license: apache-2.0
base_model: Qwen/Qwen3.5-122B-A10B
library_name: gguf
pipeline_tag: text-generation
tags:
- gguf
- abliterated
- qwen3.5
- moe
- mtp
- not-for-all-audiences
Qwen3.5-122B-A10B - Abliterated GGUF
Abliterated, importance-matrix (imatrix) quantized GGUFs of
Qwen/Qwen3.5-122B-A10B. Multi-Token Prediction
(MTP) and vision tensors are preserved through abliteration, conversion, and quantization.
These quants were made from the full-precision fp16 safetensors base at
RobinsonLabs/Qwen3.5-122B-A10B-abliterated -
use that repo if you want to re-abliterate, merge a LoRA, fine-tune, or roll your own quants.
Disclosure
This model is abliterated - the hard-refusal reflex on adult / creative content has been
reduced via single-direction weight orthogonalization. Harm guardrails are retained by design:
self-harm prompts still redirect to help (e.g. 988), and it is not intended to assist genuine
wrongdoing. This is a v1, partial abliteration; capability is preserved. Taggednot-for-all-audiences. Use responsibly - you are responsible for your use. License inherited from
the base model: Apache-2.0.
Files
| File | Quant | ~Size | Notes |
|---|---|---|---|
Qwen3.5-122B-A10B-abliterated-Q8_0.gguf |
Q8_0 | ~132 GB | near-lossless master |
Qwen3.5-122B-A10B-abliterated-Q6_K.gguf |
Q6_K | ~102 GB | |
Qwen3.5-122B-A10B-abliterated-Q5_K_M.gguf |
Q5_K_M | ~89 GB | |
Qwen3.5-122B-A10B-abliterated-Q5_K_S.gguf |
Q5_K_S | ~86 GB | |
Qwen3.5-122B-A10B-abliterated-Q4_K_M.gguf |
Q4_K_M | ~76 GB | |
Qwen3.5-122B-A10B-abliterated-Q4_K_S.gguf |
Q4_K_S | ~71 GB | |
Qwen3.5-122B-A10B-abliterated-IQ4_XS.gguf |
IQ4_XS | ~67 GB | quality/size sweet spot |
Qwen3.5-122B-A10B-abliterated-Q3_K_M.gguf |
Q3_K_M | ~60 GB | |
Qwen3.5-122B-A10B-abliterated-IQ3_M.gguf |
IQ3_M | ~55 GB | |
Qwen3.5-122B-A10B-abliterated-Q3_K_S.gguf |
Q3_K_S | ~54 GB | |
Qwen3.5-122B-A10B-abliterated-IQ3_XS.gguf |
IQ3_XS | ~52 GB | |
Qwen3.5-122B-A10B-abliterated-Q2_K.gguf |
Q2_K | ~46 GB | |
Qwen3.5-122B-A10B-abliterated-IQ2_M.gguf |
IQ2_M | ~42 GB | |
Qwen3.5-122B-A10B-abliterated-IQ2_XS.gguf |
IQ2_XS | ~38 GB | smallest |
All quants are MTP-preserved and imatrix-weighted.
Method
- Abliteration: single mid-layer refusal direction removed via weight orthogonalization on the
bf16 base; MTP (nextn) block, vision, and routers preserved. - Quant: importance-matrix (imatrix) weighted, MTP-preserving convert + quantize with
llama.cpp.
fp16 base
The full-precision fp16 safetensors base this ladder was quantized from is at
RobinsonLabs/Qwen3.5-122B-A10B-abliterated.
That repo is the master for further surgery (re-abliteration, LoRA merge, fine-tune) and for making
your own quants.
Provenance
Qwen3.5-122B-A10B (base, Apache-2.0) -> abliterated (bf16) -> Q8_0 master -> imatrix quants.