Koan-Sin Tan
Koan-Sin Tan
@mohitmundhragithub: users in the whole China market mostly don't have access to the Google Play. It seems that putting a release APK on GitHub is a "must".
- can we get an AAB and convert it back to APK or APK(s)? NOT to do this. - just use the the APK we generated before generative AAB, usually,...
when should we update the play store version? - merging of every PR - per release (backend updates, major bug fixes, etc.).
* for Android one, let's try to have a running app on Pixel phones with LLM (3B or 8B) running.
cf #979 for list of current devices
I was able to convert 3B on colab with 56 GiB (CPU instance + high RAM). Thus I guess you over-estimate the memory requirements. My impression is that 64 GiB...
@freedomtan to share the dynamic int8 tflite model later
@Mostelk let's discuss with Scott first (in the group meeting this week, to check available resources).
@freedomtan to check if he can run the script provided by @farook-edev on his personal Colab account.
@farook-edev llama 1b and 3b dynamic range int8 quantized by ai-edge-torch, https://drive.google.com/drive/folders/1ImWzf-Az5L_GrvZ2pZ21fxpJmdsH9oJm?usp=share_link