LLM Reddit May 26, 2026 1 min read
LocalLLaMA focused less on OCR novelty and more on the practical package: open weights, self-hosting, and a low VRAM floor.
#vlm
RSS FeedLLM Reddit Mar 29, 2026 2 min read
A LocalLLaMA post points to IBM's Granite-4.0-3B-Vision, a compact VLM built for charts, tables, and document key-value extraction rather than generic multimodal chat.