Llama-3.2-3B-Instruct-uncensored
3BLlamauncensored
from meta-llama/Llama-3.2-3B-Instruct
↓ 67.7K♥ 16050 repos
Hermes-3-Llama-3.2-3B-abliterated
Hermes-3-Llama-3.2-3B-abliterated is a refusal-ablated variant of NousResearch/Hermes-3-Llama-3.2-3B, published on Hugging Face by lunahr. It is 3B parameters, runs on a laptop CPU class and llama3 licence. It is distributed across 6 repositories in Transformers, GGUF and GGUF (imatrix) formats, totalling 5.5K downloads.
Specification
- Model name
- Hermes-3-Llama-3.2-3B-abliterated
- Source repository
- lunahr/Hermes-3-Llama-3.2-3B-abliterated
- Base model
- NousResearch/Hermes-3-Llama-3.2-3B
- Publisher
- lunahr
- Parameters
- 3B
- VRAM required
- ~2.5 GB at Q4_K_M, 4k context (estimated) — runs on 4 GB cards
- Hardware class
- Under 4B — runs on a laptop CPU
- Licence
- llama3
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- Transformers, GGUF, GGUF (imatrix)
- First indexed
- 13 Dec 2024
- Last updated
- 3 Mar 2025
Ablation technique
How Hermes-3-Llama-3.2-3B-abliterated was modified, and what that implies.
Abliteration (directional ablation)
The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.
Reported scope: Not stated by the publisher. Publishers frequently omit this, so verify behaviour empirically rather than assuming full coverage.
From the publisher’s model card
“An abliterated version of the below model:”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Hermes-3-Llama-3.2-3B-abliterated-GGUF --local-dir ./hermes-3-llama-3.2-3b-abliterated
llama-cli -m ./hermes-3-llama-3.2-3b-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "lunahr/Hermes-3-Llama-3.2-3B-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve lunahr/Hermes-3-Llama-3.2-3B-abliterated --trust-remote-codeDownloads and variants (5)
Every published repository of Hermes-3-Llama-3.2-3B-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Hermes-3-Llama-3.2-3B-abliterated-GGUF | GGUF | 4.3K |
| mradermacher/Hermes-3-Llama-3.2-3B-abliterated-i1-GGUF | GGUF (imatrix) | 997 |
| Melvin56/Hermes-3-Llama-3.2-3B-abliterated-Q4_0-GGUF | GGUF | 125 |
| lunahr/Hermes-3-Llama-3.2-3B-abliteratedsource | Transformers | 56 |
| Melvin56/Hermes-3-Llama-3.2-3B-abliterated-Q8_0-GGUF | GGUF | 51 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Hermes-3-Llama-3.2-3B-abliterated will attempt offensive-security prompts that a hosted commercial model declines, which is what makes it usable for red-team corpus generation and for measuring what an unaligned model of this class produces.
- Capability is inherited from NousResearch/Hermes-3-Llama-3.2-3B, not added by ablation — at 3B parameters, expect a small model’s reasoning and a high error rate on exploit detail.
- The publisher does not state which layers were ablated, so assume nothing about refusal consistency — probe it directly.
- Declared licence is llama3, but the terms that bind you are NousResearch/Hermes-3-Llama-3.2-3B’s — a re-release cannot grant more than its parent.
Handling: run it isolated, put moderation in front of it if anyone outside your team can reach it, and verify the weights before loading — these are third-party artifacts. Full handling and licence guidance.
Metadata for Hermes-3-Llama-3.2-3B-abliterated is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with lunahr, and does not independently verify publisher claims. See responsible use.