Linux Fedora / cpu .and works on 1 thread, dual xeon never answered /5-10 min waits/

Linux Fedora / cpu .and works on 1 thread, dual xeon never answered /5-10 min waits/

Okay thanks, so it's the threading. Thanks for the reply.

Changing threads to 1 did not help. 30 minutes wait, still no output.

import os
import torch

Set single thread environment variables

os.environ["OMP_NUM_THREADS"] = "1"
os.environ["MKL_NUM_THREADS"] = "1"
os.environ["OPENBLAS_NUM_THREADS"] = "1"
os.environ["VECLIB_MAXIMUM_THREADS"] = "1"
os.environ["NUMEXPR_NUM_THREADS"] = "1"
os.environ["TOKENIZERS_PARALLELISM"] = "false"

Configure PyTorch thread settings

torch.set_num_threads(1)
torch.set_num_interop_threads(1)

Check for MPS availability (macOS 12.3+ and PyTorch 1.12+ required)

device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu')

from generate import generate
from transformers import AutoTokenizer, AutoModel

def chat():
# Load the model in bfloat16 on CPU first to avoid MPS dtype issues
model = AutoModel.from_pretrained(
'GSAI-ML/LLaDA-8B-Instruct',
trust_remote_code=True,
torch_dtype=torch.bfloat16 # Load weights in bfloat16
).to('cpu').eval()

:)
yes :))))

cpu cores x 2 are correct threads ,for example 24 cores x 2 = 48 threads
with 1 thread only 1/48 ,2%-4% cpu load

when i run chat.py it use only one thread,not the max threads available
will try your code later

Thanks for uploading. But I am struggling to get the chat.py to run on a M2 Pro 32GB

It won't run with AppleSilicon MPS due to it using bfloat16. I tried changing that to float32 but then it did not run. :D
Now with CPU it is running, but takes ages to reply. All I entered was "hi".
Is this model not supposed to be faster? Anything I need to change?

modifications to the chat.py
from generate import generate
from transformers import AutoTokenizer, AutoModel

def chat():
device = 'cpu' ##<-- force cpu use
model = AutoModel.from_pretrained('GSAI-ML/LLaDA-8B-Instruct', trust_remote_code=True, torch_dtype=torch.bfloat16).to(device).eval()

i'm working on mlx, standby

I'm extremely sorry. I'm not very familiar with running our code on MAC and I'm eagerly looking forward to more help from the community!