Help: How to use KoboldCpp when unpacked to folder or launch template?

[Krxon]

Hello

1
Can someone guide me or post url to guide how to use Unpacking feature?
I'd like to avoid creating 2.76 GB of files in temp dir each time when I run Kobold, to reduce nvme wear.
Using KoboldCPP-v1.83.1.yr1-ROCm on 7900 XT atm.
I tried unpacking it, but I don't know what to do after that - how to launch it using unpacked files
with selected settings, Text model, image model, image lora and Whisper model

2
When make my settings and I create Launch template, when I launch it by dropping Run.kcppt on KoboldCPP-v1.83.1.yr1-ROCm.exe file, it launches it, but language model doesn't use GPU then.
How to solve that?

Thanks

[Krxon]

Nobody?
Not even on Kobold Reddit... sighs..

[sinni800]

1. Download python-3.10-embed
2. Save the koboldcpp .exe as koboldcpp_rocm.exe
3. download pyinstxtractor-ng.exe (just google it, you will find the project)

Then this batch file will do the rest:

set subdir=koboldcpp_rocm
set pythonsource=python-3.10.0-embed-amd64

rmdir %subdir% /s /q
pyinstxtractor-ng.exe -d %subdir%.exe
ren %subdir%.exe_extracted %subdir%
robocopy %pythonsource%\ %subdir% /s /e /r:0 /z

The same works with regular koboldcpp for CUDA, but there, it needy python 3.8 (like 3.8.10)

to run, use this batch file:

set subdir=koboldcpp_rocm

set TCL_LIBRARY=%cd%\%subdir%\_tcl_data
set TK_LIBRARY=%cd%\%subdir%\_tk_data
%subdir%\python %subdir%\koboldcpp.pyc %*

[Krxon]

Thanks :D