UNICORN WHO DEV
← Back to catalogue
Models / Unicorn Who Dev

UWD LiteRT Android Models

A UWD collection of twenty-five LiteRT variants prepared for Android dataset production, including fifteen trainable heads or adapters with frozen visual encoders.

Open the public source
Overview

A UWD collection of twenty-five LiteRT variants prepared for Android dataset production, including fifteen trainable heads or adapters with frozen visual encoders. The record preserves published qualification states and does not turn functional checks into accuracy or on-device performance claims.

Variants
25
Trainable
15
App passes
5
Source
Hugging Face
Status
Connected source
Public dossier

Documentation and scope

Structured content from the public source documentation.

01

Overview

A UWD collection of twenty-five LiteRT variants prepared for Android dataset production, including fifteen trainable heads or adapters with frozen visual encoders.

The record preserves published qualification states and does not turn functional checks into accuracy or on-device performance claims.

Visual referenceUWD LiteRT Android Models — On-device vision
02

Documented scope

  • 25 active variants
  • 15 trainable heads or adapters
  • Weights kept outside the APK
  • 5 application passes, 1 RTMDet failure and 19 pending variants in the published status
  • Contracts, hashes and reports retained per variant
  • No task accuracy or physical ARM-phone qualification claimed
03

Place in the catalogue

UWD LiteRT Android Models has a dedicated record connected to its current source.

04

Encoders and classification

  • caformer_s18_learning
  • convformer_s18
  • convformer_s18_learning
  • dinov2
  • dinov2_learning
  • edgenext_small_usi_learning
  • edgenext_x_small_learning
  • edgenext_xx_small_learning
  • efficientformer_l1_learning
  • hgnetv2_b0
  • hgnetv2_b0_learning
05

Detection, structure and pose

  • repvit_m1
  • repvit_m1_learning
  • efficientvit_sam
  • florence2
  • rfdetr
  • rfdetr_learning
  • rtmdet_tiny
  • rtmdet_tiny_learning
  • table_transformer_detection_learning
  • table_transformer_structure_learning
  • vitpose
  • vitpose_learning
06

Vision-language

  • tinyclip
  • tinyclip_learning
07

Provenance and limits

The information is linked to the stated public source. Metrics reflect a reference snapshot and may change between synchronizations.

Provenance

A record traceable to its source

Editorial information, metadata and documentation remain connected to the original public repository.

Public identifier
Unicorn Who Dev/UWD LiteRT Android Models
Latest activity
September 20, 2026
Read the complete documentation
Project sources

Every source consolidated in this record

Public sources remain individually accessible. Private components are named without exposing their repository, revision or content.

01Lite_rt_prepared_for_android_dataset_builderHugging Face
Technologies
litertandroidcomputer-visionon-device-learningtflite