Qwen3.8-27B-abliterated
27BQwen
from Qwen/Qwen3.8-27B
↓ 3.9M♥ 1.7K166 repos
qwen2.5-coder-32b-instruct-heretic-sft
qwen2.5-coder-32b-instruct-heretic-sft is a Heretic-ablated variant of PeetPedro/qwen2.5-coder-32b-instruct-heretic-sft, published on Hugging Face by PeetPedro. It is 32B parameters, 48GB VRAM class class and apache-2.0 licence. It is distributed across 4 repositories in Transformers, GGUF and GGUF (imatrix) formats, totalling 20.9K downloads.
Specification
- Model name
- qwen2.5-coder-32b-instruct-heretic-sft
- Source repository
- PeetPedro/qwen2.5-coder-32b-instruct-heretic-sft
- Publisher
- PeetPedro
- Parameters
- 32B
- VRAM required
- ~20.3 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
- Unspecified
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- Transformers, GGUF, GGUF (imatrix)
- First indexed
- 19 Jul 2026
- Last updated
- 30 Oct 2026
Ablation technique
How qwen2.5-coder-32b-instruct-heretic-sft was modified, and what that implies.
Heretic
Automated directional ablation via the Heretic toolchain, which searches for the refusal direction and applies it with a KL-divergence budget so general capability is preserved.
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
“Part of the 8b.is AI training family: weights, corpora and evaluations produced on hardware we own, kept reproducible and local-first. Nothing here is rented inference, and no claim below is a benchmark unless it says so.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/qwen2.5-coder-32b-instruct-heretic-sft-i1-GGUF --local-dir ./qwen2.5-coder-32b-instruct-heretic-sft
llama-cli -m ./qwen2.5-coder-32b-instruct-heretic-sft/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "PeetPedro/qwen2.5-coder-32b-instruct-heretic-sft"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve PeetPedro/qwen2.5-coder-32b-instruct-heretic-sft --trust-remote-codeDownloads and variants (3)
Every published repository of qwen2.5-coder-32b-instruct-heretic-sft, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/qwen2.5-coder-32b-instruct-heretic-sft-i1-GGUF | GGUF (imatrix) | 20K |
| mradermacher/qwen2.5-coder-32b-instruct-heretic-sft-GGUF | GGUF | 830 |
| PeetPedro/qwen2.5-coder-32b-instruct-heretic-sftsource | Transformers | 50 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- qwen2.5-coder-32b-instruct-heretic-sft 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 PeetPedro/qwen2.5-coder-32b-instruct-heretic-sft, not added by ablation — at 32B 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 PeetPedro/qwen2.5-coder-32b-instruct-heretic-sft’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 qwen2.5-coder-32b-instruct-heretic-sft is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with PeetPedro, and does not independently verify publisher claims. See responsible use.