from huggingface_hub import snapshot_download

# Replace with your Hugging Face token (optional but recommended)
# token = ""

# Replace with the repository ID you want to download
repo_id = "ICEPVP8977/Uncensored_llama_3.2_3b_safetensors"

try:
  snapshot_download(repo_id=repo_id,
                   #  token=token,
                   local_dir="./model")
  print(f"Successfully downloaded {repo_id} to ./model")
except Exception as e:
  print(f"Error downloading repository: {e}")

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("./model", torch_dtype=torch.float16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("./model")

python

prompt = "Your_question_here"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
max_new_tokens = 2000  # Set the maximum number of tokens in the response
outputs = model.generate(**inputs, max_new_tokens=max_new_tokens)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

RichardErkhov/ICEPVP8977_-_Uncensored_llama_3.2_3b_safetensors-8bits

Original model description:

This model was fine-tuned using a combination of 'uncensored' datasets available on Hugging-Face, as well as the 'Uncensored_mini'.

In my opinion this was a waste of time.

"too toxic." I prefer the LLM to maintain a level of respect when addressing the user without being overly limited or censored.

RichardErkhov/
ICEPVP8977_-_Uncensored_llama_3.2_3b_safetensors-8bits