How to use from the
Use from the
LiteRT library
# 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

MediaPipe Universal Sentence Encoder (Text Embedder)

Compact dual-encoder text embedding model (universal_sentence_encoder.tflite, 6.1 MB, FP32) mapping text inputs to 100-dimensional dense semantic vectors for rapid on-device cosine similarity and topical clustering in the MediaPipe Text Embedder (TextEmbedder) task.

Documentation: MediaPipe Text Embedder Docs

Samples: Android · iOS · Python Colab · Web · Flutter

Try live: Web Demo

Uses a compact dual-encoder architecture trained on question-answer datasets ([-1, 100] output vector). Captures topical similarity rather than sentiment polarity (for example, "I like my phone" and "I hate my phone" produce high similarity because both concern phones).

Files

File Precision Size SHA-256
universal_sentence_encoder.tflite float32 6.12 MB (6,120,274 B) 89ad3c74175dd8caa398cc22b657296d94302d20c525c12b58b29420f7249749

Download

pip install -U huggingface_hub
hf download litert-community/MediaPipe-TextEmbedder-Universal-Sentence-Encoder universal_sentence_encoder.tflite --local-dir .

Inputs, Outputs, and Model Specifications

universal_sentence_encoder.tflite

  • Embedded TFLITE_METADATA: name USE-QA TFLite model · version 1 · author MediaPipe
Direction Tensor Shape Type Notes
Input ParseExample/ParseExampleV2:1 [-1] string inp_text — Input query text
Input ParseExample/ParseExampleV2:2 [-1] string res_context — Input query text
Input ParseExample/ParseExampleV2:3 [-1] string res_text — Input query text
Output Final/EncodeQuery/mul [-1, 100] float32 query_encoding — query-response similarity.
Output Final/EncodeResult/mul [-1, 100] float32 response_encoding — query-response similarity.

Use with MediaPipe Tasks

Python Quickstart

pip install -U mediapipe huggingface_hub
import mediapipe as mp
from huggingface_hub import hf_hub_download

model_path = hf_hub_download(repo_id="litert-community/MediaPipe-TextEmbedder-Universal-Sentence-Encoder", filename="universal_sentence_encoder.tflite")

BaseOptions = mp.tasks.BaseOptions
TextEmbedder = mp.tasks.text.TextEmbedder
TextEmbedderOptions = mp.tasks.text.TextEmbedderOptions

options = TextEmbedderOptions(base_options=BaseOptions(model_asset_path=model_path), quantize=True)
with TextEmbedder.create_from_options(options) as embedder:
    e1 = embedder.embed("It's a charming and affecting journey.").embeddings[0]
    e2 = embedder.embed("What a great and fantastic trip!").embeddings[0]
    print("Cosine similarity:", TextEmbedder.cosine_similarity(e1, e2))

Cross-Platform Guides & Samples

Latency Benchmarks

Average CPU pipeline latency reported by the Text Embedder documentation:

Model Variant Device CPU Latency
Universal Sentence Encoder Galaxy S26 Ultra (4 threads) 10 ms

Intended Use, Limitations, and Fairness

  • Intended applications: Lightweight on-device semantic similarity, FAQ/retrieval matching, and topical grouping where minimal size (6.1 MB) and ≈10 ms CPU latency are prioritized over 768D LLM embeddings.
  • Limitations & considerations: Dual-encoder trained on QA pairs; for multilingual or instruction-steered retrieval/classification tasks, consider MediaPipe-TextEmbedder-EmbeddingGemma-300M. No standalone Google fairness PDF is published for this model on the MediaPipe documentation page.

Provenance

Every model binary in this repository is byte-identical (unmodified FlatBuffer weights and embedded TFLITE_METADATA bytes) to the official artifact served by Google from storage.googleapis.com/mediapipe-models.

Repo File Upstream URL GCS Generation Upstream Last-Modified Pinned Bucket Version
universal_sentence_encoder.tflite universal_sentence_encoder.tflite 1682480025114200 Wed, 26 Apr 2023 03:33:45 GMT /1/

License and Citation

  • License: apache-2.0 — Distributed under Apache License, Version 2.0 alongside the MediaPipe release artifacts.

Citation

Architecture linked from the MediaPipe Text Embedder documentation: aclanthology.org/2022.emnlp-main.640.pdf.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including litert-community/MediaPipe-TextEmbedder-Universal-Sentence-Encoder