Llama.cpp llama-cli error: (GGML_ASSERT(ggml_is_contiguous(src0)) failed). Is anyone know?

[AkiraToya]

Build use from: https://github.com/ggerganov/llama.cpp/releases/tag/b3827
When try cuda, I'm using the cu117 since this is what I use on my PC, already used for many cuda related usage.

Hi! Is there anyone here ever use llama.cpp the llama-cli command? If yes... Is anyone know what this error is?

llamainit_from_gpt_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
D:\a\llama.cpp\llama.cpp\ggml\src\ggml-cuda\norm.cu:212: GGML_ASSERT(ggml_is_contiguous(src0)) failed

I only use code:
./ext/llama-cli -m .\llm\falcon-mamba\falcon-mamba-7b-instruct.Q2_K.gguf -p "Hi!" -ngl 1

If I didn't use -ngl 1, it's working fine. But if my prompt also long ( not so long though, can be said short ) about 30 words, it's also give me the same error.

My GPU is NVIDIA. So far I can use llamafile with gpu, ollama too. But the main problem is I don't know why it is getting that error. @@..

Another information when using "-cnv":
system
Hi!

Hello! How are you?
I'm good! Thank you for asking. How about you? How was your day?

I'm good. I want to give you an order. can I?
Of course! Go ahead and share your order with me, and I'll be happy to help. Let me know how I can assist you!

Please write a super long article about Overall Immeasurable Prowess of Michael Jordan vs Super Strength and Power Lebron James vs Stephen Curry Unrealistic 3Pt Shoot Accuracy.
D:\a\llama.cpp\llama.cpp\ggml\src\ggml-cuda\norm.cu:212: GGML_ASSERT(ggml_is_contiguous(src0)) failed

And it just died like that if I writing longer prompt ( This is not using ngl ).

Again, another info: Somehow I can run it when using llama-cli for avx2 build. It can handle long prompt, not error whatsoever. But it can't use GPU TT__TT... So... How? Am I needed to compile the llamacpp myself to get cuda + avx2? ( I am really new to llamacpp, so... I'm not even sure I can do that.... )