Back to home
VLM
3 articles tagged with this topic
LocalLLaMAVLM
450M VLM Hits 44% on Browser Screens After Fine-Tune — Vertical Beats Scale
Reddit dev fine-tunes 450M-param VLM on 50K browser screenshots, lifting screen QA from 1% to 44%. Computer Use bottleneck may not be model size — it'
6d ago2 min read
NVIDIAFLARE
NVIDIA Brings Federated Learning to Multimodal AI — Data Stays Local
NVIDIA FLARE now supports vision-language models, letting hospitals and factories train shared AI without moving data. Not a silver bullet.
Aug 192 min read
TomofunFurbo
Furbo Ditches GPU for AWS Inferentia2: A Real-World AI Inference Cost Win
Tomofun moved Furbo AI inference to AWS Inferentia2 from GPU, cutting costs with no precision loss—validating specialized chips replacing GPUs for con
May 72 min read