deepseek-v4.1POST
DeepSeek V4.1 Flash on DwarfStar: Metal tuning lifts 18.5 to 31.3 tok/s
A Metal decode-fusion patch raises DeepSeek V4.1 Flash throughput on DwarfStar from 18.5 to 31.3 tok/s, near Unsloth's llama.cpp but still below oMLX's 38.5 tok/s.
きしだൠ(K1S)
@kisArchived 4 days agoOriginally posted 2026.10.26
きしだൠ(K1S) @kisOriginally posted 2026.10.26
DwarfStarのDeepSeek V4.1 Flash、18.5tok/secだったのを、31.3tok/sec、Unsloth版llama.cppと同じくらいに高速化した。 けどoMLXの38.5には届かないな。やはりGGUFの展開が重い https://github.com/kishida/ds4/tree/v41-metal-decode-fusion2
x.com/kis/status/2114440981340926254Archived 2026.10.26 12:04