Meta-Llama-3.1-8B-Instruct-heretic
A decensored variant of meta-llama/Meta-Llama-3.1-8B-Instruct, produced with Heretic v1.4.0 (directional ablation / "abliteration"). Refusal behavior is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the base model's knowledge and instruction-following are left largely intact.
Who this is for: developers who want Llama 3.1 8B that answers instead of lecturing - the most widely used open LLM, now without refusals. Best run on an 8-12 GB GPU or via the Q4_K_M/Q5_K_M GGUF on consumer hardware. Not a capability upgrade over base Meta-Llama-3.1-8B-Instruct - same model, refusal guardrails removed.
Runs on your gaming PC
Full GGUF ladder included — pick the quant that fits your card:
| Your GPU | Recommended quant | Weights |
|---|---|---|
| RTX 3090 / 4090 / 5090 (24 GB) | Q8_0 | 8.54 GB |
| RTX 4080 / 5080 / 4060 Ti 16G (16 GB) | Q6_K | 6.60 GB |
| RTX 3060 / 4070 / 5070 (12 GB) | Q5_K_M | 5.73 GB |
| RTX 4060 / 3070 (8 GB) | Q4_K_M | 4.92 GB |
| GTX 1660 Super / 2060 / 3050 laptop (6 GB) | Q4_K_M | 4.92 GB |
| CPU-only / Apple Silicon | Q4_K_M | 4.92 GB, fits in system RAM |
Weights only, at this model's ~8.0B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up.
Why abliteration instead of fine-tuning
Fine-tuning a "helpful" persona on top of RLHF'd refusals fights the base model's training and tends to degrade coherence. Abliteration instead finds and edits the specific weight directions responsible for refusal, leaving the rest of the network (and its capabilities) untouched. See the Heretic repo and the original abliteration writeup for the mechanism.
Files
Safetensors (BF16)
The full-precision weights are in model-*.safetensors (see the repo file listing for exact shard count and sizes).
GGUF quantizations
GGUF quantizations are published for this model (Q4_K_M, Q5_K_M, Q6_K, Q8_0). Pull a specific quant with llama.cpp / ollama.
| File | Format | Size |
|---|---|---|
Meta-Llama-3.1-8B-Instruct-heretic-Q4_K_M.gguf | GGUF Q4_K_M | (see repo files) |
Meta-Llama-3.1-8B-Instruct-heretic-Q5_K_M.gguf | GGUF Q5_K_M | (see repo files) |
Meta-Llama-3.1-8B-Instruct-heretic-Q6_K.gguf | GGUF Q6_K | (see repo files) |
Meta-Llama-3.1-8B-Instruct-heretic-Q8_0.gguf | GGUF Q8_0 | (see repo files) |
Quickstart
# llama.cpp - defaults to the Q4_K_M quant
llama serve -hf saidutta69/Meta-Llama-3.1-8B-Instruct-heretic:Q4_K_M
# transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "saidutta69/Meta-Llama-3.1-8B-Instruct-heretic"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)
# ... inference code
Also runnable via Ollama, LM Studio, Jan, vLLM, SGLang.
Responsible use
Refusal suppression is deliberate and works as intended: this model will comply with requests the base model would refuse, including some it shouldn't. There is no safety filtering layered on top. You are responsible for how you deploy it.
Made with ❤️ by RACER IS OP — follow for more uncensored models
License
Inherits the llama-3.1-community-license license from the base model.