Mar 28 · 7 min read · Your LoRA Is Synthetic Garbage (And Why That Matters More Than You Think) Let me stop you right here: If you trained a LoRA on AI-generated images, you wasted your time and your GPU budget. This isn't a hot take. It's a hard lesson learned in the CGA...
Join discussion
Mar 24 · 10 min read · Not all fine-tuning is the same. The method you pick determines your compute bill, your results, and whether your model forgets everything it already knew. Fine-tuning is not a single button you press
Join discussion
Mar 9 · 14 min read · TLDR: Full fine-tuning updates every model weight, which is expensive in memory, compute, and storage. PEFT methods update only a small trainable slice. LoRA learns low-rank adapters on top of frozen base weights. QLoRA pushes efficiency further by q...
Join discussionMar 9 · 15 min read · TLDR: Fine-tuning a 7B-parameter LLM updates billions of weights and requires expensive GPUs. LoRA (Low-Rank Adaptation) freezes the original weights and trains only tiny adapter matrices that are added on top. 90%+ memory reduction; zero inference l...
Join discussion
Feb 10 · 3 min read · In my journey to dive deeper into multimodal AI systems, I decided to fine-tune BLIP-2, a powerful vision-language model trained on the Flickr8k dataset to generate image captions. What made this more exciting was integrating LoRA (Low-Rank Adaptatio...
Join discussion
Jan 22 · 1 min read · This is an excerpt. Read the full technical analysis: https://www.vladavramut.com/articles/lora-meshtastic-encryption.html Encryption protects payload secrecy. It does not hide metadata, routing behavior, or traffic patterns. What Encryption Does •...
Join discussionJan 22 · 1 min read · This is an excerpt. Read the full technical analysis: https://www.vladavramut.com/articles/meshtastic-hardware-stability.html Hardware is not the weak link because radios fail. It becomes the weak link when design assumptions ignore real operational...
Join discussionJan 22 · 1 min read · This is an excerpt. Read the full technical analysis: https://www.vladavramut.com/articles/meshtastic-security-threat-model.html Security is not a feature checkbox. It is a system property shaped by threat assumptions and operational context. Real ...
Join discussion
