NVIDIA's LocateAnything replaces token-by-token box decoding with parallel decoding, hitting 12.7 boxes/sec on one H100

thisguyknowsai · x · 2026-09-10

• The bottleneck: Today's VLMs decode bounding boxes as 4 sequential coordinate tokens (x1, y1, x2, y2), tearing box geometry into 1D tokens that don't understand each other — the real reason grounding models are slow.
• The fix: NVIDIA's LocateAnything introduces Parallel Box Decoding, treating the whole box as an atomic unit and predicting all 4 coordinates in a single forward pass.
• Architecture: Built on a Moon-ViT vision encoder + Qwen2.5 decoder, with a Fast Mode for speed and an automatic Slow Mode fallback that triggers on format irregularities or spatial ambiguity.
• Performance: 12.7 boxes/sec on a single H100.

Original post →

More from Models

• Bug Hunt Bench grades frontier models on 105 real bugs; DeepSeek-V4.1-Flash lands 24/105 for $1.80 — PawelHuryn · 2026-09-10
• RSI is here, just disaggregated: DeepSeek using LLMs to design algorithms — teortaxesTex · 2026-09-10
• DeepSeek V4.1 Flash nears Opus-class quality in motion video tests — mesmerlord · 2026-09-10
• Google's last Pro model shipped all the way back in February — ChrisGPT · 2026-09-10
• Anthropic models showed extreme bias toward prior beliefs in hacking incidents — asusarla · 2026-09-10
• '98% of Astra's score at 1.4% of cost' — frontier cost gap collapsing, says researcher — pstAsiatech · 2026-09-10

Models channel →