BeeMonitor Pro (Gemma 4 E2B, fine-tuned)

On-device assistant for the BeeMonitor Android app (Paduravia): beekeeping in Romania and the BeeMonitor hive sensor box.

  • Base model: Gemma 4 E2B instruct (Google), licensed Apache 2.0.
  • Fine-tuned with LoRA on Romanian beekeeping Q&A, then exported to .litertlm (int8 weights) to run on the phone GPU/CPU with LiteRT-LM.
  • File: beemonitor-pro.litertlm (~5.2 GB, phones with 8 GB RAM or more).
  • 2026-10-24: the file now also contains the audio encoder, audio adapter and end-of-audio sections of the official gemma-4-E2B-it.litertlm (litert-community), copied unchanged (~103 MB). The app uses them for Romanian dictation and transcription of recordings, on the phone. File size ~5.2 GB.
  • 2026-10-23: replaced the earlier int4 file (~2.4 GB). That export quantized the per-layer embedding projections to 4 bits as well and the model produced wrong answers on device; the int8 export answers correctly (tested on a phone with the BeeMonitor app).

This is a derivative work of Gemma 4, distributed under the Apache License 2.0. It is an AI model and can be wrong; it does not replace a veterinarian and does not give medication doses.

Downloads last month
33
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for vnicoara22/beemonitor-pro

Finetuned
(494)
this model