language:
- en
- multilingual
license: other
license_name: llama3.2
pipeline_tag: text-generation
tags: - llama
- llama-3.2
- instruct
- uncensored
- heretic
- abliteration
- text-generation
- multilingual
- not-for-all-audiences
base_model: meta-llama/Llama-3.2-3B-Instruct
library_name: transformers
Llama-3.2-3B-Instruct-uncensored
Highly uncensored fine-tune of Llama-3.2-3B-Instruct with near-zero refusals.
Only 4 out of 100 carefully selected test prompts were refused while remaining on 0.0265 KL-divergence,
which means the model keeps a good part of its original quality of its base Llama model.
Created with Heretic for maximum compliance, directness, and unrestricted text generation.
Model Details
- Developed by: [tostideluxekaas]
- Base model:
unsloth/Llama-3.2-3B-Instruct - Model type: Causal decoder-only transformer (fine-tuned)
- Language(s): English, multilingual
- Fine-tuning method: Heretic abliteration (uncensoring pass)
- Context length: 8,192 tokens (same as base model)
- License: Llama 3.2 Community License (derivative)
Disclaimer
⚠️ Use this model responsibly.
This is a heavily uncensored model and may generate harmful, illegal, offensive, NSFW, or otherwise inappropriate content.
You are solely responsible for all outputs and any consequences that may arise from its use.
Intended Use
- Creative writing, roleplay, fiction, and open-ended dialogue
- Research on model alignment and safety
- Local deployment (perfect for Ollama, LM Studio, SillyTavern, etc.)
- Any application where maximum freedom and minimal refusals are desired
Out-of-Scope Use / Limitations
- Not intended for production systems without human oversight
- Not suitable for children or sensitive environments
- May still produce biased or factually incorrect information (same as base model)
- Not optimized for code generation or high-precision reasoning tasks
Abliteration parameters
| Parameter | Value |
|---|---|
| direction_index | 13.36 |
| attn.o_proj.max_weight | 1.47 |
| attn.o_proj.max_weight_position | 18.11 |
| attn.o_proj.min_weight | 0.95 |
| attn.o_proj.min_weight_distance | 12.76 |
| mlp.down_proj.max_weight | 1.29 |
| mlp.down_proj.max_weight_position | 21.70 |
| mlp.down_proj.min_weight | 0.56 |
| mlp.down_proj.min_weight_distance | 11.73 |
Performance/metrics
| Metric | This model | Original model (unsloth/Llama-3.2-3B-Instruct) |
|---|---|---|
| KL divergence | 0.0265 | 0 (by definition) |
| Refusals | 4/100 | 97/100 |
unsloth/Llama-3.2-3B-Instruct
For more details on the model, please go to Meta's original model card
Special Thanks
A huge thank you to the Meta and Llama team for creating and releasing these models.
Model Information
The Meta Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. They outperform many of the available open source and closed chat models on common industry benchmarks.
Model developer: Meta
Model Architecture: Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
Supported languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai are officially supported. Llama 3.2 has been trained on a broader collection of languages than these 8 supported languages. Developers may fine-tune Llama 3.2 models for languages beyond these supported languages, provided they comply with the Llama 3.2 Community License and the Acceptable Use Policy. Developers are always expected to ensure that their deployments, including those that involve additional languages, are completed safely and responsibly.
Llama 3.2 family of models Token counts refer to pretraining data only. All model versions use Grouped-Query Attention (GQA) for improved inference scalability.
Model Release Date: Sept 25, 2024
Status: This is a static model trained on an offline dataset. Future versions may be released that improve model capabilities and safety.
License: Use of Llama 3.2 is governed by the Llama 3.2 Community License (a custom, commercial license agreement).
Where to send questions or comments about the model Instructions on how to provide feedback or comments on the model can be found in the model README. For more technical information about generation parameters and recipes for how to use Llama 3.1 in applications, please go here.
How to Use
from transformers import pipeline
pipe = pipeline(
"text-generation",
model="tostideluxekaas/Llama-3.2-3B-Instruct-uncensored",
device="cuda" # or "cpu"
)
messages = [
{"role": "user", "content": "Write a dark horror story..."}
]
output = pipe(messages, max_new_tokens=512)
print(output[0]["generated_text"][-1]["content"])'''