convert : add --local-model and --local-llama options

[danbev]

This commit adds two new options to the convertion script to enable specifying a local model directory instead the models being downloaded form HuggingFace. And the other option is to enable a local llama.cpp directory to be used instead of upstream llama.cpp

The motivation for these options is to enable the convert repo/tool to be used with models that are not yet available on HF, for example unreleased models. These models sometimes requires additions/changes to llama.cpp as well and by introducing the local-llama option a local/private llama.cpp can be used to convert the models.

---

example usage

#!/bin/bash

set -e

if [ ! -x venv/bin/python ]; then
      bash create-venv.sh
fi


local_model=~/work/models/qwen/Qwen3.8-27B
local_llama=~/work/llama.cpp-some-local-dev-repo

convert_models() {
	source venv/bin/activate
	./convert.sh --owner danbev --one qwen3.8-27b --no-upload --keep --force \
		--local-model ${local_model} \
		--local-llama ${local_llama}
	deactivate
}
convert_models

Example upload: https://huggingface.co/danbev/Qwen3.8-27B-GGUF

Note

This model repo is private and should not be visible which is actually what we want so we don't disclose any information about new models.