toktapethe record, not a recording of it

midagedev

https://github.com/midagedev

Narrowed toengine: llama-server ×gpu: rtx-3090 ×os: linux ×Clear all

4 runs, newest first

Qwen3.8-Flash-Next UD-Q4_K_XL · 1 stream · mlock 51.0 tok/s

Qwen3.8-Flash-Next-UD-Q4_K_XL-00001-of-00004.gguf

llama-serverb0UD-Q4_K_XLrtx-3090rtx-a6000linux1 stream

2026-09-18

DeepSeek-V4.1-Flash Q3_K_M · Korean prose, no thinking 20.3 tok/s

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

llama-serverb96Q3_K_M rtx-a6000rtx-3090linux1 stream ! 2 caveats

2026-09-18

DeepSeek-V4.1-Flash Q3_K_M · 4 streams 27.4 tok/s

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

llama-serverb96Q3_K_M rtx-a6000rtx-3090linux4 streams ! 1 caveat

2026-09-18

DeepSeek-V4.1-Flash Q3_K_M · 1 stream · 347 GB model on 72 GB of VRAM 28.6 tok/s

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

llama-serverb96Q3_K_M rtx-a6000rtx-3090linux1 stream ! 1 caveat

2026-09-18