Best of LoRA — 2025

1. 1
   Article

   Michael Lynch·25w

   My First Impressions of MeshCore Off-Grid Messaging

   MeshCore enables encrypted text messaging over long-range (LoRa) radios without cellular infrastructure, forming mesh networks where devices relay messages through each other. Testing three devices (Heltec v3, SenseCAP T-1000e, and Lilygo T-Deck+) revealed significant usability challenges: confusing onboarding, poor UI/UX design, limited range (only 5 blocks in suburban areas), and critical components being closed-source despite appearing open. While the concept appeals for emergency communication, the technology isn't yet practical for non-technical users or reliable disaster scenarios.

   66

   6
2. 2
   Article

   Daily Dose of Data Science | Avi Chawla | Substack·40w

   Fine-tuning Gemma 3 270M Locally

   Google's Gemma 3 270M model can be fine-tuned locally using just 0.5 GB RAM. The tutorial demonstrates using Unsloth and HuggingFace transformers to fine-tune the model for chess move prediction. The process involves loading the model, configuring LoRA for efficient training, preparing a chess dataset, and training with decreasing loss. After fine-tuning, the model successfully predicts missing chess moves instead of generating random moves.

   28
3. 3
   Article

   Hugging Face·48w

   (LoRA) Fine-Tuning FLUX.1-dev on Consumer Hardware

   QLoRA enables fine-tuning of FLUX.1-dev diffusion models on consumer hardware with under 10GB VRAM by combining 4-bit quantization with Low-Rank Adaptation. The approach uses bitsandbytes for quantization, 8-bit AdamW optimizer, gradient checkpointing, and cached latents to dramatically reduce memory usage from ~120GB to ~9GB. Training on RTX 4090 takes 41 minutes for 700 steps, while FP8 training with torchao on H100 reduces time to 20 minutes. The technique maintains high-quality results while making advanced model customization accessible to developers without enterprise-grade hardware.

   11