Huihui · Qwen3.6 · 35B-A3B · abliterated

_{EXL3 · 4.5 bpw · 21.6 GB · Mixture‑of‑Experts · 40 layers × 256 experts}

An ExLlamaV3 build of huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated at 4.5 bits per weight — quality-leaning build — comfortable on a 24 GB card with usable context (~22 GB weights). See Quants for sibling repos at other bit‑widths or browse the collection.

Quants

BPW	Head bits	Calibration rows	Size	Status
4.0	8	128	19.5 GB	link
4.5	8	250	21.6 GB	`this repo`

Inference

Loader	Use it for
TabbyAPI	OpenAI‑compatible HTTP server. Drop‑in for OpenAI clients.
text‑generation‑webui	Local chat UI. Pick the ExLlamaV3 loader from the model dropdown.
ExLlamaV3	Direct Python API for embedding the model in your own code or pipeline.

VRAM at 4.5 bpw: ~21.6 GB weights + ~2 GB context. Comfortable on a single 24 GB card; tight on 16 GB with reduced context.

Download

pip install -U huggingface_hub

hf download \
  blockblockblock/Huihui-Qwen3.6-35B-A3B-abliterated-exl3-4.5bpw \
  --local-dir ./Huihui-Qwen3.6-35B-A3B-abliterated-exl3-4.5bpw

Quantization recipe _{(advanced — embedded in quantization_config.json)}

Setting	Value
Format	`EXL3`
Bits per weight	`4.5`
Head bits	`8`
Calibration rows	`250`
Codebook	`MCG`
Out‑scales	`always`
Parallel mode	`enabled` (MoE expert batching)

Loaded automatically by every ExLlamaV3 loader; reproduced here for searchability.

License & use

Use and license follow the base model. Quantization adds no additional restrictions. Refer to the upstream repository for terms, citation, and safety documentation.

_{Quantized with BlockQuant · convention {org}/{model}-exl3-{bpw}bpw}