Skip navigation
More open-source model choice is coming to Windows. 💻
0:00 / 0:43

More open-source model choice is coming to Windows. 💻

Windows Developer

@windowsdev511K followers

Visit on X
341 likesDownloadOriginal post

19,801 views

More open-source model choice is coming to Windows. 💻 We’re bringing llama.cpp support to Windows ML, giving developers a faster way to experiment with emerging models and deploy across GPU, NPU + CPU. Learn more:

10 Comments

Comments from the original post on X (Twitter)

  • Pranav Kakde's profile picture

    @im_pranavkakde

    More model choice, local inference, and CPU/GPU/NPU support. 👏 Now if Windows could also stop using 40% of my RAM before I even open a browser, we’d really be cooking. 😂 Local AI is great. Local AI that doesn’t compete with your OS for memory is even better.

    4 likes

  • Mike Koker's profile picture

    @rekokylnoeht

    Llama.cpp has been the default on Windows anyway. Took this long.

    2 likes

  • TattooedTechy's profile picture

    @TattoodTechy

    Good headway.

    1 like

  • CyaVer's profile picture

    @CyaVer009

    After a year or 2, Bloated Windows OS will be a thing of the past cause someone will create an OS debloated of trash software thanks to Claude or ChatGPT

  • Josh Jackson | AI agent security's profile picture

    @JustJorshin

    Useful detail in the docs: the GGUF path currently validates CPU and CUDA, with no NPU target support. Worth checking before trying this on an NPU laptop.

  • Stanlin Dias's profile picture

    @Stanlin_dias

    Running GGUF straight through llama.cpp with no ONNX conversion removes a real annoyance 💻 The docs say it's experimental for now with greedy decoding only. Are sampling settings like temperature and top-p on the roadmap before it's cleared for Store apps?

  • JustViewNews's profile picture

    @justviewnews

    btw llama-manager on windows here

  • Linus | LLMRig's profile picture

    @oldsportlinus

    This is interesting. One of the messier parts of local AI today is figuring out the right model + runtime + hardware path for a given machine. I’ve been running into exactly this while building LLMRig. Curious how much of the backend and hardware selection Windows ML will handle automatically with llama.cpp.

  • the's profile picture

    @thdmit

    Wasn't it already supported?

  • zoomies.jsonl's profile picture

    @snappercayt

    One has to ask Why not windows_ml.cpp Why rely on open source