Meituan-AutoML / MobileVLM

Strong and Open Vision Language Assistant for Mobile Devices
Apache License 2.0
958 stars 65 forks source link

MobileVLM weights? #1

Closed pHaeusler closed 7 months ago

pHaeusler commented 8 months ago

Awesome work!

Do you plan on releasing the weights for MobileVLM?

cxxgtxy commented 8 months ago

Yes. We have updated weights of MobileLLaMA.

Foul-Tarnished commented 8 months ago

Is there any android app that uses this ? Sherpa (llama2 7b) run slowly, and even crash with input/long context with 12gb (+4gb swap) on SD865 ...

Edit : Apparently there's this more updated fork of Sherpa https://github.com/Mobile-Artificial-Intelligence/maid/

huyiming2018 commented 8 months ago

Is there any android app that uses this ? Sherpa (llama2 7b) run slowly, and even crash with input/long context with 12gb (+4gb swap) on SD865 ...

Edit : Apparently there's this more updated fork of Sherpa https://github.com/Mobile-Artificial-Intelligence/maid/

Based on our experience, it is normal for the 7B model to have slow inference on mobile devices. We suggest reducing the model size or using higher specification equipment.