BFloat16 is not supported on MPS (MacOs)

Question

I accessed a Llama-based model on Huggingface named: "LeoLM/leo-hessianai-7b-chat". I downloaded the model on my Mac with the device set as 'MPS'. The download worked, however when I want to test the model I get following error:

TypeError: BFloat16 is not supported on MPS

Above I see the hint:

FP4 quantization state not initialized. Please call .cuda() or .to(device) on the LinearFP4 layer first.

Here is my code:

from torch import cuda, bfloat16
import transformers

device = torch.device("mps")

model_id = 'LeoLM/leo-hessianai-7b-chat'

#device = f'cuda:{cuda.current_device()}' if cuda.is_available() else 'cpu'

# set quantization configuration to load large model with less GPU memory
# this requires the `bitsandbytes` library
bnb_config = transformers.BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type='nf4',
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=bfloat16
)

# begin initializing HF items, need auth token for these
hf_auth = 'HF_KEY'
model_config = transformers.AutoConfig.from_pretrained(
    model_id,
    use_auth_token=hf_auth
)

model = transformers.AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=False, # True for flash attention
    config=model_config,
    quantization_config=bnb_config,
    device_map='auto',
    use_auth_token=hf_auth
)
model.eval()
print(f"Model loaded on {device}")

tokenizer = transformers.AutoTokenizer.from_pretrained(
    model_id,
    use_auth_token=hf_auth
)

generate_text = transformers.pipeline(
    model=model, tokenizer=tokenizer,
    return_full_text=True,  # langchain expects the full text
    task='text-generation',
    # we pass model parameters here too
    temperature=0.0,  # 'randomness' of outputs, 0.0 is the min and 1.0 the max
    max_new_tokens=512,  # mex number of tokens to generate in the output
    repetition_penalty=1.1  # without this output begins repeating
)

res = generate_text("Explain the difference between a country and a continent.")
print(res[0]["generated_text"])

What do i need to change to make it run?

Answer 1

Use the nightly build of pytorch.

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cpu

Official source: https://pytorch.org/get-started/locally/