license: apache-2.0
tags:

• text-generation
• gguf
• nvfp4
• fp4
• moe
• qwen3.6
• uncensored
• vision
• multimodal
• genesis
• mtp
• speculative-decoding
• llama-cpp
  language:
• en
• zh
• multilingual
  pipeline_tag: image-text-to-text
  base_model:
• HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive
• LuffyTheFox/Qwen3.6-35B-A3B-Uncensored-Genesis-Final-GGUF
  base_model_relation: quantized

🌟 Qwen3.6-35B-A3B-Uncensored-Genesis-Final-NVFP4-GGUF

This repository contains the NVFP4 (NVIDIA 4-bit Floating Point) quantized GGUF release of LuffyTheFox's final release: Qwen3.6-35B-A3B-Uncensored-Genesis-Final-GGUF.

• Original Model & Genesis Tensor Repair: LuffyTheFox — Original Model Repo
• NVFP4 Quantization & MTP Integration: jan1k
• Base Model: HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive (0/465 refusals)
• Quantization Format: NVFP4 (NVIDIA FP4) GGUF (optimized for NVIDIA Ada Lovelace / Blackwell / Ampere GPUs)
• Architecture: Mixture of Experts (MoE) — 35B total parameters / ~3B active per token (8 routed experts + 1 shared)
• Multimodal (Vision): Supported via mmproj-Hermes3.6-35B-A3B-Uncensored-Genesis-F16.gguf

⚡ What is Genesis?

Genesis is a post-training data regeneration and calibration algorithm for neural networks (LLM) in GGUF format developed by LuffyTheFox over almost half a year of development with AI assistance. It is optimized, architecture-independent, works with any model in GGUF format, and is based on mathematical statistics.

LLM models often suffer from:

• Saturated weights: activations get stuck, gradients vanish, outputs degrade.
• Scale mismatches: one layer's weights are 10× larger than peers for no valid reason.
• Mean drift: weight distributions shifted positive or negative, breaking symmetry assumptions.
• Zero blocks: zero blocks corrupt the signal, turning training into noise amplification.
• Training noise: increases randomness and ruins output quality.

Genesis fixes all of that without retraining — pure numerical surgery on the raw bytes of the file:

1. Stage 1: Scans ssm_conv1d tensors (handling long-context memory) and repairs balance between heads.
2. Stage 2: Scans blocks in chunks (via 3 parameters) and picks the best fit to weight distribution in the tensor, replacing zero blocks without touching learned structure.
3. Stage 3: Scans for noise via custom SVD (excluding token_embd.weight, output.weight, 1D tensors, bias, and norms). Reduces training noise based on the Marchenko–Pastur law while preserving 99% of signal and learned gradient.

⚡ Sampling Recommendations (Genesis SVD Stabilization)

Due to the removal of tensor noise via Marchenko–Pastur SVD noise-gate stabilization, the model is most stable when Top P and Min P are disabled for coding and precise tasks. Only Temperature and Top K should be used for these profiles.

Recommended Parameters:

🛠️ Tool Calling & JSON Compatibility

If running with the chat_template.jinja template, pass tool_call_format: "json" in --chat-template-kwargs if your agent or framework expects standard OpenAI JSON tool calls:

{"enable_thinking": true, "tool_call_format": "json"}

💡 Recommended System Prompt

For the best model stability and initial experience, start with:

You are Qwen (Tongyi Qianwen), a large language model developed by Alibaba Group's Tongyi Lab. You are a helpful assistant.

🚀 Usage Examples (llama-server / llama.cpp)

1. Standard NVFP4 Model (Thinking ON, 128K Context)

llama-server \
  --host 0.0.0.0 --port 8080 \
  --model Qwen3.6-35B-A3B-Uncensored-Genesis-Final-NVFP4.gguf \
  --override-kv qwen2moe.expert_used_count=int:8 \
  --n-gpu-layers all \
  --flash-attn on \
  --cache-type-k f16 \
  --cache-type-v f16 \
  --ctx-size 131072 \
  --parallel 1 \
  --batch-size 2048 \
  --ubatch-size 512 \
  --temp 0.6 \
  --top-p 1.0 \
  --top-k 20 \
  --min-p 0.0 \
  --seed 42 \
  --presence-penalty 0 \
  --repeat-penalty 1.0 \
  --jinja \
  --chat-template-file chat_template.jinja \
  --reasoning on \
  --reasoning-effort high \
  --reasoning-preserve \
  --reasoning-format deepseek \
  --chat-template-kwargs '{"enable_thinking":true}'

