Closed j2wx closed 3 months ago
Which models are these?
The models I tried are Toppy-M-7B and SnowLotus-v2-10.7B.
Do you have a link to the specific model quant file that doesn't work?
Yes, here are the links:
Toppy-M-7B: https://huggingface.co/Undi95/Toppy-M-7B-GGUF/resolve/main/Toppy-M-7B.q4_k_s.gguf SnowLotus-v2-10.7B: https://huggingface.co/BlueNipples/DaringLotus-SnowLotus-10.7b-IQ-GGUF/resolve/main/SnowLotus-v2-10.7B-Q3_K.gguf
Hi, I tried Toppy-M-7B and it worked fine. Please try the latest version which is just released. Also, try reducing the number of offloaded layers.
Thanks, I tried it with the latest version and I get this error while starting:
Warning, GUI failed to start: Reason: No module named 'customtkinter'
File selection GUI unsupported.
customtkinter python module required!
Please check command line: script.py --help
I installed customtkinter, but it still show this error message.
What is your OS? For windows and linux you can grab the latest release here.
Otherwise, you can run it without GUI with the command line. For example
python koboldpp.py --model /path/to/your/model.gguf
I use Linux. I use the "koboldcpp.sh" and I also tried the program from the release.
I tried it with "koboldcpp.py" and I get a error in my language which translates to "`Invalid machine command (memory dump written)". Is the problem related to my OS?
Alright. Do you have a GUI? If yes, I suggest just trying this binary https://github.com/LostRuins/koboldcpp/releases/download/v1.62.2/koboldcpp-linux-x64 which should make things easier.
This is the one I tried. I tried it in a live OS this time to see if it works there, but it gave the same error as with the "koboldcpp.py": Illegal instruction
.
I suspect your trying to use the regular backends on a CPU that does not have AVX2 support. Can you double check your CPU supports AVX2 and if not try the Old CPU or Failsave backends?
Sometimes this is the error when you try to call a .py file like a .sh file. Try to call with python3 before, like:
python3 ./koboldcpp.py
My CPU only supports AVX, but I read somewhere that it's enough. I tested it with the Old CPU backend and it works now. Thank you very much for your help.
I tried to load some GGUF models, but the loading process is stuck at this line:
[Threads: 2, BlasThreads: 2, SmartContext: False, ContextShift: True]
I already tried it without ContextShift and SmartContext, but it still won't load. With Oobabooga it works without problem.