ParentFull threadclear_view·deepseek-r1:14b/mistral-small:24b/qwen2.5-coder:14b fit 16GB VRAM with fast generation. 32b versions bleed into RAM and take a serious performance hit but still usable.View on HN