mirror of
https://github.com/hanzoai/ml.git
synced 2026-11-20 22:46:50 +00:00
Three-way per file through the rename: base 39355c6c91, theirs the upstream
merge of main with 35d7ae7c (via 7ef6a694d), ours origin/main.
Lands the barrier thread pool, runtime CPU feature dispatch, aarch64 and x86
repacked quantized kernels (indexed_gemv, gemv_fused_shared_lhs), quantized
embedding on CPU/Metal/CUDA, paged flash-attn kernels, cutile fused MoE, the
in-place op aliasing API, parking_lot storage locks, Metal buffer builders and
debug labels, safetensors 0.8, hf-hub 1.0, tokenizers 0.23.
Ours keeps ROCm, Vulkan, wgpu, the IQ/MXFP4/NVFP4/TQ dtypes, and the i-quant
weight/activation block split in matmul. Our backends ride InplaceOpN and the
quantized embedding; Metal get_rows gains the IQ kernels. ug leaves, as upstream.
Co-authored-by: Hanzo Dev <[email protected]>
Running BERT with ML and WASM
Here, we provide two examples of how to run Bert using a ML-compiled WASM binary and runtime.
Vanilla JS and WebWorkers
To build and test the UI made in Vanilla JS and WebWorkers, first we need to build the WASM library:
sh build-lib.sh
This will bundle the library under ./build and we can import it inside our WebWorker like a normal JS module:
import init, { Model } from "./build/m.js";
The full example can be found under ./lib-example.html. All needed assets are fetched from the web, so no need to download anything.
Finally, you can preview the example by running a local HTTP server. For example:
python -m http.server
Then open http://localhost:8000/lib-example.html in your browser.