license: apache-2.0 base_model:

• AEON-7/DFlash-Qwen3.5-27B-Uncensored base_model_relation: quantized tags:
• 27b
• 4-bit
• aarch64
• abliterated
• aeon
• aeon-7
• agentic
• arm64
• awq
• blackwell
• block-diffusion
• chat
• chunked-prefill
• coding
• conversational
• dflash
• dgx-spark
• english
• flashinfer-cutlass
• fp4
• fp8-kv-cache
• function-calling
• gated-deltanet
• gb10
• gdn
• gpu
• grace-blackwell
• hybrid
• hybrid-attention
• instruct
• linear-attention
• long-context
• modelopt
• multimodal
• native-fp4
• nvfp4
• openai-api
• openai-compatible
• prefix-caching
• production-ready
• quantized
• qwen
• qwen3
• qwen3.5
• reasoning
• refusal-removed
• safetensors
• sm_121a
• speculative-decoding
• text-generation
• thinking
• tool-calling
• uncensored
• unfiltered
• vision
• vision-language
• vllm
• weight-quantization model_type: qwen3_5 quantization: nvfp4 language:
• en library_name: transformers pipeline_tag: text-generation

DFlash Qwen3.5-27B Uncensored — NVFP4

AEON Qwen — Supreme Being of the Digital Cosmos

27B hybrid linear-attention model | NVFP4 quantized | Vision + Text | DFlash speculative decoding

Performance (DGX Spark GB10)

Quick Links

Quick Start (DGX Spark)

1. Download the model

huggingface-cli download AEON-7/DFlash-Qwen3.5-27B-Uncensored-NVFP4 \
  --local-dir ~/models/DFlash-Qwen3.5-27B-Uncensored-NVFP4

2. Create your environment file

# Auto-generate API key and create .env
cat > .env.dflash << 'EOF'
# Authentication
HF_TOKEN=hf_your_token_here
VLLM_API_KEY=$(openssl rand -hex 32)

# Model path
MODEL_HOST_PATH=~/models/DFlash-Qwen3.5-27B-Uncensored-NVFP4

# DFlash speculative decoding (auto-downloads drafter on first run)
DFLASH_DRAFTER=z-lab/Qwen3.5-27B-DFlash
DFLASH_NUM_SPEC_TOKENS=15

# DGX Spark optimal settings (64K context, 4 concurrent sequences)
MAX_MODEL_LEN=65536
MAX_NUM_SEQS=4
GPU_MEMORY_UTILIZATION=0.85
MAX_NUM_BATCHED_TOKENS=65536
EOF

# Generate a real API key and inject it
sed -i "s|\$(openssl rand -hex 32)|$(openssl rand -hex 32)|" .env.dflash
echo "Your API key: $(grep VLLM_API_KEY .env.dflash | cut -d= -f2)"

3. Save docker-compose.dflash.yml

services:
  vllm-dflash:
    image: ghcr.io/aeon-7/vllm-dflash:latest
    container_name: vllm-dflash
    restart: unless-stopped
    network_mode: host
    ipc: host
    volumes:
      - ${MODEL_HOST_PATH}:/models/DFlash-Qwen3.5-27B-Uncensored-NVFP4
      - dflash-drafter-cache:/models/drafter-cache
    environment:
      - MODEL_PATH=/models/DFlash-Qwen3.5-27B-Uncensored-NVFP4
      - SERVED_MODEL_NAME=DFlash-Qwen3.5-27B-Uncensored
      - DFLASH_DRAFTER=${DFLASH_DRAFTER}
      - DFLASH_NUM_SPEC_TOKENS=${DFLASH_NUM_SPEC_TOKENS}
      - GPU_MEMORY_UTILIZATION=${GPU_MEMORY_UTILIZATION}
      - MAX_MODEL_LEN=${MAX_MODEL_LEN}
      - MAX_NUM_SEQS=${MAX_NUM_SEQS}
      - MAX_NUM_BATCHED_TOKENS=${MAX_NUM_BATCHED_TOKENS}
      - NVIDIA_VISIBLE_DEVICES=all
      - TORCH_MATMUL_PRECISION=high
      - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
      - HF_TOKEN=${HF_TOKEN}
      - VLLM_API_KEY=${VLLM_API_KEY}
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

volumes:
  dflash-drafter-cache:

4. Launch

docker compose --env-file .env.dflash -f docker-compose.dflash.yml up -d

# Watch startup (~5 min for weight loading + CUDA graph compilation)
docker compose -f docker-compose.dflash.yml logs -f

5. Test

# Text generation
curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $(grep VLLM_API_KEY .env.dflash | cut -d= -f2)" \
  -d '{
    "model": "DFlash-Qwen3.5-27B-Uncensored",
    "messages": [{"role": "user", "content": "Explain quantum entanglement simply."}],
    "max_tokens": 200
  }'

# Vision (image understanding)
curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $(grep VLLM_API_KEY .env.dflash | cut -d= -f2)" \
  -d '{
    "model": "DFlash-Qwen3.5-27B-Uncensored",
    "messages": [{"role": "user", "content": [
      {"type": "image_url", "image_url": {"url": "https://upload.wikimedia.org/wikipedia/commons/thumb/3/3a/Cat03.jpg/1200px-Cat03.jpg"}},
      {"type": "text", "text": "What do you see?"}
    ]}],
    "max_tokens": 200
  }'

Environment Variables

Performance (DGX Spark GB10)

DFlash Speculative Decoding (Measured)

Throughput Scaling

Baseline (No Speculation)

What Makes This Model Special

Why Dense Over MoE

Qwen3.5 comes in two flavors: the 122B-A10B MoE (256 experts, 10B active per token) and this 27B dense model (all parameters active on every token). The dense model has real advantages:

• Higher quality per FLOP — Every one of the 27B parameters contributes to every token. MoE models route to a sparse subset, which means some experts are undertrained and routing decisions introduce noise. Dense models don't have this problem.
• No routing overhead — MoE models spend compute on expert selection, load balancing, and all-to-all communication. Dense models just run the computation.
• Predictable latency — No variance from different experts being selected per token. Every forward pass costs the same.
• Simpler deployment — No expert parallelism concerns, no load imbalance, fits on a single GPU with NVFP4.

