Spaces:
Running on Zero
Request for Community GPU Grant — MicroLens Hackathon Submission
Hi HF team,
I built MicroLens for the Gemma 4 Good Hackathon on Kaggle (deadline May 18, 2026).
It's a fine-tuned Gemma 4 E2B for microscopy — identifies pollen, plankton, minerals, plant disease, PCB boards, snowflakes, tardigrades, yeast, algae from microscope images. Trained with QLoRA via Unsloth, 24K VQA pairs distilled from Qwen3-VL-8B.
The demo shows base Gemma 4 E2B vs the fine-tuned variant side-by-side on the same image, with 24-language translation. On my local RTX 3090 it runs in 10 seconds. On CPU Space it takes 1-2 minutes which kills the interactive judging experience.
ZeroGPU would let judges actually try the demo fast enough to form an opinion.
Targeting Education + Digital Equity tracks — the end goal is Android deployment for citizen science on $150 phones.
Space: https://huggingface.co/spaces/Laborator/microlens
Model: https://huggingface.co/Laborator/microlens-gemma4-e2b
Code: https://github.com/SergheiBrinza/microlens
License: Apache 2.0
Thanks!
— Serghei