Light-PDF-Web-QA-Chatbot / Generation speed GPU test.txt
Sean-Case
Added working like buttons. Added model choice. Modified requirements txt
2e536f9
raw history blame
No virus
1.44 kB
With 5 gpu layers, batch size 8
Num of generated tokens: 113
Time for complete generation: 115.42684650421143s
Tokens per secound: 0.9789750255013432
Time per token: 1021.4765177363843ms
With 5 gpu layers, batch size 512
Num of generated tokens: 102
Time for complete generation: 40.369266986846924s
Tokens per secound: 2.5266745624396285
Time per token: 395.77712732202866ms
With 6 gpu layers -
Num of generated tokens: 113
Time for complete generation: 46.37785983085632s
Tokens per secound: 2.4365074285902764
Time per token: 410.42353832616215ms
With 6 gpu layers, batch size 1024 -
Five pillars Q:
Num of generated tokens: 102
Time for complete generation: 41.85241961479187s
Tokens per secound: 2.4371350793766346
Time per token: 410.31783936070457ms
With 8 threads
Num of generated tokens: 102
Time for complete generation: 40.64410996437073s
Tokens per secound: 2.5095887224351774
Time per token: 398.4716663173601ms
Vision statement Q:
Num of generated tokens: 84
Time for complete generation: 35.57932233810425s
Tokens per secound: 2.360921863597128
Time per token: 423.5633611679077ms
Commitments Q:
Num of generated tokens: 50
Time for complete generation: 23.73319172859192s
Tokens per secound: 2.106754142965266
Time per token: 474.6638345718384ms
Outcomes Q
Num of generated tokens: 167
Time for complete generation: 52.302518367767334s
Tokens per secound: 3.1929628861412094
Time per token: 313.1887327411217ms