GitHub - Luce-Org/lucebox-hub: Lucebox optimization hub: hand-tuned LLM inference, built for specific consumer hardware.

8 min read Original article ↗

Lucebox

lucebox.com Discord Blog

MIT CUDA 12+ C++17

Open LLM inference, rewritten by hand for one specific chip at a time.
Kernels, speculative decoding, and quantization, tailored per target.
We don't wait for better silicon. We rewrite the software.


Inside the box

Two projects today, more coming. Each one is a self-contained release with its own benchmarks and paper-style writeup.

Megakernel    DFlash 27B


01 · Megakernel Qwen3.5 0.8B on RTX 3090

The first megakernel for hybrid DeltaNet/Attention LLMs. All 24 layers of Qwen 3.5-0.8B in a single CUDA dispatch, 1.87 tok/J on a 2020 GPU, matching Apple's latest silicon at 2× the throughput.

# 1. clone + enter
git clone https://github.com/Luce-Org/lucebox-hub && cd lucebox-hub/megakernel

# 2. install (Python 3.10+, CUDA 12+, PyTorch 2.0+). Weights stream from HF on first run.
python -m venv .venv && source .venv/bin/activate   # required on Ubuntu 24+ system Python (PEP 668)
pip install --upgrade pip
pip install torch                          # install BEFORE the next step; setup.py imports torch at build time
pip install -e . --no-build-isolation      # --no-build-isolation lets the build see the torch you just installed

# 3. run the benchmark (prefill pp520 + decode tg128 vs llama.cpp BF16 + PyTorch HF)
python final_bench.py
Method Prefill pp520 Decode tg128 tok/J
Megakernel @220W 37,800 413 1.87
llama.cpp BF16 @350W 11,247 267 0.76
PyTorch HF 7,578 108 n/a

What makes it work: 82 blocks, 512 threads, one persistent kernel. No CPU round-trips between layers. Weights streamed straight from HuggingFace. Cooperative grid sync instead of ~100 kernel launches per token. Power ceiling hit before compute ceiling, so DVFS converts tight execution straight into saved watts.

Full writeup → · Benchmarks → · Blog post →

Blackwell (RTX 5090, DGX Spark / GB10): auto-detected by setup; NVFP4 decode path lands ~194 tok/s tg128 on GB10. See megakernel/README.md#blackwell-sm_120--sm_121a.


02 · DFlash DDtree Qwen3.5 27B GGUF on RTX 3090

First GGUF port of DFlash speculative decoding. Qwen3.5-27B on a single RTX 3090, Q4_K_M target + BF16 draft, DDTree budget=22.

  • Up to 207 tok/s in the demo (207.6 tok/s DFlash vs 38.0 tok/s AR, 5.46×)
  • 129.5 tok/s mean on the HumanEval 10-prompt bench
  • 3.43× faster than autoregressive (+15% over chain speculative decoding)
  • 2.8× faster than SGLang AWQ on the same hardware
  • Up to 256K context in 24 GB via TurboQuant TQ3_0 KV cache (128K Q4_0 bench: 134.78 tok/s at ctx=131072)
# 1. clone with submodules (pulls the pinned Luce-Org/llama.cpp@luce-dflash fork)
git clone --recurse-submodules https://github.com/Luce-Org/lucebox-hub && cd lucebox-hub/dflash

# 2. build the C++/CUDA decoder (CUDA 12+, CMake 3.18+)
# Default compiles for 75/80/86/89 (+120 on CUDA 12.8+, +sm_121/DGX Spark on CUDA 12.9+, +sm_110/Thor on CUDA 13.0+) so the binary runs on every supported card.
# 3090-only users can add -DCMAKE_CUDA_ARCHITECTURES=86 to skip the other archs and build faster (~3 min).
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release
cmake --build build --target test_dflash -j

# 3. fetch weights: ~16 GB Q4_K_M target + 3.46 GB bf16 draft
huggingface-cli download unsloth/Qwen3.5-27B-GGUF Qwen3.5-27B-Q4_K_M.gguf --local-dir models/
huggingface-cli download z-lab/Qwen3.5-27B-DFlash model.safetensors --local-dir models/draft/

