Does loading in 4bit override an 8bit model?

🤗Transformers
1

Having saved a model in 8bit:

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    load_in_8bit = True,
    device_map = 'auto')

model.save_pretrained('model_weights_test')

I then loaded it in 4bit (with a different script):

double_quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)
    
model = AutoModelForCausalLM.from_pretrained(
    'model_weights_test',
    quantization_config=double_quant_config,
    device_map = 'auto')

However I am then running into OOM issues I was not seeing by just initially loading it in 4bit:

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=double_quant_config,
    device_map = 'auto')

My question is therefore: does the 4bit config fail to override the 8bit saved model? Am I essentially just training an 8bit model, hence the OOM issues?