toktapethe record, not a recording of it

midagedev

https://github.com/midagedev

3 runs, newest first

ik_llama.cpp · V4.1 Flash Q3_K_M · 3090 + A6000 · coding review 507 in / 1500 out · DSpark n_max=3 28.8 tok/sprefill 156 tok/s

DeepSeek-V4.1-Flash-Q3_K_M-00001-of-00009.gguf

ik_llama.cpp81af2772Q3_K_M748B · 16B activeP507 · G1500 · cache 0%ctx 16koffload partialrtx-3090rtx-a6000linux1 stream! 2 caveats

midagedev

2026-09-30

Qwen2.5-7B-Instruct Q3_K_M · 4 streams · dense control 134 tok/sprefill 4159 tok/s

qwen2.5-7b-instruct

ik_llama.cpp7cd62a3eQ3_K_M7.6BP5665 · G934 · cache 0% · min 703ctx 8kfa onrtx-3090rtx-a6000linux4 streams! 3 caveats

midagedev

2026-09-20

DeepSeek-V4.1-Flash Q3_K_M · 1 stream · 347 GB model on 72 GB of VRAM 23.8 tok/sprefill 52.3 tok/s

DeepSeek-V4.1-Flash-Q3_K_M-00001-of-00009.gguf

llama-serverb102Q3_K_M748B · 18B activeP6509 · G66 · cache 0%ctx 8koffload partialrtx-3090rtx-a6000linux1 stream! 3 caveats

midagedev

2026-09-20

Page views are counted with Cloudflare Web Analytics: no cookies, no fingerprinting. The toktape binary itself sends nothing anywhere.