Hans

For now, I'm prereleased [here](https://www.npmjs.com/package/@fugood/transformers) I will try my best to keep up to date with the latest version. P.S. Our company is heavily on React Native and has some...

> Also, to help test, would you like to create a demo model on the HF Hub which contains all these different variants? Here is my convertion: https://huggingface.co/BricksDisplay/stablelm-2-1_6b-q4 https://huggingface.co/BricksDisplay/stablelm-2-1_6b-bnb4 https://huggingface.co/BricksDisplay/phi-1_5-q4...

For benchmark, the quantized `all-MiniLM-L6-v2` model is https://huggingface.co/BricksDisplay/all-MiniLM-L6-v2-q4 https://huggingface.co/BricksDisplay/all-MiniLM-L6-v2-bnb4

> I think it would be good to also name/suffix these models slightly differently, so you can export (and use) multiple versions. Did you mean the model file name like...

> Yes, exactly :) I think this PR can help standardize which suffixes we should use (as quantized will be overloaded). It might be worthwhile aligning this with other libraries'...

That is because `sharp` no prebuild binaries for your platform, yout should build from source Or You could use [overrides](https://docs.npmjs.com/cli/v8/configuring-npm/package-json#overrides) to force `sharp>=0.33.0` And install [WASM binary](https://sharp.pixelplumbing.com/install#webassembly)

In my debug. The compute shaders for Q4_K and Q5_K are unsupported on Qualcomm Adreno. Without these, it will work. For more info: Failed shaders are matmul_q4_k_f32_l matmul_q4_k_f32_m matmul_q4_k_f32_s matmul_q4_k_f32_aligned_l...

https://hans00.me 這個 WebView 是放在左下一個小區塊，當時其它地方是有放東西的 (忘記是放什麼了

@xenova Seems `interpolate_data` is incorrect. Will get  My implement 

It roughly done, except for performance issues with `onnxruntime-react-native`.