Instructions to use litert-community/Laya-Multilingual-LiteRT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LiteRT
How to use litert-community/Laya-Multilingual-LiteRT with LiteRT:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Laya
How to use litert-community/Laya-Multilingual-LiteRT with Laya:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Commit History
25 days ago
-
Hexagon NPU next to GPU FP32: fp16-safe graphs (exact in fp32) and the Android NPU path
Verified signatureThe S256 and S512 graphs (same file names) gain three rewrites that keep fp16 finite and give identical results in fp32: a power-of-two pre-scale in each large LayerNorm (eps scaled by the square), -1e4 attention masks instead of -1e9, and a power of two moved from the GeGLU product into the output projection in layers 11-12. Galaxy S26, LiteRT 2.2.0, 201 rows: NPU (Hexagon v81, JIT) 36.5 ms per question, argmax 81/81, max probability difference 0.0069; GPU FP32 51.9 ms, 0.0014. The Android sample selects GPU, NPU or CPU and packages the Qualcomm libraries through scripts/fetch_npu_libs.sh. Card and SHA256SUMS updated.32f1b84
-
Super-squash branch 'main' using huggingface_hub
eae8f62