作者

Brian

Staff Technical Writer

2 posts

Brian educates on using software and tools effectively. He is pragmatic in each of his articles and provides value in terms of work that can actually be done today.

$GGUF, GPTQ, AWQ, EXL2 quantization formats compared: how model weights, runtime overhead, and KV cache stack up in memory$

AI 与机器学习

GGUF、GPTQ、AWQ、EXL2:LLM量化格式究竟如何使用内存

对比GGUF、GPTQ、AWQ和EXL2的内存使用情况,从Q4_K_M文件大小到KV缓存增长和运行时开销。

Brian Jul 2, 2026 12 分钟阅读

Unified memory explained: discrete GPU memory requires a copy across PCIe between system RAM and VRAM, while unified memory is one shared pool the CPU and GPU both access directly

AI 与机器学习

什么是统一内存,为什么它能让一台迷你主机运行235B模型?

统一内存让紧凑型AI主机能够加载单块24-32GB GPU都装不下的235B级模型。它是什么、为什么有效,以及为什么更大不等于更快。

Brian Jul 2, 2026 11 分钟阅读