Open ldjpr opened 1 year ago
have the same issue here, did not find a solution unfortunately
On Fri 2. Jun 2023 at 01:58, ldjpr @.***> wrote:
How can I configure to run in a Intel Mac with AMD GPU?
I get the error: "Torch not compiled with CUDA enabled"
Has anyone been able to run it?
— Reply to this email directly, view it on GitHub https://github.com/haoheliu/AudioLDM/issues/89, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHV4JVOEODIKBJ62S576F6LXJEUCHANCNFSM6AAAAAAYXTOU2Y . You are receiving this because you are subscribed to this thread.Message ID: @.***>
-- David Kamp I studiokamp.com http://www.studiokamp.com
I got it running on my M1 using the AudioLDM pipeline from HuggingSpace.
(https://huggingface.co/docs/diffusers/v0.17.1/en/api/pipelines/audioldm)
from diffusers import AudioLDMPipeline
import torch
repo_id = "cvssp/audioldm-m-full"
pipe = AudioLDMPipeline.from_pretrained(repo_id, torch_dtype=torch.float64)
# pipe = pipe.to("cude")
pipe = pipe.to("cpu")
prompt = "mystical, dreamy analog synth strings"
negative_prompt = "low quality, average quality, noisey, high pitch, artefacts"
audio = pipe(prompt=prompt, audio_length_in_s=5.0, num_inference_steps=10, guidance_scale=2.5, negative_prompt=negative_prompt).audios[0]
from IPython.display import Audio
Audio(audio, rate=44100)
with the conda dependencies:
name: audio_env
channels:
- defaults
dependencies:
- python=3.11.0
- pip
- jupyter
- pip:
- torch==2.0.1
- diffusers
- transformers
seems also to work and be a bit faster with
pipe = pipe.to("mps")
@suckrowPierre Thanks for the code and conda env. I tried to use pipe = pipe.to("mps")
but got the error
TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.
Did you run unto this problem at all?
How can I configure to run in a Intel Mac with AMD GPU?
I get the error: "Torch not compiled with CUDA enabled"
Has anyone been able to run it?