What is the proper way to point abliterator to the model on my local computer?

[TheSilverSmith]

Here is the script I'm using. It's just copy-pasted from the example.

import abliterator

model = "D:\AI\abliterator\model\Qwen2.5-3B-Instruct"
dataset = [abliterator.get_harmful_instructions(), abliterator.get_harmless_instructions()]split by harmful/harmless
device = 'cuda'
n_devices = None
cache_fname = 'my_cached_point.pth'         
activation_layers = None                    
chat_template = None                       
negative_toks = [33260]                    
positive_toks = [39814]             

my_model = abliterator.ModelAbliterator(
  model,
  dataset,
  device='cuda',
  n_devices=None,
  cache_fname=None,
  activation_layers=None,
  chat_template=None,
  positive_toks=positive_toks,
  negative_toks=negative_toks
)

But I keep getting this result, suggesting that I'm not correctly pointing to the file the right way.

[yamaton]

HookedTransformer.from_pretrained_no_processing takes a kwarg local_files_only=True. So you can submit a PR, I suppose.
Ref: https://huggingface.co/blog/mlabonne/abliteration