license: llama3.1
license_link: https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct/blob/main/LICENSE
language:

• en
  pipeline_tag: text-generation
  base_model: meta-llama/Llama-3.1-8B-Instruct
  tags:
• llama-3
• llama
• peft
• PyTorch
• causal-lm
  library_name: transformers

Llama-3.1-8B-Instruct (Fine-Tuned / Modified)

This repository contains a modified/fine-tuned variant of meta-llama/Llama-3.1-8B-Instruct processed using direction-based representation refinement (Abliteration methodology).

Model Details

• Base Model: meta-llama/Llama-3.1-8B-Instruct
• Architecture: Transformer Causal LM (Llama 3 family)
• Parameters: 8 Billion
• Context Length: 128k tokens
• Format: PyTorch / safetensors (fp16)

Quickstart & Usage

You can run this model using Hugging Face transformers on a GPU with at least 16 GB of VRAM (or ~8 GB using 4-bit/8-bit quantization).

1. Requirements

Ensure you have the required packages installed:

pip install transformers torch accelerate bitsandbytes

2. Python Inference Code

Use the following Python snippet to run inference using Hugging Face transformers:

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "2023mt13042/Llama-3.1-8B-uncensored-abliterated"

# Load Tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Load Model in fp16
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Prepare Prompt
messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "Explain quantum computing in simple terms."}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    do_sample=True,
    temperature=0.7
)

response = tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True)
print(response)

4-Bit Quantization (Low VRAM / Google Colab T4)

For GPUs with lower VRAM (~8 GB VRAM), load the model with bitsandbytes 4-bit NF4 quantization:

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

model_id = "2023mt13042/Llama-3.1-8B-uncensored-abliterated"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16
)

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto"
)

Local Usage via GGUF / Ollama

To run this model locally with Ollama:

1. Convert the model to .gguf format using llama.cpp.
2. Create a file named Modelfile with the following configuration:

FROM ./llama-3.1-8b-abliterated-Q4_K_M.gguf

PARAMETER temperature 0.7
PARAMETER top_p 0.9

TEMPLATE """{{ if .System }}<|start_header_id|>system<|end_header_id|>

{{ .System }}<|eot_id|>{{ end }}{{ if .Prompt }}<|start_header_id|>user<|end_header_id|>

{{ .Prompt }}<|eot_id|>{{ end }}<|start_header_id|>assistant<|end_header_id|>

{{ .Response }}<|eot_id|>"""

Build and launch the model in Ollama:

ollama create llama3-abliterated -f Modelfile
ollama run llama3-abliterated

Limitations & Licensing

• Base Model License: Subject to the Meta Llama 3.1 Community License Agreement.
• Responsibility: This model has undergone representation editing. Users are responsible for testing and evaluating outputs prior to deployment in production environments.