Open hjh0119 opened 1 month ago
The training process is quite slow, whereas using 8-bit hqq speeds it up by more than tenfold. Is this normal? Or have I missed any code?
import torch from transformers import EetqConfig, AutoModelForCausalLM config = EetqConfig("int8") # from transformers import HqqConfig # HqqConfig(nbits=8) model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1", quantization_config=config) # train...
Sorry for your trouble. The backward propagation process of EETQ has not been fully optimized.
got it. Is the optimization in schedule?
Not yet.
The training process is quite slow, whereas using 8-bit hqq speeds it up by more than tenfold. Is this normal? Or have I missed any code?