license: other
license_name: qwen-community-1.0
license_link: https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/LICENSE
base_model: mrfakename/Qwen3.8-Flash-Next-abliterated
pipeline_tag: image-text-to-text
tags:
- gguf
- abliterated
- uncensored
Qwen3.8-Flash-Next (abliterated) GGUF
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 :)
Join the Discord: https://discord.com/invite/foss

GGUF quants of mrfakename/Qwen3.8-Flash-Next-abliterated, which is Qwen/Qwen3.8-Flash-Next with its refusal behaviour removed by directional ablation.
These are the unsloth/Qwen3.8-Flash-Next-GGUF quants with the edit applied in place. Only the edited tensors were dequantized, edited and requantized to their original type. Everything else is byte-identical.
| Folder | Size | Files | Notes |
|---|---|---|---|
Q8_0 |
175 GiB | 6 | Closest to the original weights. |
UD-Q4_K_XL |
104 GiB | 4 | Good default. |
UD-Q2_K_XL |
74 GiB | 4 | Smallest. The edit still survives in full (see below). |
Each folder holds a split GGUF. Point llama.cpp at the -00001-of- file and it loads the rest.
What changed
- 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_projon the 12 full-attention layers,linear_attn.out_projon the 36 linear-attention layers, andmlp.shared_expert.down_projon every layer. That is 96 tensors.
Nothing else was touched: routed experts, embeddings, the vision projector and the MTP file are byte-identical to the original quants.
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.
Does the edit survive quantization?
Yes, in every folder here. Low-bit quants can round a small edit away, but these unsloth UD quants keep the edited tensors at Q6_K or Q8_0 even in UD-Q2_K_XL, and the patch log shows essentially none of the edit lost (at most 2%). See patch-stats.txt.
Running
Mainline llama.cpp runs these; they were patched and tested at commit 2cbfc3c.
Files
Q8_0/,UD-Q4_K_XL/,UD-Q2_K_XL/: the patched quants, e.g.llama-server -m UD-Q4_K_XL/Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-….gguf.mmproj-*.gguf: the unchanged vision projector, and the unchangedmtp-*.gguf.abliteration-lora.gguf: the edit as a LoRA, if you'd rather use another quant of the original:--lora abliteration-lora.gguf.abliteration-edit.npz: the subspace basis and per-layer strengths.patch-stats.txt: per-tensor patch log.
This model will comply with harmful requests. You are responsible for how you use it.