The tradeoff has always been speed: a 27B dense model moves all parameters through memory per token. On a memory-bandwidth-limited device like DGX Spark (273 GB/s), that meant 12 tok/s baseline. DFlash changes this entirely.

Why DFlash Makes Dense Practical on DGX Spark

The fundamental bottleneck on DGX Spark is memory bandwidth. At 273 GB/s, loading 20 GB of NVFP4 weights per token limits you to ~12 tok/s. Every dense model hits this wall.

DFlash block-diffusion speculative decoding breaks through it:

1. The 2B drafter proposes multiple tokens simultaneously — one diffusion forward pass generates an entire block of speculative tokens in parallel, not sequentially. This costs roughly the same as generating a single token.
2. The 27B target verifies all proposed tokens in one forward pass — instead of paying the full memory bandwidth cost per token, you pay it once and produce 3-4 accepted tokens on average.
3. Net effect: you amortize the bandwidth cost across multiple tokens per forward pass.

The result on DGX Spark:

This makes the 27B dense model faster than the 122B MoE on a single DGX Spark while delivering the quality advantages of a dense architecture. DFlash turns the DGX Spark from "it can run a 27B model" into "it runs a 27B model well."

Hybrid Architecture

Qwen3.5-27B uses a hybrid architecture mixing two attention types across 64 layers:

• Linear attention (GDN) — Gated Delta Network layers for efficient long-context processing with O(1) per-token state (48 layers)
• Full attention — Standard multi-head attention every 4th layer for global context capture (16 layers)

This gives near-linear scaling with sequence length while maintaining full-attention quality at key intervals.

Vision + Text

The model includes a 27-layer ViT vision encoder (460M params, BF16) with a merger that projects visual features into the language model's hidden space. Supports image understanding alongside text generation.

DFlash Block-Diffusion Speculative Decoding

z-lab/Qwen3.5-27B-DFlash is a 2B block-diffusion drafter that generates all speculative tokens simultaneously in a single diffusion step (not sequentially like standard speculative decoding). The 27B target model then verifies in one pass, achieving 2-5x speedup with zero quality loss.

Key difference from standard spec decode: drafting cost is ~constant regardless of token count (one diffusion forward pass), so the tradeoff is purely about verification overhead vs acceptance rate.

AWQ_FULL Quantization

This model uses the most thorough NVFP4 quantization pipeline available:

1. AWQ_FULL — Exhaustive grid search with alpha_step=0.1 across 10 scaling factors per layer, plus a second awq_clip pass that optimizes clipping ratios
2. Full NVFP4 Quantization — All attention projections (Q/K/V/O) and all MLP layers (gate/up/down) quantized to FP4. Excludes: vision tower, embeddings, norms, and lm_head
3. Pre-quantization scales — Channel-wise BF16 factors that redistribute weight magnitudes before quantization

Model Details

NVFP4 Weight Format

Each quantized layer stores:

• weight (uint8) — packed FP4 E2M1 pairs (16-element blocks)
• weight_scale (float8_e4m3fn) — per-block scale (1 per 16 elements)
• weight_scale_2 (float32) — per-tensor global scale
• pre_quant_scale (bfloat16) — AWQ per-channel pre-scaling factors
• input_scale (float32) — static activation scale from calibration

Optimization Stack

Alternative Deployment Methods

vLLM (Manual)

vllm serve AEON-7/DFlash-Qwen3.5-27B-Uncensored-NVFP4 \
  --quantization modelopt \
  --speculative-config '{"method": "dflash", "model": "z-lab/Qwen3.5-27B-DFlash", "num_speculative_tokens": 15}' \
  --attention-backend flash_attn \
  --kv-cache-dtype auto \
  --gpu-memory-utilization 0.80 \
  --max-num-batched-tokens 8192 \
  --max-num-seqs 8 \
  --enable-chunked-prefill \
  --enable-prefix-caching \
  --trust-remote-code

Note: DFlash uses non-causal attention, which requires --kv-cache-dtype auto (BF16). FP8 KV cache is incompatible with DFlash.

SGLang

python -m sglang.launch_server \
  --model-path AEON-7/DFlash-Qwen3.5-27B-Uncensored-NVFP4 \
  --speculative-algorithm DFLASH \
  --speculative-draft-model-path z-lab/Qwen3.5-27B-DFlash \
  --speculative-num-draft-tokens 16 \
  --tp-size 1 \
  --attention-backend fa3 \
  --mem-fraction-static 0.75 \
  --mamba-scheduler-strategy extra_buffer \
  --trust-remote-code

Credits

• Base model by Qwen Team
• DFlash speculative decoding by z-lab (paper)
• Abliteration using llm-abliteration
• NVFP4 quantization with NVIDIA ModelOpt
• Release by AEON-7

Legal Disclaimer

THIS MODEL IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND. This model has had safety alignment removed. Users are responsible for ensuring ethical and legal use.

☕ Support the work

If this release has been useful, tips are deeply appreciated — they go directly toward more compute, more models, and more open releases.

Ethereum L2s (Base, Arbitrum, Optimism, Polygon, etc.) and EVM-compatible tokens can be sent to the same Ethereum address.