[FEEDBACK] Local apps

#31
by kramp - opened
Hugging Face org

Please share your feedback about the Local Apps integration in model pages.

On compatible models , you'll be proposed to launch some local apps:

In your settings, you can configure the list of apps and their order:

The list of available local apps is defined in https://github.com/huggingface/huggingface.js/blob/main/packages/tasks/src/local-apps.ts

I think the tensor-core fp16 FLOPS should be used for GPUs supporting that. I note that V100 counts as way less than the theoretical 125 TFLOPS, listed e.g. here: https://images.nvidia.com/content/technologies/volta/pdf/tesla-volta-v100-datasheet-letter-fnl-web.pdf

Hey! Have you guys heard of LangFlow? It is a neat solution for developing AI-powered apps as well!

The GPU list is missing the RTX A4000 (16GB)

PR: https://github.com/huggingface/huggingface.js/pull/817

Would be nice to get ollama integration

I suggest adding Ollama as local app to run LLM's

I use GPT4All and it is not listed herein

Hi! I'm Danny, CPO of Atomic Chat. We were merged in huggingface/huggingface.js#2076 ( https://github.com/huggingface/huggingface.js/pull/2076) and our app is present in @huggingface/tasks since v0.20.24. However, Atomic Chat doesn't appear on the Settings → Local Apps page (huggingface.co/settings/local-apps). It seems this page uses a separate registry on the Hub frontend side. Could you add Atomic Chat to it? cc @Wauplin @julien-c

For Local Apps, one useful direction could be making the “handoff” metadata explicit: what model artifact is being opened, what local app/version receives it, and what command/config was used.

That matters once local apps move beyond simple model loading into agentic workflows. If a local app starts an agent that can call tools or write files, users will eventually want to answer: which local app launched this run, with which model, and what did it do afterwards?

The feature already feels like a natural bridge between HF models and local developer workflows. Adding a little more launch/run metadata would make it easier to debug and trust those workflows.

The GPU list is missing the Intel Arc Pro B70

EDIT: PR here: https://github.com/huggingface/huggingface.js/pull/2175#pullrequestreview-4309161150

Hugging Face org

@j1mc want to open a quick PR to add it?

Please add https://omlx.ai to local apps

Please add vmlx or MXL studio. https://mlx.studio/download/

Hi,

Missing from the hardware list:

  • NVIDIA RTX™ 5000 Ada Generation Laptop GPU (16GB GDDR6) powers advanced AI with 682 AI TOPS
  • Intel® Core™ i9 processor HX series (14th Gen)

go add it then

Please add vmlx or MXL studio. https://mlx.studio/download/

why not add it yourself?

Sign up or log in to comment