|
Post by nathanmyersc on Mar 24, 2024 19:48:42 GMT
I created a Google Colab file using my new Intention Repeater Simple CUDA. It has hashing where the old colab workbook didn't.
I also made a minor update to the CUDA files, so you can update if you wish.
"I am Love." with 5GB RAM runs at 453YHz.
Cool im GOOGLE COLLABing it right now.
|
|
|
Post by reden on Mar 24, 2024 21:50:18 GMT
Ok, I updated the Simple, Simple Multithreaded and Simple CUDA to allow for command-line flags/parameters. github.com/tsweet77/intention-repeater-simpleUse /? or --help after the .exe for list of flags.
I'm keeping the Multithreading one separate to keep the files simpler, and to help others learn a bit easier.
Should I add in LIBZ compression option? No idea its just as experimental as hashing is. i have no clue whether it conveys the intent in a smaller package same as hashing. its hard to know these things except to test them. Compression has been proven to work
|
|
|
Post by reden on Mar 24, 2024 21:50:57 GMT
Ok, I updated the Simple, Simple Multithreaded and Simple CUDA to allow for command-line flags/parameters. github.com/tsweet77/intention-repeater-simpleUse /? or --help after the .exe for list of flags.
I'm keeping the Multithreading one separate to keep the files simpler, and to help others learn a bit easier.
Should I add in LIBZ compression option? Add Zlib and check if it's better.
|
|
|
Post by nathanmyersc on Mar 24, 2024 22:48:46 GMT
The cuda version keeps stopping at 4 minutes in google collab for me. dunno why even if i set --dur to much higher.
|
|
|
Post by AnthroHeart on Mar 24, 2024 23:15:38 GMT
Ok, I updated the Simple, Simple Multithreaded and Simple CUDA to allow for command-line flags/parameters. github.com/tsweet77/intention-repeater-simpleUse /? or --help after the .exe for list of flags.
I'm keeping the Multithreading one separate to keep the files simpler, and to help others learn a bit easier.
Should I add in LIBZ compression option? Add Zlib and check if it's better. I have added compression option to the Intention Repeater Simple and Intention Repeater Simple (Multithreaded).
I couldn't get the ZLIB library to work with the NVCC compiler for CUDA. So if anyone knows how to link that library in compile time, let me know.
You can run like following:
Intention_Repeater_Simple_Multithreaded.exe --intent "I am Love." --imem 1 --compress y --hashing y --dur 00:00:10
Intention_Repeater_Simple.exe --intent "I am Love." --imem 1 --compress y --hashing y --dur 00:00:10
Updated code and binaries:
|
|
|
Post by AnthroHeart on Mar 24, 2024 23:52:44 GMT
I implemented compression on CUDA but it runs slower on CUDA for some reason. About 1/2 the speed. I'm not sure why.
So I probably won't update the one online. It would require an additional .h file as well as the .exe requiring a .dll to run.
C:\Repeater\Simple_Multithreaded>Intention_Repeater_Simple_CUDA.exe --intent "I am Love." --imem 1 --hashing y --compress n --dur 00:00:10 Intention Repeater Simple CUDA by Anthro Teacher, WebGPT and Claude 3 Opus
[00:00:10] Repeating: (318.672S / 31.673YHz): I am Love.
C:\Repeater\Simple_Multithreaded>Intention_Repeater_Simple_CUDA.exe --intent "I am Love." --imem 1 --hashing y --compress y --dur 00:00:10 Intention Repeater Simple CUDA by Anthro Teacher, WebGPT and Claude 3 Opus
[00:00:10] Repeating: (163.500S / 16.484YHz): I am Love.
C:\Repeater\Simple_Multithreaded>
|
|
|
Post by AnthroHeart on Mar 25, 2024 16:21:33 GMT
I updated the Simple, Multithreaded and the CUDA.
Added Compression to CUDA. It runs slower with compression though for some reason.
CUDA executable requires: zlib.dll in the same folder. Available in the Github repo.
CUDA code requires picosha2.h, zconf.h, zlib.h to compile. Also requires the miniconda3 library for some reason. Directions for compiling are on the Github repo.
It was possible that the compiler was optimizing out the repeated string assignment (repeater functionality), so I used a new technique from GPT-4 to make it run better, and not possibly get optimized out.
I also updated the Intention Repeater MAX to v5.9 to use this new technique. I default it now to EXACT timer which runs a bit slower than the INEXACT method. You can run INEXACT method using flag: --timer INEXACT I wanted the MAX to run similarly to the Simple. They run at the same speed now. Use --timer INEXACT to run faster, but time display will be off.
|
|