Trying to run llama-server, but llama_model_load: error loading model: missing tensor 'token_embd.weight'

[MichelleTanPY]

Hi everyone, I'm new to this repo and trying to learn and pick up some easy issue to contribute to. I'm following all the steps in this README , trying to run llama-server locally, but I ended up with this error: llama_model_load: error loading model: missing tensor 'token_embd.weight'. I've tried all the .gguf in models server, and was wondering if the README is slightly dated as I don't see this models\7B\ggml-model.gguf as well.
Appreciate anyone's time on explaining this.

[MichelleTanPY]

I don't completely understand why. But the *.gguf models in the models/ folder doesn't seem to work. I downloaded another model, that manage to get the llama-server up and running.

[slaren]

If you mean the ggml-vocab-*.gguf files, these only contain the tokenizers and are meant to be used with the tests only. I think it would be good to move them to /tests to avoid confusion.

[MichelleTanPY]

@slaren Yes, it was quite confusing. Thanks for clarifying. Will it be ok if I submit a PR for moving these under a new test directory within models? /models/tests

[slaren]

Yes, /models/tests sounds good to me.