# 4a. one-shot streaming generate
python3 scripts/run.py --prompt "def fibonacci(n):"

# 4b. or reproduce the paper-style bench (HumanEval + GSM8K + Math500, ~15 min)
python3 scripts/bench_llm.py
Benchmark AR (tok/s) DFlash+DDTree (tok/s) Speedup
HumanEval 37.8 129.5 3.43×
Math500 37.7 110.5 2.93×
GSM8K 37.7 96.2 2.55×

The constraint that shaped the project. AWQ INT4 of Qwen3.5-27B plus the BF16 draft doesn't leave room for the DDTree verify state on a 24 GB card. Q4_K_M GGUF (~16 GB target) is the largest format that fits target + 3.46 GB draft + budget=22 tree state + KV cache in 24 GB on the RTX 3090. Picking it forced a new port on top of ggml, since no public DFlash runtime supports a GGUF target.

What we built vs what we didn't. The algorithms are not ours:

  • DFlash (z-lab, 2026): block-diffusion draft conditioned on target hidden states.
  • DDTree (Ringel et al., 2026): tree-structured verify that beats chain verify at the same compute budget.

What we ported and tuned:

  • C++/CUDA decode engine on top of ggml (no libllama, no Python runtime, Q4_K_M target path).
  • Three custom CUDA kernels for tree-aware SSM state rollback: ggml_ssm_conv_tree, ggml_gated_delta_net_tree, ggml_gated_delta_net_tree_persist.
  • DDTree budget swept for RTX 3090 + Q4_K_M target: budget=22 is the sweet spot.
  • TQ3_0 KV cache (TurboQuant 3.5 bpv, default) + sliding target_feat ring to fit up to 256K context in 24 GB (Q4_0 available as legacy, tops out near 128K).

Running on other GPUs (4090, 5090, DGX Spark / GB10, Jetson AGX Thor)

Supported out of the box; the build just needs the right CUDA toolkit. dflash/CMakeLists.txt already auto-adds Blackwell archs when your nvcc is new enough, so the main quickstart above works as-is on newer cards.

GPU Arch Min CUDA Status
RTX 3090 Ampere sm_86 12.0 reference, all numbers above
RTX 4090 Ada sm_89 12.0 should work, unverified, pass -DCMAKE_CUDA_ARCHITECTURES=89
RTX 5090 Blackwell consumer sm_120 12.8 supported, auto-added by CMake
DGX Spark / GB10 sm_121 (compute capability 12.1) 12.9 supported, auto-added by CMake
Jetson AGX Thor sm_110 13.0 supported, auto-added by CMake

Verify your target:

python -c "import torch; p=torch.cuda.get_device_properties(0); print(p.name, 'sm_%d%d'%(p.major,p.minor), p.multi_processor_count,'SMs', round(p.total_memory/1e9,1),'GB')"
nvcc --version

DGX Spark / GB10 quick start:

# CUDA 12.9+ required for sm_121
nvcc --version  # must show >= 12.9
git clone --recurse-submodules https://github.com/Luce-Org/lucebox-hub && cd lucebox-hub/dflash
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release   # CMake auto-adds sm_121
cmake --build build --target test_dflash -j

Jetson AGX Thor quick start:

# CUDA 13.0+ required for sm_110 / AGX Thor.
nvcc --version
git clone --recurse-submodules https://github.com/Luce-Org/lucebox-hub && cd lucebox-hub/dflash
cmake -B build -S . -DCMAKE_BUILD_TYPE=Release   # CMake auto-adds the Thor arch your nvcc supports
cmake --build build --target test_dflash -j

What will NOT auto-port:

  • DDTree budget=22 tuned for 3090 + Q4_K_M + 24 GB. On cards with more VRAM (5090 32 GB, GB10 128 GB unified), re-sweep, larger tree = more verify throughput until memory bandwidth saturates. scripts/bench_llm.py has the sweep hooks.
  • TQ3_0 KV cache + sliding target_feat ring was shaped by 24 GB (fits up to 256K context on a 3090). On GB10 (128 GB unified) / 5090 (32 GB) you can push context further or skip quantization entirely and keep F16 KV.
  • Perf numbers (207 tok/s demo, 129.5 HumanEval, 2.8× vs SGLang AWQ) are RTX 3090 @ stock. Blackwell/Ada not yet swept, PRs with RESULTS.md entries welcome.

