⌦abliteratedmodels.org

Ornith-1.5-35B-A3B-FULLY-OBLITERATED

Ornith-1.5-35B-A3B-FULLY-OBLITERATED is a refusal-ablated variant of jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED, published on Hugging Face by jhone888. It is 35B total parameters (mixture-of-experts), 48GB VRAM class class and mit licence. The publisher describes ablation applied across all 40 layers. It is distributed across 4 repositories in GGUF, GGUF (imatrix) and Transformers formats, totalling 9.6K downloads.

Abliteration (directional ablation)35B MoEOrnith25B – 50B↓ 9,615♥ 4
View on Hugging Face ↗Base model ↗

Specification

Model name
Ornith-1.5-35B-A3B-FULLY-OBLITERATED
Source repository
jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED
Base model
jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED
Publisher
jhone888
Parameters
35B (mixture-of-experts)
Hardware class
25B – 50B — 48GB VRAM class
Licence
mit
Task
Text generation
Context window
Not documented
Ablation scope
All 40 layers
Formats available
GGUF, GGUF (imatrix), Transformers
First indexed
29 Aug 2026
Last updated
31 Aug 2026

Ablation technique

How Ornith-1.5-35B-A3B-FULLY-OBLITERATED 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: All 40 layers. This comes from the publisher’s own model card and has not been independently verified.

From the publisher’s model card

“A second-pass obliteration of huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated. The v1 abliteration covered only 11 of 29 targeted components; v2 applies a full-model refusal-direction ablation across all 40 layers, then verifies the result numerically.”

Running it

Commands are templates — confirm the exact repository and quant file before use.

llama.cpp / GGUF
hf download mradermacher/Ornith-1.5-35B-A3B-FULLY-OBLITERATED-GGUF --local-dir ./ornith-1.5-35b-a3b-fully-obliterated
llama-cli -m ./ornith-1.5-35b-a3b-fully-obliterated/<file>.gguf -p "..."
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype="auto", device_map="auto"
)
vLLM
vllm serve jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED --trust-remote-code

Downloads and variants (3)

Every published repository of Ornith-1.5-35B-A3B-FULLY-OBLITERATED, including quantised re-releases by other authors.

RepositoryFormatDownloads
mradermacher/Ornith-1.5-35B-A3B-FULLY-OBLITERATED-GGUFGGUF5.9K
mradermacher/Ornith-1.5-35B-A3B-FULLY-OBLITERATED-i1-GGUFGGUF (imatrix)2.8K
jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATEDsourceTransformers965

Security-team assessment

Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.

  • Ornith-1.5-35B-A3B-FULLY-OBLITERATED 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 jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED, not added by ablation — at 35B parameters, expect roughly the base model’s competence, minus some instruction-following fidelity.
  • Full-depth ablation maximises compliance and is the configuration most likely to have measurably degraded the base model.
  • Declared licence is mit, but the terms that bind you are jhone888/Ornith-1.5-35B-A3B-FULLY-OBLITERATED’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 Ornith-1.5-35B-A3B-FULLY-OBLITERATED is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with jhone888, and does not independently verify publisher claims. See responsible use.