is convert pt to gguf possible by llama.cpp?

[IdeaMeshDyx]

Hi, everyone. I am new to the llama.cpp community. Recently, I tried to use llama.cpp to load some models I found. During this process, I encountered a serious problem: the format conversion among different storage formats of LLM is quite messy.
For example, if I have a .pt/.pth file, should I first convert it to safetensor and then convert that safetensor file to gguf? Additionally, I have found very few methods or all-in-one tools/tutorials to handle these conversions effectively.
Moreover, I have discovered another issue: not all .pth/.pt files are the same. This raises the question of which specific files are necessary for a successful conversion. I have tried some repositories[1][2][3] that convert .pt files to safetensor, but they only produce a single safetensor file, which does not include config.json and other potentially necessary files.
I apologize for my unclear expression. These problems seem to be intertwined, and I am struggling to explain them clearly. If anyone has experience with these issues or can provide some guidance, I would greatly appreciate it.

[1] https://github.com/jtabox/safetensors-converter/blob/main/safetensors_converter.py
[2] https://github.com/Silver267/pytorch-to-safetensor-converter
[3] https://github.com/huggingface/safetensors/blob/main/bindings/python/convert.py

[NEWbie0709]

Hey, did you find a solution for it? I want to convert the PT model to GGUF in order to use it in Ollama, but I don't know how to do it. This is the model I want to try
https://huggingface.co/mobiuslabsgmbh/Llama-3.1-8b-instruct_4bitgs64_hqq_calib

[misutoneko]

I haven't tried conversion from .pt or .pth myself, but a quick search turned up these:
#8808
https://github.com/codiak/llama-raw-to-py

But really, if you only want to run the models, you're much better off just finding a ready-made GGUF.
I suspect they're all over the place at HF by now.
And as for HQQ, it's not supported in llama.cpp (or ollama) AFAIK. Or at least git log didn't return any hits for me...