TEST #2012 RESULTS
12/21/2025 - 1:28 PM
ACCELERATOR
MODEL
59.8
tokens/s
generation
756
ms
time to first token
1693
tokens/s
prompt
512
LocalScore
HOW YOU STACK UP
Explore All ResultsMeta Llama 3.1 8B Instruct - Q4_K - Medium
SYSTEM
CPU
AMD Ryzen 9 5900X 12-Core Processor (znver3)
RAM
31.9GB
OS
Windows
Kernel Release
10.0
Architecture
x86_64
Version
Cosmopolitan 3.9.7 MODE=x86_64
RUNTIME
Name
llamafile
Version
0.9.3
Commit Hash
a30b324
DETAILED RESULTS
TEST NAME
PROMPT
GENERATION
TTFT
pp1024+tg16
1995
tokens/s
62.3
tokens/s
530
ms
pp4096+tg256
1651
tokens/s
52.1
tokens/s
2.50
sec
pp2048+tg256
1862
tokens/s
58.6
tokens/s
1.12
sec
pp2048+tg768
1865
tokens/s
57.7
tokens/s
1.12
sec
pp1024+tg1024
1971
tokens/s
61.0
tokens/s
537
ms
pp1280+tg3072
1890
tokens/s
55.9
tokens/s
693
ms
pp384+tg1152
2057
tokens/s
63.4
tokens/s
203
ms
pp64+tg1024
1471
tokens/s
64.0
tokens/s
59
ms
pp16+tg1536
477
tokens/s
63.3
tokens/s
49
ms