Tesla A10 and cuBLAS and cuDNN 8.x libs versions #197
1001ruchka
started this conversation in
General
Replies: 3 comments 9 replies
-
What is "Version 4.7" & "Version 2.7"? |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Beta Was this translation helpful? Give feedback.
3 replies
-
That's probably an hallucination from custom initial prompt, I'll change it in the next version. |
Beta Was this translation helpful? Give feedback.
6 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
It turned out that depending on the version of cuBLAS and cuDNN 8.x libs - errors may appear or disappear on Tesla a10.
Version 4
Timing 02:01:56,430 - ... instead of text
Purfview-Whisper-Faster log-4.zip
Version 2
Timing 02:01:01,750 - Speech is normally recognized
Purfview-Whisper-Faster log-5.zip
At the same time, if you compare the quality of recognition of text not in English (I used Russian) - the recognition quality is higher by v4
Command line arguments: --sentence --compute_type=float16
I gave up using int8 with them Whisper makes huge long sentences. float32 is slower than float16
Whisper-Faster r172.3 version was used.
Beta Was this translation helpful? Give feedback.
All reactions