In QLoRA the base models parameters are quantized and frozen, and a small number of trainable low rank adapters are added.
So, want to check if my understanding correct.
Is QLoRA a form of Quantization Aware Training (QAT)?
Thank you ![]()
In QLoRA the base models parameters are quantized and frozen, and a small number of trainable low rank adapters are added.
So, want to check if my understanding correct.
Is QLoRA a form of Quantization Aware Training (QAT)?
Thank you ![]()
Yes it is a QAT related method.