request for the training code

[leonardxie]

hi, thank you for your excellent work.
Do you have any plans to share the training code?
i want to reproduce the training but raises error as followers

RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed). Saved intermediate values of the graph are freed when you call .backward() or autograd.grad(). Specify retain_graph=True if you need to backward through the graph a second time or if you need to access saved tensors after calling backward.

[XiaoYiWeio]

me too

[marcio-afr]

I also tried to run a training script using Trainer class from Huggingface and I faced several issues and errors including:

• Diverging tensor devices (lora weights created on cpu when the model is in gpu)
• Different tensor dtype (multiplication between float and bfloat16 when the model is on bfloat16)
• Missing peft configs in xLoRAConfig class.