Weiheng Liu

I have no the choice of "extension" in my window menu T.T 

> I found a workaround to get the training started (can't promise on the model training quality as the optimizer changes). > > The CUDA out-of-memory error during optim.step() or...

> I also face this problem, I set batch_size=1, It is also out of memory. Did you train this torch model with lora?

> > > I found a workaround to get the training started (can't promise on the model training quality as the optimizer changes). > > > The CUDA out-of-memory error...

> > Thanks a lot！Based on your previous training experience, does using an 8-bit optimizer affect the final performance? > > No, based on my tests with the model I...