Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-Reasoning
8BLlamaheretic
from allura-forge/Llama-3.3-8B-Instruct
↓ 75.1K♥ 8219 repos
Dolphin3.0-Llama3.1-8B-abliterated
Dolphin3.0-Llama3.1-8B-abliterated is a refusal-ablated variant of dphn/Dolphin3.0-Llama3.1-8B, published on Hugging Face by skilledu. It is 8B parameters, single consumer GPU class and llama3.1 licence. It is distributed across 5 repositories in GGUF, GGUF (imatrix) and Transformers formats, totalling 6.4K downloads.
Specification
- Model name
- Dolphin3.0-Llama3.1-8B-abliterated
- Source repository
- skilledu/Dolphin3.0-Llama3.1-8B-abliterated
- Base model
- dphn/Dolphin3.0-Llama3.1-8B
- Publisher
- skilledu
- Parameters
- 8B
- Hardware class
- 4B – 10B — single consumer GPU
- Licence
- llama3.1
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF, GGUF (imatrix), Transformers
- First indexed
- 8 Jan 2025
- Last updated
- 8 May 2026
Ablation technique
How Dolphin3.0-Llama3.1-8B-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
“This is an uncensored version of cognitivecomputations/Dolphin3.0-Llama3.1-8B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
Ollama (publisher-documented)
ollama run huihui_ai/dolphin3-abliteratedllama.cpp / GGUF
hf download mradermacher/Dolphin3.0-Llama3.1-8B-abliterated-GGUF --local-dir ./dolphin3.0-llama3.1-8b-abliterated
llama-cli -m ./dolphin3.0-llama3.1-8b-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "skilledu/Dolphin3.0-Llama3.1-8B-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve skilledu/Dolphin3.0-Llama3.1-8B-abliterated --trust-remote-codeDownloads and variants (4)
Every published repository of Dolphin3.0-Llama3.1-8B-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Dolphin3.0-Llama3.1-8B-abliterated-GGUF | GGUF | 3.9K |
| mradermacher/Dolphin3.0-Llama3.1-8B-abliterated-i1-GGUF | GGUF (imatrix) | 1.9K |
| G4E-Unknown/Dolphin3.0-Llama3.1-8B-abliterated-Q4_K_M-GGUF | GGUF | 352 |
| skilledu/Dolphin3.0-Llama3.1-8B-abliteratedsource | Transformers | 258 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Dolphin3.0-Llama3.1-8B-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 dphn/Dolphin3.0-Llama3.1-8B, not added by ablation — at 8B 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 llama3.1, but the terms that bind you are dphn/Dolphin3.0-Llama3.1-8B’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 Dolphin3.0-Llama3.1-8B-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 skilledu, and does not independently verify publisher claims. See responsible use.