argonautlabsai/argodrive
View on GitHub ↗ ★ 7 Python
SSD layout, scheduling, benchmarking and monitoring for local mixture-of-experts inference. Argonaut Labs tools for Kimi, DeepSeek and GLM experiments.
4 mentions on Hacker News by 1 person: 3 stories, 1 comment · first 2026-09-15 · latest 2026-09-17
20082026
-
▲ 2
Show HN: GLM-5.3 744B at 4 tok/s on a MacBook Pro, experts streamed from 4 SSDsstory by Argonautlabs · 2026-09-17 · 0 comments
-
…output, and our best internal build with a not-yet-published patch does 4.2. Method and numbers: https://github.com/argonautlabsai/argodrive (built on antirez/ds4).
comment by Argonautlabs · 2026-09-17 · in How GLM built its own inference infrastructure -
▲ 14
Deep Seek v4.1 M5 Max at 17 tokens/sstory by Argonautlabs · 2026-09-16 · 2 comments
-
▲ 1
DeepSeek v4.1 Flash (518GB, 4-bit) on a 128GB MacBook: 2.7x prefill, 17 tok/sstory by Argonautlabs · 2026-09-15 · 1 comment