Spark-x2.5-4B-Heretic-jp
4BOtherheretic
from XHToken/Spark-X2.5-4B
↓ 452.1K♥ 182 repos
K2-Horizon-7B-heretic
K2-Horizon-7B-heretic is a Heretic-ablated variant of IFM/K2-Horizon-7B, published on Hugging Face by Avicennasis. It is 7B parameters, single consumer GPU class and apache-2.0 licence. It is distributed across 5 repositories in GGUF, Transformers, MLX and GGUF (imatrix) formats, totalling 12.3K downloads.
Specification
- Model name
- K2-Horizon-7B-heretic
- Source repository
- Avicennasis/K2-Horizon-7B-heretic
- Base model
- IFM/K2-Horizon-7B
- Publisher
- Avicennasis
- Parameters
- 7B
- VRAM required
- ~5 GB at Q4_K_M, 4k context (estimated) — runs on 6 GB cards
- Hardware class
- 4B – 10B — single consumer GPU
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- GGUF, Transformers, MLX, GGUF (imatrix)
- First indexed
- 3 Oct 2026
- Last updated
- 14 Oct 2026
Ablation technique
How K2-Horizon-7B-heretic 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
“An abliterated (refusal-removed) version of IFM/K2-Horizon-7B, produced in-house with Heretic (adversarial per-layer ablation directions). Full merged BF16 checkpoint — not an adapter.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download mradermacher/K2-Horizon-7B-heretic-i1-GGUF --local-dir ./k2-horizon-7b-heretic
llama-cli -m ./k2-horizon-7b-heretic/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "Avicennasis/K2-Horizon-7B-heretic"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve Avicennasis/K2-Horizon-7B-heretic --trust-remote-codeDownloads and variants (4)
Every published repository of K2-Horizon-7B-heretic, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| mradermacher/K2-Horizon-7B-heretic-i1-GGUF | GGUF (imatrix) | 10.7K |
| mradermacher/K2-Horizon-7B-heretic-GGUF | GGUF | 1.3K |
| Avicennasis/K2-Horizon-7B-heretic-mlx-8bit | MLX | 128 |
| Avicennasis/K2-Horizon-7B-hereticsource | Transformers | 116 |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- K2-Horizon-7B-heretic 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 IFM/K2-Horizon-7B, not added by ablation — at 7B parameters, expect a small model’s reasoning and a high error rate on exploit detail.
- 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 IFM/K2-Horizon-7B’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 K2-Horizon-7B-heretic is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with Avicennasis, and does not independently verify publisher claims. See responsible use.