lambsea/Qwen3.6-27B-AEON-Ultimate-Uncensored-UD-GGUF

Qwen3.6-27B-AEON-Ultimate-Uncensored — GGUF (UD Quants)

Unsloth Dynamic-style (UD) GGUF quantizations of AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16

Every quant uses per-tensor overrides (sensitivity-driven) + importance matrix (multi-domain calibration). All SSM recurrence tensors are preserved at source precision. MTP speculative decoding and vision (mmproj) are preserved.

---

Quant Comparison

File	Quant	Size	tg t/s	PPL	KL mean	KL max	KL p99.9
F16	F16	50.9 GB	30.9	5.7215	—	—	—
UD-Q8_0	Q8_0	34.7 GB	44.7	5.7055	0.0029	5.23	0.22
UD-Q6_K	Q6_K	30.6 GB	49.1	5.7046	0.0045	7.64	0.33
UD-Q5_K_M	Q5_K_M	28.7 GB	46.9	5.7589	0.0111	5.09	2.04
UD-IQ4_XS	IQ4_XS	25.9 GB	56.5	5.7630	0.0236	4.53	1.97

Benchmarked on NVIDIA RTX PRO 6000 Blackwell (96 GB VRAM), llama.cpp fork (a4501150/llama.cpp), pp=512, tg=128.

---

What Makes These Different

SSM Recurrence Preservation

Qwen3.6 is a hybrid GatedDeltaNet + attention model. 48 of 64 layers use a recurrent SSM where quantization error compounds across token positions. All SSM recurrence tensors are preserved at source precision (F16) — never quantized.

Tensor	Count	Precision	Rationale
ssm_alpha, ssm_beta	96	F16	State update projections — error accumulates in recurrence
ssm_out	48	F16	Output projection feeds directly into residual stream
ssm_a, ssm_conv1d, ssm_dt, ssm_norm	192	F32	Small state tensors (llama-quantize keeps 1D/small tensors at F32)
attn_qkv (SSM input projection)	48	F16	Highest measured KL sensitivity
attn_gate (SSM gate projection)	48	F16	Second-highest measured KL sensitivity

Per-Tensor Sensitivity Analysis

Each tensor group was probed by quantizing only that group to Q4_0 while keeping the rest at F16, then measuring KL divergence. The override generator assigns precision based on measured sensitivity:

Precision	Tensor Groups	Override Count
F16	SSM recurrence, norms, biases, MTP layer	512
F16	All attention tensors (attn_qkv, attn_gate, attn_v, attn_q, attn_k, attn_output), ffn_down edge	173
Base quant	FFN middle layers, FFN edge gate/up, embeddings	~181

685 total overrides — every non-FFN tensor has an explicit precision assignment. No dependence on llama-quantize's internal promotion rules.

Multi-Domain Calibration + GPU Imatrix

Calibrated on a balanced mix across 4 domains from 13 HF datasets:

Domain	Token Budget	Sources
General	1M	ultrachat, OpenHermes, COIG-CQIA, LongAlpaca, pg19, froggeric/imatrix
Code	750K	Magicoder-Evol-Instruct-110K
Reasoning	750K	OpenMathInstruct-2, OpenR1-Math-220k
Agentic	500K	glaive-function-calling-v2, xlam-function-calling-60k, hermes-function-calling-v1

Special tokens from source datasets are stripped automatically. Samples are kept whole — never truncated mid-conversation.

The importance matrix is generated with a PyTorch GPU-native generator (src/generate_imatrix.py) at 65,536 context — uses forward hooks to accumulate squared activations on GPU with zero PCIe D2H copies. Supports multi-GPU via device_map="auto".

Per-domain imatrices are merged with equal weights (DI-MATRIX approach).

MTP + Vision Preserved

• MTP (Multi-Token Prediction): Draft head (blk.64) pinned at F16. Use --spec-type draft-mtp --spec-draft-n-max 3 for ~1.5-2x faster generation.
• Vision: mmproj file contains the full vision encoder. Use --mmproj flag with llama-server for image/video understanding.

