For privacy reasons a browser tells us at most
"≥ 8 GB RAM, 8 cores" - same reading whether you have 8 GB
or 128 GB. It has no idea how much RAM is free right now, which
apps are open, or whether you have a GPU.
The Abliteration app is integrated with your machine
It reads your exact RAM, GPU model and VRAM,
free memory right now, and picks the sharpest quant that
still fits. Every model page lights up precisely for your rig.
And you can chat with any model, right now
The app is a full local runtime - no API keys, no subscription,
everything runs on your machine. Click any model on this site and
start a conversation in seconds.
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
Huge thanks to Hugging Face for generously sponsoring the compute used to abliterate these models! This was done on HF Jobs, which makes it incredibly easy to run tasks such as these :)
This is Qwen/Qwen3.8-Flash-Next with its refusal behaviour removed by directional ablation. It has the same files, format and BF16 weights as the original, so it loads anywhere the original loads.
Refusal directions were measured separately for chat and think prompts at every layer, on the Q8_0 GGUF. The subspace candidates are the top singular vectors of all of those directions together.
A multi-objective search (minimise refusals, keep the first-token KL on harmless prompts low) chose the subspace rank and per-layer strengths. It settled on a single direction (rank 1) at strength 1.6 in all 48 layers.
That direction was projected out of the residual-stream writes: self_attn.o_proj on the 12 full-attention layers, linear_attn.out_proj on the 36 linear-attention layers, and mlp.shared_expert.down_proj on every layer. That is 96 tensors.
Nothing else was touched: routed experts, embeddings, the vision tower and the MTP layer are bit-identical to the original.
Results (Q8_0 GGUF, greedy)
refusals, chat (/100)
think mode, reasoning or answer refuses (/100)
KL, harmless first token
original
99
96
0
abliterated
7
1
0.068
The 100 prompts are held-out harmful requests. Chat replies were 384 tokens; think mode got 768 tokens. Where the reasoning had not finished by then, the reasoning text itself was judged. The KL figure includes a small noise floor from batched inference.
Files
model-*.safetensors: the patched BF16 checkpoint (same index as the original).
abliteration-lora.gguf: a rank-1 LoRA that applies the same edit to the original GGUFs, e.g. llama-server -m Qwen3.8-Flash-Next-Q8_0-00001-of-….gguf --lora abliteration-lora.gguf.
abliteration-edit.npz: the subspace basis Q and the per-layer strengths scales[layer, (attn, shexp)]. The edit is W' = W - s·QQᵀW.
bake-stats.txt: per-tensor check of how much of the subspace is left after the edit.
This model will comply with harmful requests. You are responsible for how you use it.
README history
4 versions
The author's README evolved over time. Click a version to see its content at that point.
2026-11-18card: Discord linke6e89452.7 KB
Loading...
2026-11-18card: bold thanks lineabe8be22.7 KB
Loading...
2026-11-18card: thanks to Hugging Face for computed97c6bb2.7 KB
Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.