Why is my model bigger after compression?

Hello,

I am using this config on a translation model (Helsinki-NLP/opus-mt-zh-en), and I check the size of the model using the following function before and after running init_compression and deepspeed.initialize:

def print_size_of_model(model, label=""):
    torch.save(model.state_dict(), "temp.p")
    size=os.path.getsize("temp.p")
    print("model: ",label,' \t','Size (KB):', size/1e3)
    os.remove('temp.p')
    return size

Weirdly, the size of the model increases after running init_compression and deepspeed.initialize. Even after I use redundancy_clean at the end of training and save the model to disk, the size of the model stays what had been returned by print_size_of_model after running init_compression and deepspeed.initialize.

Am I missing something? Can you please explain?

Thanks a lot

Start with model_compression/bert/config/XTC/ds_config_W1A8_Qgroup1_fp32.json and reproduce the reported size comparison around init_compression and deepspeed.initialize using the provided print_size_of_model function. Trace the resulting state_dict and redundancy_clean behavior; done means explaining why the serialized model grows and whether the final saved size is expected.