How to fix: ValueError: Failed to detect model architecture

[VenominousX]

Hi,

I'm currently learning the basics of creating my own AI models. I've already managed to get as far as the .pth and .safetensors files, but I'm stuck when it comes to converting .safetensors to gguf.
When I run the convert_hf_to_gguf.py script, I get the following error:

ValueError: Failed to detect model architecture

The Config.json file must contain a “model_type” key, but I don't know which of the options are allowed for a custom model.

I am using a GPT2 Tiktoken, but when I enter “gpt2” as the model_type, I also get the error.

What can I do to prepare a self-built model so that I can use it in Ollama?

It's probably a stupid question, but I'm stuck here^^.

Best regards

[WaterYue]

same question...
convert_hf_to_gguf.py , line 10123, in get_model_architecture
raise ValueError("Failed to detect model architecture")
ValueError: Failed to detect model architecture

[BazzFar]

Any luck? I am facing the exact same in convert_hf_to_gguf.py

[misutoneko]

Hmm if only there was an LLM that could answer any questions about llama.cpp codebase... :D

In the meanwhile you could take a look at this:
#16770

And here's what it says in the source:

$ grep @ModelBase.register convert_hf_to_gguf.py |grep -i gpt2
@ModelBase.register("GPT2LMHeadModel")

I suppose that @ModelBase.register call must match what you have in config.json.