Windows build CUDA Toolkit

[QuantForgeSoftware]

I have the latest versions of msvc and cuda installed and it will not build. It would be helpful if you would update your tools and/or document what version are needed for DFlash which looks really interesting.

[davide221]

Can you paste the actual build output error? Windows support got added in pr#6 against one toolchain combo, once I see the error we can fix it

[cfelicio]

Anyone managed to get a successful Windows build on Github? Everything compiles and runs, but connecting Claude Code or Kilo Code via VS Code does not seem to work for complex prompts with Qwen 3.6 on a RTX 3090. I'm running with

python scripts/server.py --bin bin/test_dflash.exe --target models/Qwen3.6-27B-UD-Q4_K_XL.gguf --port 8080 --ctk tq3_0 --ctv tq3_0

Build I used: https://github.com/cfelicio/lucebox-hub/actions/runs/25122880343

[LeiYangGH]

on 29th Apr 2026 i built on windows using latest code. but performance is no much better than llama.cpp (i my smoke test).
i've put my logs in folk branch

[davide221]

@LeiYangGH instead of spamming would be useful to try to tune your ddtree configurations to see if that help to gain speed.

[davide221]

Closing the issue because the author didn't provide useful specification