- MXFP4/MXFP8量子化 — ローカルLLM実行の新常識
MXFP4/MXFP8はLLM向け新4ビット量子化標準。120Bパラメータモデルが1枚の80GB GPUで動作、llama.cpp/vLLMが対応。
3 min🇯🇵 日本語 - MXFP4/MXFP8 Quantization — Running LLMs Locally
MXFP4/MXFP8 is the new 4-bit quantization standard for LLMs. Enables 120B-param models on a single 80GB GPU via llama.cpp and vLLM.
3 min🇺🇸 English - 2026年の主要ニューラルプロセッシングユニット
Apple ANE、Intel NPU、AMD XDNA、Qualcomm Hexagon、Google TPU、Samsung Exynos、MediaTek APU — 主要NPUを総まとめ。
3 min🇯🇵 日本語 - Training on Apple Neural Engine with Private APIs
ANE project reverse-engineers Apple Neural Engine private APIs to train transformers directly on ANE — no CoreML, no GPU, no Metal.
3 min🇺🇸 English - Training trên Apple Neural Engine qua private API
Dự án ANE reverse-engineer private API Apple Neural Engine cho phép training transformer trực tiếp trên ANE — không CoreML, không GPU, không Metal.
3 min🇻🇳 Tiếng Việt - vLLM — High-Performance LLM Serving Engine
vLLM is the leading open-source LLM inference engine with PagedAttention, continuous batching, FP4 quantization — up to 24x faster than naive transformers.
2 min🇺🇸 English - vLLM — 高性能LLM推論エンジン
vLLMはPagedAttention、continuous batching、FP4量子化を備えたオープンソースLLM推論エンジン。単純なtransformersと比べ最大24倍高速。
3 min🇯🇵 日本語 - vLLM — Serving Engine cho LLM hiệu suất cao
vLLM là open-source inference engine cho LLM với PagedAttention, continuous batching, FP4 quantization — nhanh hơn 24 lần so với naive transformers.
2 min🇻🇳 Tiếng Việt
Quay lại