Full finetune?

[sgt101]

Hi - I have had success fine tuning with lora and dora, but I tried full for fun and it blew up with the message:

RuntimeError: QuantizedMatmul::vjp no gradient wrt the quantized matrix yet.

Is this just because I don't have the memory on my machine.

Thanks !

[awni]

You can't do full fine-tuning on a quantized model. You can use a non-quantized model for that (e.g. fp16, bf16, or fp32).