Qwen2.5-Coder-14B-Instruct-Abliterated
14BQwen
from Qwen/Qwen2.5-Coder-14B-Instruct
↓ 56.2K♥ 7417 repos
Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1
Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1 is a refusal-ablated variant of Qwen/Qwen2.5-Coder-14B-Instruct, published on Hugging Face by Goekdeniz-Guelmez. It is 14B parameters, 24GB VRAM class class and apache-2.0 licence. It is distributed across 5 repositories in Transformers, MLX and GGUF formats, totalling 17.2K downloads.
Specification
- Model name
- Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1
- Base model
- Qwen/Qwen2.5-Coder-14B-Instruct
- Publisher
- Goekdeniz-Guelmez
- Parameters
- 14B
- VRAM required
- ~9.3 GB at Q4_K_M, 4k context (estimated) — runs on 12 GB cards
- Hardware class
- 10B – 25B — 24GB VRAM class
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- Transformers, MLX, GGUF
- First indexed
- 11 Jan 2025
- Last updated
- 18 Apr 2025
Ablation technique
How Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1 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 a abliterated model, and further finetuned model on a custom dataset for more uncensoredness, recomendet system prompt is:”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1-GGUF --local-dir ./josiefied-qwen2.5-coder-14b-instruct-abliterated-v1
llama-cli -m ./josiefied-qwen2.5-coder-14b-instruct-abliterated-v1/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "Goekdeniz-Guelmez/Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve Goekdeniz-Guelmez/Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1 --trust-remote-codeDownloads and variants (4)
Every published repository of Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1, 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.
- Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1 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 Qwen/Qwen2.5-Coder-14B-Instruct, not added by ablation — at 14B 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 apache-2.0, but the terms that bind you are Qwen/Qwen2.5-Coder-14B-Instruct’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 Josiefied-Qwen2.5-Coder-14B-Instruct-abliterated-v1 is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with Goekdeniz-Guelmez, and does not independently verify publisher claims. See responsible use.