fix: improved error handling when llama.cpp build fails #2358

[Hansehart]

Hi, this has the same changes like the reverted pr, but with fixed code. Copy and pasted from before:

When I tried to run save_pretrained_gguf() I had multiple different errors. All problems were related to a failing LLama.cpp build. As you can see unter this pull request , curl for llama cpp is now activated by default, which is resulting for me in:

CMake Error at common/CMakeLists.txt:90 (message):
  Could NOT find CURL.  Hint: to disable this feature, set -DLLAMA_CURL=OFF

(System is a Ubuntu 24.04 with newest version of unsloth and llama.cpp. Curl is installed.)

So then I installed llama.cpp manually, but then got the error:

Unsloth: The file 'llama.cpp/llama-quantize' or `llama.cpp/quantize` does not exist.\n"\
            "But we expect this file to exist! Maybe the llama.cpp developers changed the name or check extension of the llama-quantize file."

Thats right, because is it still under /llama.cpp/build/bin. As the save.py shows, all llama-* files gets copied to llama.cpp, when the build runs successfully. But in my case that dont happend and I had some misleading error messages. I hope that pull request decreases the amount of people having these issues or a clearer direction for debugging!

[danielhanchen]

Thanks for the PR! This looks good!