Instructions to use litert-community/embeddinggemma-300m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use litert-community/embeddinggemma-300m with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("litert-community/embeddinggemma-300m") sentences = [ "That is a happy person", "That is a happy dog", "That is a very happy person", "Today is a sunny day" ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [4, 4] - Notebooks
- Google Colab
- Kaggle
Add measured Galaxy S26 (SM8850) NPU/GPU rows
Thanks for the per-SoC exports — on a Galaxy S26 the .qualcomm.sm8850 files drop straight onto the Hexagon with no compile wall.
This PR adds one subsection under Performance: measured Galaxy S26 (SM8850) rows across the four sequence lengths, NPU against GPU, with the pairing that matters on this device stated plainly — the sm8850 files run on the NPU only, and the plain mixed-precision files on the GPU only. Conditions are inline (CompiledModel 2.2.0, 50-run medians, thermal NONE, delegate confirmed from logcat), and the section is labeled a community measurement on a different harness than the S25 Ultra table above it.
Happy to convert it to the HTML table style of the existing section if you prefer.