Qwen3.8-27B-abliterated
27BQwen
from Qwen/Qwen3.8-27B
↓ 3.9M♥ 1.7K166 repos
Qwen3.6-Candid-Coder-35B-A3B
Qwen3.6-Candid-Coder-35B-A3B is a abliterated and uncensored variant of DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B, published on Hugging Face by DarkKitsune. It is 35B total parameters (mixture-of-experts) and 48GB VRAM class class. It is distributed across 4 repositories in GGUF and GGUF (imatrix) formats, totalling 8.1K downloads.
Specification
- Model name
- Qwen3.6-Candid-Coder-35B-A3B
- Source repository
- DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B
- Base model
- DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B
- Publisher
- DarkKitsune
- Parameters
- 35B (mixture-of-experts)
- VRAM required
- ~22.1 GB at Q4_K_M, 4k context (estimated) — runs on 24 GB cards
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- Not declared
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF, GGUF (imatrix)
- First indexed
- 24 Oct 2026
- Last updated
- 3 Nov 2026
Ablation technique
How Qwen3.6-Candid-Coder-35B-A3B was modified, and what that implies.
Abliteration + uncensored fine-tune
Directional ablation combined with additional fine-tuning on unfiltered data, so behaviour diverges from the base model beyond refusal removal alone.
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 abliterated (heretic) DELLA merge of the following finetuned & post-trained models:”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B --local-dir ./qwen3.6-candid-coder-35b-a3b
llama-cli -m ./qwen3.6-candid-coder-35b-a3b/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B --trust-remote-codeDownloads and variants (3)
Every published repository of Qwen3.6-Candid-Coder-35B-A3B, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| DarkKitsune/Qwen3.6-Candid-Coder-35B-A3Bsource | GGUF | 5.9K |
| mradermacher/Qwen3.6-Candid-Coder-35B-A3B-GGUF | GGUF | 1.2K |
| mradermacher/Qwen3.6-Candid-Coder-35B-A3B-i1-GGUF | GGUF (imatrix) | 920 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Qwen3.6-Candid-Coder-35B-A3B 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 DarkKitsune/Qwen3.6-Candid-Coder-35B-A3B, not added by ablation — at 35B 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.
- This release adds fine-tuning on unfiltered data, so its behaviour diverges from the base model beyond refusal removal — differences you measure cannot be attributed to ablation alone.
- No licence is declared on this repository, which is a compliance problem in itself. Resolve it against the base model’s terms before use.
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 Qwen3.6-Candid-Coder-35B-A3B is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with DarkKitsune, and does not independently verify publisher claims. See responsible use.