MediaPipe Text Summarizer (200M)

Compact 200M-parameter Gemma model quantized with mixed precision (int4 + int8) (summarization_quant_200m_2modes.litertlm, 117.6 MB) fine-tuned for two real-time on-device summarization modes (TLDR 1–2 sentence paragraphs and KEYPOINTS 3–5 bullets) in the MediaPipe Text Summarizer (TextSummarizer) task.

Documentation: MediaPipe Text Summarizer Docs

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

Try live: Browser Web Demo is not yet available for TextSummarizer (use the samples above; see MediaPipe Web Demos).

Select mode via TextSummarizerMode.TLDR (concise 1–2 sentence overview) or TextSummarizerMode.KEYPOINTS (3–5 bulleted takeaways). Peak RAM usage ranges from β‰ˆ200 MB to β‰ˆ350 MB across 100–2,000 word inputs.

Files

File Precision Size SHA-256
summarization_quant_200m_2modes.litertlm β€” 117.59 MB (117,587,968 B) 8b2d4ef09236adb9ead3127325526ba1aa5a59feb7c5de2d3f5958f27479de59

Download

pip install -U huggingface_hub
hf download litert-community/MediaPipe-TextSummarizer-200M summarization_quant_200m_2modes.litertlm --local-dir .

Inputs, Outputs, and Model Specifications

summarization_quant_200m_2modes.litertlm

Self-contained .litertlm package bundling the quantized LiteRT prefill/decode subgraphs, SentencePiece tokenizer, and prompt template metadata in a single memory-mapped file.

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-TextSummarizer-200M", filename="summarization_quant_200m_2modes.litertlm")

BaseOptions = mp.tasks.BaseOptions
TextSummarizer = mp.tasks.text.TextSummarizer
TextSummarizerOptions = mp.tasks.text.TextSummarizerOptions
TextSummarizerMode = mp.tasks.text.TextSummarizerMode

options = TextSummarizerOptions(
    base_options=BaseOptions(model_asset_path=model_path),
    mode=TextSummarizerMode.TLDR,
)
with TextSummarizer.create_from_options(options) as summarizer:
    result = summarizer.summarize("MediaPipe Tasks runs lightweight on-device ML models locally across mobile, web, and desktop.")
    print(result.summary)

Cross-Platform Guides & Samples

Latency Benchmarks

Average Pixel 10 Pro CPU benchmarks across TLDR and KEYPOINTS modes from the Text Summarizer documentation (peak RAM β‰ˆ200 MB–350 MB):

Device Input Words (tokens) Prefill (tokens/s) Decode (tokens/s) E2E Latency (ms)
Pixel 10 Pro CPU 100 (β‰ˆ200 tokens) 980 133 900
Pixel 10 Pro CPU 500 (β‰ˆ900 tokens) 950 130 1,790
Pixel 10 Pro CPU 1000 (β‰ˆ1900 tokens) 690 100 5,000
Pixel 10 Pro CPU 2000 (β‰ˆ3600 tokens) 470 55 12,000–17,000

Intended Use, Limitations, and Fairness

  • Intended applications: Condensing private notes, emails, transcripts, and articles locally into TLDR paragraphs or KEYPOINTS bullets without cloud round-trips.
  • Terms & limitations: Subject to the Gemma Terms of Use. Supported on Android, iOS, Flutter (Android/iOS), and desktop Python (LiteRT-LM backend; not available on Web or Flutter web). Generated summaries should be verified for factual preservation on long or technical texts.

Provenance

Every model binary in this repository is byte-identical (unmodified LiteRT-LM package 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
summarization_quant_200m_2modes.litertlm summarization_quant_200m_2modes.litertlm 1786566818453053 Wed, 12 Aug 2026 20:33:38 GMT /1/

License and Citation

  • License: gemma β€” Subject to the Gemma Terms of Use, as specified on the MediaPipe task documentation page.

Citation

Gemma family model packaged for MediaPipe Text Summarizer (LiteRT-LM).

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

Collection including litert-community/MediaPipe-TextSummarizer-200M