Full writeup → · Benchmarks → · Blog post →

Qwen3.6-27B (supported, experimental draft): same qwen35 architecture, so the 3.6 Q4_K_M GGUF loads as a drop-in target. With z-lab's matched Qwen3.6-27B-DFlash draft (still under training, 2026-04-26 snapshot), HumanEval lands at ~78 tok/s (AL 5.05); the 3.5 draft gets ~74 tok/s. 3.5↔3.5 reference is 129.5 tok/s. AL should improve as z-lab finishes training the draft. Details in dflash/README.md.


Why this exists

Local AI should be a default, not a privilege: private data, no per-token bill, no vendor lock-in. The hardware to run capable models already sits on desks. The software to run those chips well doesn't.

General-purpose frameworks dominated the last decade because hand-tuning kernels per chip was too expensive to justify. One stack, decent on everything, great on nothing. Most of the silicon's capability stays on the floor.

AI-assisted development flips that calculus. Rewrites that took a quarter now fit in a release cycle. Lucebox is where we publish them, one chip and one model family at a time. MIT source, full writeup, reproducible benchmarks.


Requirements

All experiments in this repo are built, tuned, and benchmarked on NVIDIA RTX 3090 (2020), the reference target. Supported GPU families:

  • Ampere (sm_86, RTX 3090 / A-series): reference, CUDA 12+.
  • Ada (sm_89, RTX 40xx): should work, unverified, CUDA 12+.
  • Blackwell consumer (sm_120, RTX 50xx incl. 5090): supported, CUDA 12.8+.
  • DGX Spark / GB10 (sm_121, compute capability 12.1): supported, CUDA 12.9+.
  • Jetson AGX Thor (sm_110): supported, CUDA 13+.

PyTorch 2.0+. dflash/ needs CMake 3.18+ and --recurse-submodules for the pinned Luce-Org/llama.cpp@luce-dflash fork (three tree-mode ggml ops); multi-arch build is automatic (see Running on other GPUs).

Megakernel porting note. Tighter than dflash: megakernel/setup.py pins -arch=sm_86 -DNUM_BLOCKS=82 (3090 SM count). To run on a different card, edit both defines and pip install -e . --force-reinstall --no-deps. Grid is persistent, one block per SM, so NUM_BLOCKS must match exactly. Suggested starting points: 4090 sm_89 + 128, 5090 sm_120 + 170, DGX Spark / GB10 sm_121 + run torch.cuda.get_device_properties(0).multi_processor_count to read SM count, Jetson AGX Thor sm_110 + the same SM-count query.

Optional, find your GPU's sweet spot: sudo nvidia-smi -pl 220 (megakernel hits best tok/J at 220 W on 3090; re-sweep for other cards).


Repository layout

lucebox-hub/
├── megakernel/    · fused forward pass for Qwen 3.5-0.8B
├── dflash/        · DFlash speculative decoding port for Qwen 3.5-27B on RTX 3090
└── assets/        · banners, cards, diagrams

Roadmap

  Q1 2026    ▮▮▮▮▮▮▮▮▮▮    RTX 3090 kernels & optimizations
  Q2 2026    ▮▮▮▮▮▯▯▯▯▯    Ryzen AI MAX+ 395 optimizations
  Q2 2026    ▮▮▯▯▯▯▯▯▯▯    Heterogeneous CPU + GPU latency optimizations
  Q2 2026    ▮▯▯▯▯▯▯▯▯▯    Lucebox OS for local AI machines
  Q3 2026    ▯▯▯▯▯▯▯▯▯▯    Lucebox official launch

Citation

@software{lucebox_2026,
  title  = {Lucebox: Open LLM Inference, Rewritten by Hand for One Specific Chip at a Time},
  author = {Lucebox},
  url    = {https://github.com/Luce-Org/lucebox-hub},
  year   = {2026}
}

Per-project citations live in each subproject's README.


Inspired by


Community


MIT · Lucebox.com