All quants generated with importance matrix (imatrix) for optimal quality preservation on abliterated weights.

Specs

• 35B total parameters, ~3B active per forward pass (MoE)
• 256 experts, 8 routed + 1 shared per token
• Hybrid architecture: Gated DeltaNet linear attention + full softmax attention (3:1 ratio)
• 40 layers, pattern: 10 x (3 x DeltaNet-MoE + 1 x Attention-MoE)
• 262K native context (extendable to 1M with YaRN)
• Natively multimodal (text, image, video)
• Multi-token prediction (MTP) support
• 248K vocabulary, 201 languages
• Based on Qwen/Qwen3.5-35B-A3B

Recommended Settings

From the official Qwen authors:

Thinking mode (default):

• General: temperature=1.0, top_p=0.95, top_k=20, min_p=0, presence_penalty=1.5
• Coding/precise tasks: temperature=0.6, top_p=0.95, top_k=20, min_p=0, presence_penalty=0

Non-thinking mode:

• General: temperature=0.7, top_p=0.8, top_k=20, min_p=0, presence_penalty=1.5
• Reasoning tasks: temperature=1.0, top_p=1.0, top_k=40, min_p=0, presence_penalty=2.0

Important:

• Keep at least 128K context to preserve thinking capabilities
• Use --jinja flag with llama.cpp for proper chat template handling
• Vision support requires the mmproj file alongside the main GGUF

Usage

Works with llama.cpp, LM Studio, Jan, koboldcpp, and other GGUF-compatible runtimes.

# Text only
llama-cli -m Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf \
  --jinja -c 131072 -ngl 99

# With vision
llama-cli -m Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf \
  --mmproj mmproj-Qwen3.5-35B-A3B-Uncensored-HauhauCS-Aggressive-f16.gguf \
  --jinja -c 131072 -ngl 99

Note: LM Studio may show 256x2.6B in the params column instead of 35B-A3B — this is a cosmetic metadata quirk, the model runs correctly.

Other Formats

• GGUF (this repo)
• GPTQ — coming soon

Other Models

• Qwen3.5-27B-Uncensored-HauhauCS-Aggressive