Running a vector-search on Android with wgpu (Rust API). The compute time is not good and the shader needs to be improved. I would also test on other mobile phones with better GPU cores.

SmolChat is now available on Google Play in 170+ countries and in two languages, English and simplified Chinese.

SmolChat allows users to download LLMs and use them offline, with a clean and easy-to-use interface.

Why does L1 regularization lead to sparse models? Shubham Panchal's newest article explains the mechanics behind L1's feature selection power, connecting it to the concept of covariance and the way gradient descent updates parameters.

Having recently started learning iOS app development (with KMP), I wanted to study how Apple computers, macOS and iOS evolved to their present states. I compiled all my findings and research in a blogpost, which explains the 'Apple' computing stack.

Blog: shubham0204.github.io/blogpost/pro...

[Project-Update] Sentence-Embeddings-Android now supports model2vec sentence embeddings in Kotlin.

A sentence embedding generated by a model, like, all-MiniLM-L6-v2 is obtained after pooling the encoder hidden states of a BERT-style transformer along the sequence-length axis.

Microsoft has released a new repo 'typescript-go' that provides a native-port to the TypeScript (TS) compiler.

The current TS compiler 'tsc' translates TS code to optimized JavaScript (JS) code which then executes in one of the available JS runtimes (Node, V8 etc.).

✨ How do you search for images given a textual query, just like Google Photos?

On-device inference for OpenAI's CLIP using GGUF, enabling image-text similarity & zero-shot classification.

GitHub: github.com/shubham0204/...

#android #softwaredevelopment #programming #machinelearning #ondeviceml