David Soušek
sousekd
AI & ML interests
Running AI on-prem with agents processing confidential data. Supporting government, private sector, and all kinds of weirdos.
Recent Activity
liked a model 14 days ago
deepseek-ai/DeepSeek-V4-Flash-0731 liked a model 16 days ago
lowbitcoffee/GLM-5.2-W4A16 liked a model about 1 month ago
tencent/Hy3-FP8Organizations
None yet
Context Length for 2X6000 Pros (2x96 = 192GB VRAM)
5
#2 opened 4 months ago
by 
mtcl
Q4_X on Blackwell + EPYC + DDR 5
👍🔥 4
13
#3 opened 4 months ago
by sousekd
Nex-N2-Pro
1
#11 opened 2 months ago
by sousekd
MTP
4
#2 opened 3 months ago
by 
RedDragonGecko
Rambling nonsense
6
#3 opened 3 months ago
by sousekd
SGLang kernel version exception
#8 opened 3 months ago
by sousekd
Q5_K_M inference speed on llama.cpp vs ik_llama.cpp
👍❤️ 2
#1 opened 3 months ago
by sousekd
Great model for single GPU use cases.
🔥 4
16
#1 opened 4 months ago
by 
phakio
No think tags.
10
#4 opened 4 months ago
by 
DrRos
Working configuration for Nvidia Blackwell
13
#4 opened 4 months ago
by 
luismiguelsaez
The Best Kimi Quant!
🔥 4
13
#1 opened 4 months ago
by 
anikifoss
Unable to use full 192k context in SGLang with MiniMax-M2.7-NVFP4 (runtime capped at ~80,964 tokens)
3
#9 opened 4 months ago
by mtcl
IQ5_K 136.891 GiB
🔥 2
30
#9 opened 6 months ago
by 
Hunterx
IQ3_KS on EPYC 9355 + 1x RTX 5090
🔥 2
5
#5 opened 5 months ago
by sousekd
1x RTX 5090 on EPYC, CPU-only speeds
❤️ 3
11
#5 opened 6 months ago
by sousekd
EPYC 9355 CPU-only sweep-bench
5
#3 opened 6 months ago
by sousekd
EPYC 9355 CPU-only sweep-bench
🔥 1
#6 opened 6 months ago
by sousekd
EPYC, RTX 5090 vs RTX 6000
🔥 1
7
#7 opened 8 months ago
by sousekd
Benchmarks (Q8_0-Q4_0, RTX 5090, EPYC)
🔥 1
5
#5 opened 9 months ago
by sousekd
Going to fit it into RTX PRO 6000!
👍 1
12
#1 opened 8 months ago
by mtcl