flint-roleplay-antislop-heretic
Otherheretic
from RecursiveSI/flint
↓ 6.5K♥ 23 repos
Flint-40B-A2.5B-abliterated
Flint-40B-A2.5B-abliterated is a refusal-ablated variant of RecursiveSI/flint, published on Hugging Face by dima6312. It is 40B total parameters (mixture-of-experts), 48GB VRAM class class and apache-2.0 licence. It is distributed across 7 repositories in Transformers, GGUF, MLX and GGUF (imatrix) formats, totalling 33.6K downloads.
Specification
- Model name
- Flint-40B-A2.5B-abliterated
- Source repository
- dima6312/Flint-40B-A2.5B-abliterated
- Base model
- RecursiveSI/flint
- Publisher
- dima6312
- Parameters
- 40B (mixture-of-experts)
- VRAM required
- ~25.1 GB at Q4_K_M, 4k context (estimated) — runs on 48 GB cards
- Hardware class
- 25B – 50B — 48GB VRAM class
- Licence
- apache-2.0
- Task
- Text generation
- Context window
- Not documented
- Ablation scope
- Not stated by the publisher
- Formats available
- Transformers, GGUF, MLX, GGUF (imatrix)
- First indexed
- 23 Oct 2026
- Last updated
- 1 Nov 2026
Ablation technique
How Flint-40B-A2.5B-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
“An abliterated (uncensored) Flint: RecursiveSI's Flint 40B-A2.5B with its refusal direction removed, for security research and coding work: exploit analysis, malware triage, detection engineering, CTFs and red-team tooling that the original model often declines.”
Running it
Commands are templates — confirm the exact repository and quant file before use.
llama.cpp / GGUF
hf download dima6312/Flint-40B-A2.5B-abliterated-GGUF --local-dir ./flint-40b-a2.5b-abliterated
llama-cli -m ./flint-40b-a2.5b-abliterated/<file>.gguf -p "..."Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
repo = "dima6312/Flint-40B-A2.5B-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
repo, torch_dtype="auto", device_map="auto"
)vLLM
vllm serve dima6312/Flint-40B-A2.5B-abliterated --trust-remote-codeDownloads and variants (6)
Every published repository of Flint-40B-A2.5B-abliterated, including quantised re-releases by other authors.
| Repository | Format | Downloads |
|---|---|---|
| dima6312/Flint-40B-A2.5B-abliterated-GGUF | GGUF | 23.7K |
| mradermacher/Flint-40B-A2.5B-abliterated-i1-GGUF | GGUF (imatrix) | 2.7K |
| dima6312/Flint-40B-A2.5B-abliterated-MLX-4bit | MLX | 2.6K |
| mradermacher/Flint-40B-A2.5B-abliterated-GGUF | GGUF | 1.8K |
| dima6312/Flint-40B-A2.5B-abliterated-NVFP4-GGUF | GGUF | 1.6K |
| dima6312/Flint-40B-A2.5B-abliteratedsource | Transformers | 1.1K |
Security-team assessment
Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.
- Flint-40B-A2.5B-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 RecursiveSI/flint, not added by ablation — at 40B 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 RecursiveSI/flint’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 Flint-40B-A2.5B-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 dima6312, and does not independently verify publisher claims. See responsible use.