한국어EN
On-Device & Local Inferencellama.cpp communityllama.cppCommunity

llama.cpp Discussion #23853 shares an NVFP4/MXFP6 mixed-precision tool

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Discussion #23853 — released a tool for NVFP4/MXFP6 mixed precision + imatrix per-layer optimization. Comparisons shared in the discussion found NVFP4 faster but below Q4_K_M in quality

Why it matters

Real measurements showing hardware-native FP4 has not yet beaten community quantization. The format race is unresolved

Source — Primary sources first; secondary reports are labelled “Press”

Officialgithub.comhttps://github.com/ggml-org/llama.cpp/discussions/23853

Added in Week 35, 2026 · See that week

On the map — ⑲ On-Device line

  1. Previous stopLiteRT-LM
  2. This recordllama.cpp Discussion #23853 shares an NVFP4/MXFP6 mixed-precision tool
  3. Next stopApple expands the Foundation Models framework at WWDC 2026
See it on the map

Same week

NVIDIA's Vera Rubin NVL72 enters volume productionOfficialNVIDIA releases the Isaac GR00T open humanoid reference designOfficialAnthropic raises $65 billion in its Series HOfficialGuni Sharon publishes Tree of Thoughts as a Classical Heuristic Search ProblemPaper

Nearby on the same axis

Ollama 0.30 adds llama.cpp-based GGUF compatibility to its MLX engineOfficialGoogle makes LiteRT-LM generally available (GA)OfficialApple expands the Foundation Models framework at WWDC 2026PressChrome 148 stable opens the Prompt API to the webOfficial