ipynb on an RTX3090 with 24GB, I encountered an Out of Memory error right at the onset, specifically when loading the model in the first cell. Is there a possibility to successfully run SongComposer on single/multiple commercial-grade GPUs like RTX3090/4090 ?~
Thanks for sharing your excellent work. ~While attempting to execute inference.ipynb on an RTX3090 with 24GB, I encountered an Out of Memory error right at the onset, specifically when loading the model in the first cell. Is there a possibility to successfully run SongComposer on single/multiple commercial-grade GPUs like RTX3090/4090 ?~
just change
.cuda().half()to.half().cuda()...