Text Generation
Transformers
Safetensors
English
llama
conversational
Inference Endpoints
text-generation-inference

BFloat16 is not supported on MPS

#13
by nhannn - opened

Hi,
I'm using MacOS Sonoma (14.2.1)

Getting this code from the Model card

import torch
from transformers import pipeline

pipe = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0", torch_dtype=torch.bfloat16, device_map="auto")

Executed it and ran into this problem

TypeError: BFloat16 is not supported on MPS

Please let me know if you need any furthur information

Thank you and best regards

Same for me :(

This should be fixed by the last version of accelerate (0.26.1)

Same for me even with accelerate==0.26.1.

Same here m2 MacBook Air

Same here, MacBook Pro Sonoma.

Same problem when getting Zephyr Beta... There are some solutions online but this happens when loading the model, so I can't edit the code or anything.

Also seeing this on Macbook Pro, M3 Max.

accelerate==0.26.1 does not resolve the issue.

Same here

Sign up or log in to comment