Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Prefilling Profiling on WikiText (test)
Loading...
38
Time (s)
W4A4
9.84
199.92
390
580.08
Dec 5, 2025
Time (s)
Effective BW (GB/s)
Speedup (x)
Updated 1mo ago
Evaluation Results
Method
Method
Links
Time (s)
Effective BW (GB/s)
Speedup (x)
W4A4
Model=Llama-3-8B, bank...
2025.12
38
40.61
1.26
SQ-format
Model=Llama-3-8B, Spar...
2025.12
41
38.19
1.17
SQ-format
Model=Llama-3-8B, Spar...
2025.12
42
36.9
1.14
SQ-format
Model=Llama-3-8B, Spar...
2025.12
45
35.75
1.07
W4A8
Model=Llama-3-8B, bank...
2025.12
48
34.69
1
W4A8
Model=Llama-3-70B, ban...
2025.12
108
10.84
1
W4A4
Model=Llama-3-70B, ban...
2025.12
516
20.86
1.92
SQ-format
Model=Llama-3-70B, Spa...
2025.12
555
18.54
1.71
SQ-format
Model=Llama-3-70B, Spa...
2025.12
629
16.94
1.56
SQ-format
Model=Llama-3-70B, Spa...
2025.12
742
14.27
1.32
Feedback
Search any
task
Search any
task