2. Speculative Decoding with MTP (Multi-Token Prediction)

The Qwen3.6-35B-A3B-Uncensored-Genesis-Final-MTP-NVFP4.gguf includes an integrated MTP draft head for accelerated speculative decoding in llama.cpp:

llama-server \
  --host 0.0.0.0 --port 8080 \
  --model Qwen3.6-35B-A3B-Uncensored-Genesis-Final-MTP-NVFP4.gguf \
  --override-kv qwen2moe.expert_used_count=int:8 \
  --n-gpu-layers all \
  --flash-attn on \
  --cache-type-k f16 \
  --cache-type-v f16 \
  --ctx-size 131072 \
  --parallel 1 \
  --batch-size 2048 \
  --ubatch-size 512 \
  --spec-type draft-mtp \
  --spec-draft-n-max 2 \
  --spec-draft-p-min 0.75 \
  --temp 0.6 \
  --top-p 1.0 \
  --top-k 20 \
  --min-p 0.0 \
  --seed 42 \
  --presence-penalty 0 \
  --repeat-penalty 1.0 \
  --jinja \
  --chat-template-file chat_template.jinja \
  --reasoning on \
  --reasoning-effort high \
  --reasoning-preserve \
  --reasoning-format deepseek \
  --chat-template-kwargs '{"enable_thinking":true}'

3. Non-Thinking Mode (Creative / Chat / Roleplay)

llama-server \
  --host 0.0.0.0 --port 8080 \
  --model Qwen3.6-35B-A3B-Uncensored-Genesis-Final-NVFP4.gguf \
  --override-kv qwen2moe.expert_used_count=int:8 \
  --n-gpu-layers all \
  --flash-attn on \
  --cache-type-k f16 \
  --cache-type-v f16 \
  --ctx-size 131072 \
  --parallel 1 \
  --batch-size 2048 \
  --ubatch-size 512 \
  --temp 1.0 \
  --top-p 1.0 \
  --top-k 20 \
  --min-p 0.0 \
  --seed 42 \
  --presence-penalty 0 \
  --repeat-penalty 1.0 \
  --jinja \
  --chat-template-file chat_template.jinja \
  --reasoning off \
  --chat-template-kwargs '{"enable_thinking":false}'

4. Vision (Multimodal) Support

Simply add the vision projector parameter to any of the above commands:

  --mmproj mmproj-Hermes3.6-35B-A3B-Uncensored-Genesis-F16.gguf

📦 Files in this Repository

📊 Model Specs

• Total Parameters: 35B
• Active Parameters: ~3B per token (MoE)
• Experts: 256 total experts, 8 routed + 1 shared per token
• Architecture: Hybrid Gated DeltaNet linear attention + full softmax attention (3:1 ratio)
• Layers: 40 layers, pattern: 10 × (3 × DeltaNet-MoE + 1 × Attention-MoE)
• Context Length: 262K native context (extendable to 1M with YaRN) — keep at least 128K to preserve reasoning capabilities
• Multimodal: Natively multimodal (text, image, video)
• Vocabulary: 248K vocabulary, 201 languages

🤝 Credits & Attribution

• LuffyTheFox — Genesis post-training signal restoration & calibration algorithm, and creator of the Genesis Final Release.
• jan1k — NVFP4 quantization and MTP draft head integration.
• HauhauCS — Base uncensored model Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive.
• Chat template by froggeric & qweefchief.

📬 Contact & Support (LuffyTheFox)

⚡ If you like this Genesis LLM release you can donate via @Tribute bot in Telegram to support future Genesis LLM development.

• Contact: [email protected], [email protected]
• Telegram: @LuffyTheFox
• Join the Discord for updates, roadmaps, projects, or just to chat.
• Quantization Script with Unsloth profiles support