Qwen3.8-27B-abliterated
27BQwen
from Qwen/Qwen3.8-27B
↓ 3.9M♥ 1.7K166 repos
Qwen3.8-27B-S-mirai
Qwen3.8-27B-S-mirai is a refusal-ablated variant of trymirai/Qwen3.8-27B-S-experimental, published on Hugging Face by alesha-pro. It is 27B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 2 repositories in GGUF format, totalling 16K downloads.
Specification
- Model name
- Qwen3.8-27B-S-mirai
- Source repository
- alesha-pro/Qwen3.8-27B-S-mirai-GGUF
- Base model
- trymirai/Qwen3.8-27B-S-experimental
- Publisher
- alesha-pro
- Parameters
- 27B
- VRAM required
- ~17.2 GB at Q4_K_M, 4k context (estimated) — runs on 24 GB cards
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- 128K tokens (as documented)
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF
- First indexed
- 26 Sept 2026
- Last updated
- 7 Oct 2026
Ablation technique
How Qwen3.8-27B-S-mirai 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
“compressed trellis codes are copied into the GGUF bit for bit, nothing is re-quantized. A 27B model in 11.2 GB that runs on one 12 GB GPU, with tools, reasoning, images, and an optional file that removes refusals.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download alesha-pro/Qwen3.8-27B-S-mirai-GGUF --local-dir ./qwen3.8-27b-s-mirai
llama-cli -m ./qwen3.8-27b-s-mirai/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "alesha-pro/Qwen3.8-27B-S-mirai-GGUF"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve alesha-pro/Qwen3.8-27B-S-mirai-GGUF --trust-remote-codeDownloads and variants (1)
The only published repository of Qwen3.8-27B-S-mirai.
| Repository | Format | Downloads |
|---|---|---|
| alesha-pro/Qwen3.8-27B-S-mirai-GGUFsource | GGUF | 16K |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Qwen3.8-27B-S-mirai 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 trymirai/Qwen3.8-27B-S-experimental, not added by ablation — at 27B 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 trymirai/Qwen3.8-27B-S-experimental’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 Qwen3.8-27B-S-mirai is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with alesha-pro, and does not independently verify publisher claims. See responsible use.