MediaPipe Decision Maker — GLiNER2.5-Decide (256 & 512 Token)

GLiNER2.5-Decide (DeBERTa-v3-large encoder) bidirectional cross-encoder bundles (gliner_s256.task, 981.2 MB; gliner_s512.task, 1.08 GB) converted from fastino/GLiNER2.5-Decide for the MediaPipe Decision Maker (DecisionMaker) task.

Documentation: MediaPipe Decision Maker Docs

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

Try live: Web Demo

Packages the FP16-weight (wfp16) DeBERTa-v3-large classification encoder and head converted from fastino/GLiNER2.5-Decide in litert-community/GLiNER2.5-Decide-LiteRT.

Jointly attends across the input context and up to 32 label markers ([L]) in a single cross-attention pass within a 256- or 512-token window.

Files

File Precision Size SHA-256
gliner_s256.task — 981.22 MB (981,216,352 B) 18e8eb07a06e3833c4ebb4b0a32c76bff0e1197acba0e4c1658b121d01c92674
gliner_s512.task — 1.08 GB (1,081,879,648 B) 23a02e1b286fab7573ef2a00acbd8bf80fd2b1284de25eb8fafcd74f8bccd7ec

Companion Files in Repo Root

File Role Size SHA-256
LICENSE Apache-2.0 license text (from GLiNER2.5-Decide-LiteRT) 11.4 KB (11,357 B) c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4
NOTICE Component attributions & license notices 2.7 KB (2,745 B) 00870472fb48c9617aa2272f8aca7998aefbe4f3d036e15979de57a2998de4a5

Download

pip install -U huggingface_hub
hf download litert-community/MediaPipe-DecisionMaker-GLiNER2.5-Decide gliner_s256.task --local-dir .
hf download litert-community/MediaPipe-DecisionMaker-GLiNER2.5-Decide gliner_s512.task --local-dir .

Inputs, Outputs, and Model Specifications

Bundle Contents (gliner_s256.task)

Bundled Member Uncompressed Size SHA-256 (prefix)
model.tflite 710.72 MB (710,715,520 B) 00922aba05428737…
tokenizer.json 8.33 MB (8,333,952 B) 3ad87d9ffe669147…
token_embeddings_fp16.bin 262.17 MB (262,166,528 B) f90f5d837721b3d4…

Sub-model model.tflite

Direction Tensor Shape Type Notes
Input serving_default_args_0 [1, 256, 1024] float32 —
Input serving_default_args_1 [1, 256] float32 —
Input serving_default_args_2 [1, 32, 256] float32 —
Output serving_default_output_0_output [1, 1, 1, 32] float32 —

Bundle Contents (gliner_s512.task)

Bundled Member Uncompressed Size SHA-256 (prefix)
model.tflite 811.38 MB (811,378,816 B) bbe150fa6f7bf29c…
tokenizer.json 8.33 MB (8,333,952 B) 3ad87d9ffe669147…
token_embeddings_fp16.bin 262.17 MB (262,166,528 B) f90f5d837721b3d4…

Sub-model model.tflite

Direction Tensor Shape Type Notes
Input serving_default_args_0 [1, 512, 1024] float32 —
Input serving_default_args_1 [1, 512] float32 —
Input serving_default_args_2 [1, 32, 512] float32 —
Output serving_default_output_0_output [1, 1, 1, 32] float32 —

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-DecisionMaker-GLiNER2.5-Decide", filename="gliner_s256.task")

BaseOptions = mp.tasks.BaseOptions
DecisionMaker = mp.tasks.decision.DecisionMaker
DecisionMakerOptions = mp.tasks.decision.DecisionMakerOptions
ChoiceQuestion = mp.tasks.decision.ChoiceQuestion

options = DecisionMakerOptions(
    base_options=BaseOptions(model_asset_path=model_path),
    max_num_tokens=256,
)
with DecisionMaker.create_from_options(options) as maker:
    q = ChoiceQuestion(
        criteria={
            "refund_request": "Customer wants money back or a refund.",
            "technical_support": "Customer reports a broken or defective product.",
        },
        normalize_prior=True,
    )
    res = maker.evaluate_choice("Headphones arrived with a cracked case; please refund me.", q)
    print("Choice:", res.selected_key, res.probabilities)

Cross-Platform Guides & Samples

Latency Benchmarks

The MediaPipe Decision Maker documentation does not include a standalone latency table. On the source wfp16 graphs in litert-community/GLiNER2.5-Decide-LiteRT, Galaxy S26 GPU (FP32) opening-request latency measures 174.9 ms (back-to-back median 198.8 ms) at s256 and 575.2 ms (back-to-back median 733.1 ms) at s512 (graph only).

Intended Use, Limitations, and Fairness

  • Intended applications: English zero-shot single-label choice routing, boolean gating, and multi-task schema classification via MediaPipe DecisionMaker with cross-attention between context and labels.
  • Limitations & license notices: English only; schema + text must fit within 256 or 512 encoded tokens (max 32 label slots per graph). Entity/span extraction heads from GLiNER2 are not included. See bundled LICENSE (Apache-2.0) and NOTICE (includes MIT notice for DeBERTa-v3-large).

Provenance

Every model binary in this repository is byte-identical (unmodified weights, metadata, and .task bundle archive 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
gliner_s256.task gliner_s256.task 1791242665856875 Mon, 05 Oct 2026 23:24:25 GMT /1/
gliner_s512.task gliner_s512.task 1791242707069212 Mon, 05 Oct 2026 23:25:07 GMT /1/

License and Citation

  • License: apache-2.0 — Distributed under Apache License, Version 2.0 (see bundled LICENSE and NOTICE files for component attributions).

Citation

@misc{zaratiana2025gliner2efficientmultitaskinformation,
  title         = {GLiNER2: An Efficient Multi-Task Information Extraction System with Schema-Driven Interface},
  author        = {Urchade Zaratiana and Gil Pasternak and Oliver Boyd and George Hurn-Maloney and Ash Lewis},
  year          = {2025},
  eprint        = {2507.18546},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2507.18546}
}
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for litert-community/MediaPipe-DecisionMaker-GLiNER2.5-Decide

Quantized
(15)
this model

Collection including litert-community/MediaPipe-DecisionMaker-GLiNER2.5-Decide

Paper for litert-community/MediaPipe-DecisionMaker-GLiNER2.5-Decide