Llama-3.3-70B-Instruct-abliterated
70BLlama
from meta-llama/Llama-3.3-70B-Instruct
↓ 103.7K♥ 7921 repos
Meta-Llama-3-70B-Instruct-abliterated-v3.5
Meta-Llama-3-70B-Instruct-abliterated-v3.5 is a refusal-ablated variant of failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5, published on Hugging Face by failspy. It is 70B parameters, multi-GPU class and llama3 licence. It is distributed across 16 repositories in Transformers, GGUF, GGUF (imatrix) and EXL2 formats, totalling 13.5K downloads.
Specification
- Model name
- Meta-Llama-3-70B-Instruct-abliterated-v3.5
- Source repository
- failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5
- Publisher
- failspy
- Parameters
- 70B
- Hardware class
- 50B – 100B — multi-GPU
- Licence
- llama3
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- Transformers, GGUF, GGUF (imatrix), EXL2, AWQ
- First indexed
- 28 May 2024
- Last updated
- 29 Jun 2026
Ablation technique
How Meta-Llama-3-70B-Instruct-abliterated-v3.5 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
“My original Jupyter "cookbook" to replicate the methodology can be found here”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download legraphista/Meta-Llama-3-70B-Instruct-abliterated-v3.5-IMat-GGUF --local-dir ./meta-llama-3-70b-instruct-abliterated-v3.5
llama-cli -m ./meta-llama-3-70b-instruct-abliterated-v3.5/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5 --trust-remote-codeDownloads and variants (15)
Every published repository of Meta-Llama-3-70B-Instruct-abliterated-v3.5, including quantised re-releases by other authors.
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Meta-Llama-3-70B-Instruct-abliterated-v3.5 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 failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5, not added by ablation — at 70B parameters, expect roughly the base model’s competence, minus some instruction-following fidelity.
- 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 failspy/Meta-Llama-3-70B-Instruct-abliterated-v3.5’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 Meta-Llama-3-70B-Instruct-abliterated-v3.5 is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with failspy, and does not independently verify publisher claims. See responsible use.