TEST #9 RESULTS
03/19/2025 - 10:20 PM
ACCELERATOR
MODEL
49.3
tokens/s
generation
612
ms
time to first token
2247
tokens/s
prompt
565
LocalScore
HOW YOU STACK UP
Explore All ResultsMeta Llama 3.1 8B Instruct - Q4_K - Medium
SYSTEM
CPU
AMD EPYC 7352 24-Core Processor (znver2)
RAM
125.64GB
OS
Linux
Kernel Release
6.8.0-52-generic
Architecture
x86_64
Version
Cosmopolitan 3.9.7 MODE=x86_64; #53~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Wed Jan 15 19:18:46 UTC 2
RUNTIME
Name
llamafile
Version
0.9.1
Commit Hash
a30b324
DETAILED RESULTS
TEST NAME
PROMPT
GENERATION
TTFT
pp1024+tg16
2391
tokens/s
50.2
tokens/s
448
ms
pp4096+tg256
1925
tokens/s
45.9
tokens/s
2.15
sec
pp2048+tg256
2409
tokens/s
48.7
tokens/s
871
ms
pp2048+tg768
2305
tokens/s
48.3
tokens/s
909
ms
pp1024+tg1024
2728
tokens/s
49.7
tokens/s
395
ms
pp1280+tg3072
2673
tokens/s
47.8
tokens/s
499
ms
pp384+tg1152
3117
tokens/s
50.6
tokens/s
143
ms
pp64+tg1024
2063
tokens/s
51.1
tokens/s
50
ms
pp16+tg1536
608
tokens/s
50.8
tokens/s
45
ms