deepseek-v4.1IMAGE
Tuning DeepSeek V4.1 Flash 1017 performance on portable AI compute boxes
The author is tuning six AI compute boxes in parallel: two for DeepSeek V4.1 Flash 1017, two for GLM 5.3 Flash FP4 local, and two for Qwen3.8 Flash Next FP4 QAD quantization.
Andy Stewart
@manateelazycatArchived 11 days agoOriginally posted 2026.10.18
Andy Stewart @manateelazycatOriginally posted 2026.10.18
上高铁了，开始干活 1. 前两台懒猫AI算力舱正在优化 DeepSeek V4.1 Flash 1017 的性能 2. 中间两台懒猫AI算力舱， 正在测试 Nvidia 发布的 GLM 5.3 Flash FP4 Local 的性能 3. 最后两台懒猫AI算力舱， 正在测试 Qwen3.8 Flash Next 新的 Nvidia FP4 QAD 量化性能 看看今天能不能把这三个模型的性能再提升许多？
This post quotes another post: x.com/manateelazycat/status/2108849375623700649
x.com/manateelazycat/status/2111629407769297123Archived 2026.10.18 16:26