Qwopus3.5-122B-A10B-Kimi-K2.6-destill-healed-abliterated
122BKimi
from Qwen/Qwen3.5-122B-A10B
↓ 6.6K♥ 176 repos
Kimi-Linear-48B-A3B-Instruct-abliterated
Kimi-Linear-48B-A3B-Instruct-abliterated is a refusal-ablated variant of moonshotai/Kimi-Linear-48B-A3B-Instruct, published on Hugging Face by mradermacher. It is 48B total parameters (mixture-of-experts), 48GB VRAM class class and mit licence. It is distributed across 7 repositories in GGUF (imatrix), GGUF and Full precision formats, totalling 20K downloads.
Specification
- Model name
- Kimi-Linear-48B-A3B-Instruct-abliterated
- Source repository
- mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-i1-GGUF
- Base model
- moonshotai/Kimi-Linear-48B-A3B-Instruct
- Publisher
- mradermacher
- Parameters
- 48B (mixture-of-experts)
- VRAM required
- ~30 GB at Q4_K_M, 4k context (estimated) — runs on 48 GB cards
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- mit
- Task
- Unspecified
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF (imatrix), GGUF, Full precision
- First indexed
- 19 Feb 2026
- Last updated
- 1 Nov 2026
Ablation technique
How Kimi-Linear-48B-A3B-Instruct-abliterated 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
“For a convenient overview and download list, visit our model page for this model.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-i1-GGUF --local-dir ./kimi-linear-48b-a3b-instruct-abliterated
llama-cli -m ./kimi-linear-48b-a3b-instruct-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-i1-GGUF"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-i1-GGUF --trust-remote-codeDownloads and variants (6)
Every published repository of Kimi-Linear-48B-A3B-Instruct-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-i1-GGUFsource | GGUF (imatrix) | 7.7K |
| mradermacher/Huihui-Kimi-Linear-48B-A3B-Instruct-abliterated-GGUF | GGUF | 5K |
| mradermacher/Huihui-Kimi-Linear-48B-A3B-Instruct-abliterated-i1-GGUF | GGUF (imatrix) | 3.5K |
| therickcain/Huihui-Kimi-Linear-48B-A3B-Instruct-abliterated-i1-GGUF | GGUF (imatrix) | 1.7K |
| mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-GGUF | GGUF | 1.6K |
| ashwin-sreedhar/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16 | Full precision | 525 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Kimi-Linear-48B-A3B-Instruct-abliterated 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 moonshotai/Kimi-Linear-48B-A3B-Instruct, not added by ablation — at 48B 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 mit, but the terms that bind you are moonshotai/Kimi-Linear-48B-A3B-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 Kimi-Linear-48B-A3B-Instruct-abliterated is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with mradermacher, and does not independently verify publisher claims. See responsible use.