WHAT NEW IN X10.29 THU

AI collections from X · 22 collections · 58,925 posts

deepseek-v4.1VIDEO

Tide LM Rust rewrite: DeepSeek V4.1 hits 31.8 tok/s on two DGX Sparks

After rewriting its engine in Rust, Tide LM runs native MXFP4 DeepSeek V4.1 on two DGX Sparks at 31.8 tok/s on prose and about 38 tok/s in real use.

nunomo
@nunomo1Archived 2 days agoOriginally posted 2026.10.26
Tide LM Rust rewrite: DeepSeek V4.1 hits 31.8 tok/s on two DGX Sparks
nunomo @nunomo1Originally posted 2026.10.26
Open original on X

Tide LM Update: - Engine rewritten by Rust. Token/s is boosted to 31.8 tok/s on prose. In real use case about 38 tok/s. (DS v4.1) For two DGX sparks. Two options: - DeepSeek v4.1 native MXFP4 - GLM 5.3 Full EXL3 https://github.com/raspy135/tideLM

x.com/nunomo1/status/2114418988025594116Archived 2026.10.26 12:03