license: apache-2.0
base_model: Qwen/Qwen3.5-122B-A10B
library_name: transformers
pipeline_tag: text-generation
tags:
- abliterated
- uncensored
- mtp
- qwen3.5
- moe
- not-for-all-audiences
Qwen3.5-122B-A10B - Abliterated (fp16 base)
Abliterated, Native-MTP-Preserved fp16 base of
Qwen/Qwen3.5-122B-A10B, in safetensors.
Multi-Token Prediction (MTP) and vision tensors are preserved through abliteration.
This is the full-precision base that the
RobinsonLabs/Qwen3.5-122B-A10B-abliterated-GGUF
quant ladder was quantized from. If you want a ready-to-run quant, use that repo. This repo is the
master for further surgery (re-abliteration, LoRA merge, fine-tune) and for making 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.
Method
- Abliteration: single mid-layer refusal direction removed via weight orthogonalization on the
fp16 base; MTP (nextn) block, vision, and routers preserved. - Format: safetensors, sharded, with config + tokenizer + index. Native MTP head preserved.
Files
| Format | Precision | ~Size | Notes |
|---|---|---|---|
| safetensors (sharded) | fp16 | ~234 GB | abliterated base; MTP preserved |
Quants
GGUF quants (Q8_0 down to IQ2_XS, MTP-preserved, imatrix-weighted) are published at
RobinsonLabs/Qwen3.5-122B-A10B-abliterated-GGUF.
Provenance
Qwen3.5-122B-A10B (base, Apache-2.0) -> abliterated (fp16) -> GGUF quant ladder.