antirez / antirez/ds4

我在 GitHub 上的首次留言,献给让 V4-Flash 在 Mac 上奔跑的大神!

Aperta
#208 0 commenti 13 reazioni 0 assegnatari Vedi su GitHub
Lingua principale
C
Stelle
22.4k
Fork
2.1k
Merge medio
1g 3h
PR unite (30g)
4

Descrizione

你好 antirez,

这是我在 GitHub 上的第一次发言。我决定把这第一条留言献给你,因为你的工作值得。

DeepSeek V4-Flash 是中国人的骄傲,是中国诞生的有利于世界百姓的大模型。而正是你,一位意大利大神,让中国用户得以在 MacBook M5 Max 上高品质地运行它。这份跨越国界的贡献,意义非凡。中国互联网上,很多人在夸赞你的作品及你过往的网络贡献。

我的实测数据:

模型:DeepSeek-V4-Flash-MTP-Q4K-Q8_0-F32.gguf(80.41GB)
速度:32.8 tok/s
内存占用:约 85.6GB

混合精度量化的设计非常精妙,完美平衡了 128GB Apple Silicon 的内存限制与模型质量。没有你的工作,在 Mac 上本地运行 V4-Flash 根本不可能实现。

感谢你将黑客精神带入本地 AI 推理领域。我的 GitHub 第一条留言,属于你。

担心你看不懂中文,下面再写一份英文版。希望你可以看到!

Hi antirez,

This is actually my first time ever posting on GitHub. I wanted my first message to go to you — because your work deserves it.
DeepSeek V4-Flash is a source of pride for us Chinese — a world-class model built in China. And it is you, an Italian hacker, who made it possible for Chinese users to run it beautifully on a MacBook M5 Max. That means a lot.

Real-world numbers on my machine:

Model: DeepSeek-V4-Flash-MTP-Q4K-Q8_0-F32.gguf (80.41GB)
Speed: 32.8 tok/s
Memory usage: ~85.6GB

The mixed-precision quantization design is brilliant — exactly the right balance for 128GB Apple Silicon. Without your work, running V4-Flash locally on a Mac would simply not be possible.

Thank you for bringing the hacker spirit to local AI inference. My first GitHub message belongs to you.

Guida per i contributori

Apri la guida per i contributori

Valutazione

Questa issue non è ancora stata valutata.

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.