deepseek-ai/deepseek-coder-7b-instruct-v1.5 · Problems fine tuning with PEFT

I tried to fine-tune the deepseek-coder-7b-instruct-v1.5 model on A10 (AWS g5.12xlarge) using 4bit quantization with bitsandbytes. My bitsandbytes configuration:

bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type="nf4",
            bnb_4bit_compute_dtype=torch.bfloat16,
            gradient_checkpointing=True,
        )

with device_map="auto" for loading.

I kept running into the error message:

You can't train a model that has been loaded
in 8-bit precision on a different device than the one you're training
on.

However both 8bit bitsandbytes loading and single GPU blow my GPU memory. Any suggestions?