FlashAttention v2 - Where the 1.7x Speedup Actually Comes From

Two weeks ago I built FlashAttention v1 from scratch in Triton and got 22% of peak bandwidth on an A10G. The v1 to v2...

Building FlashAttention from Scratch on an A10G - What the Numbers Actually Say

I spent the last two weeks building FlashAttention from scratch in Triton. Not to use it in production - vLLM already ships a better one....

How to save file to external storage in Android 10 and Above

Since Android 10, Environment.getExternalStoragePublicDirectory method was deprecated, then you was unable access public external storage directory directly to save your file. This short post...

Inspect network traffic on Android

As an Android developer, you might want to inspect network traffic in your Android device. With the apps you build, in fact, there are a...

Thread lifecycle in Java

As an Android developer, you have to work directly or indirectly with threads in your apps. So understanding Thread Lifecycle in Java very important.