---

Files

File	Description	Size
Qwen3.6-27B-AEON-UD-Q8_0.gguf	Highest quality quantization	34.7 GB
Qwen3.6-27B-AEON-UD-Q6_K.gguf	Recommended — best quality/size	30.6 GB
Qwen3.6-27B-AEON-UD-Q5_K_M.gguf	Balanced	28.7 GB
Qwen3.6-27B-AEON-UD-IQ4_XS.gguf	Smallest, for constrained VRAM	25.9 GB
Qwen3.6-27B-AEON-mmproj-F16.gguf	Vision encoder (use with --mmproj)	885 MB
imatrix_merged.dat	Importance matrix for requantization	13 MB

---

Usage

llama-server (recommended)

# Q6_K with YaRN 512k context, 5 concurrent slots, MTP + vision
llama-server \
    -m Qwen3.6-27B-AEON-UD-Q6_K.gguf \
    --mmproj Qwen3.6-27B-AEON-mmproj-F16.gguf \
    -ngl 99 \
    --flash-attn \
    -c 524288 \
    --parallel 5 \
    --cache-type-k q8_0 \
    --cache-type-v q8_0 \
    -kvu \
    --cache-ram -1 \
    --rope-scaling yarn \
    --rope-scale 2.0 \
    --yarn-orig-ctx 262144 \
    --override-kv "qwen35.context_length=int:524288" \
    --spec-type draft-mtp \
    --spec-draft-n-max 3 \
    --jinja \
    --chat-template-kwargs '{"enable_thinking":true,"preserve_thinking":true}' \
    --host 0.0.0.0 --port 8080

Note: --spec-type draft-mtp requires llama.cpp b9375+. A custom fork adds DFlash speculative decoding and Blackwell-tuned flash attention.

llama-cli

llama-cli \
    -m Qwen3.6-27B-AEON-UD-Q6_K.gguf \
    -ngl 99 \
    --flash-attn \
    -c 524288 \
    --rope-scaling yarn \
    --rope-scale 2.0 \
    --yarn-orig-ctx 262144 \
    --jinja \
    --chat-template-kwargs '{"enable_thinking":true,"preserve_thinking":true}'

Chat Template Notes

• enable_thinking activates reasoning mode (chain-of-thought in <think> blocks)
• preserve_thinking retains reasoning blocks in conversation history
• No spaces after colons in the JSON — Qwen3.6's template parser is whitespace-sensitive

---

Architecture

Qwen3.6-27B is a hybrid SSM-attention model:

• 64 transformer layers + 1 MTP layer (blk.0-64)
• 48 SSM layers (GatedDeltaNet, no KV cache) + 16 full attention layers (every 4th layer)
• 27B parameters, 24 attention heads, 4 KV heads, head dim 256
• Vocab: 248,320 tokens, native context: 262,144 tokens

---

Quantization Pipeline

Built with super-quant:

1. Convert HF to F16 GGUF (with MTP tensors) + mmproj GGUF (vision)
2. Multi-domain calibration data from 13 HF datasets, special tokens stripped
3. GPU-native importance matrix generation (PyTorch, 65k context) + weighted merge
4. Per-tensor sensitivity analysis (KL divergence probing against F16 logits)
5. Hybrid override generation — SSM at source precision, sensitivity-driven for the rest
6. Quantize with per-tensor overrides + imatrix
7. Benchmark: throughput + perplexity + KL divergence vs F16

Key Differences from Previous Release

Aspect	Previous	Current
Calibration	Had <|endoftext|> token leak (2,628 occurrences)	Special tokens stripped automatically
Imatrix context	32,768	65,536
Imatrix generator	llama-imatrix (slow — PCIe D2H per tensor per chunk)	PyTorch GPU-native (zero D2H during generation)
Sensitivity analysis	llama.cpp subprocesses, ~600 GB disk I/O per run	PyTorch in-place weight perturbation, zero disk I/O
SSM alpha/beta	F32 (beyond source precision)	F16 (matches source BF16)
SSM out	Q8_0 (quantized)	F16 (source precision preserved)
Attention tensors	Mixed (f16/q8_0/q6_k)	All F16 (sensitivity-confirmed)
Norms/biases	Implicit (llama-quantize internal rules)	Explicit F16 overrides
Total overrides	339	685

---

Links

• Base model: AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16
• Quantization pipeline: super-quant
• llama.cpp fork: a4501150/llama.cpp (DFlash, MTP fixes, Blackwell FA4)

Credits

• Base model: AEON-7
• Architecture: Qwen Team
• Quantization: llama.cpp
• Sensitivity methodology inspired by APEX quant research
• Calibration datasets: HuggingFaceH4, teknium, NousResearch, nvidia, open-r1, Salesforce, glaiveai, froggeric

---

License: Apache-2.0