Model status now says, per backend, which Python packages cannot be
imported (importlib.util.find_spec, nothing imported): llama-cpp-python
for GGUF, torch/transformers/accelerate/bitsandbytes for HuggingFace,
onnxruntime for WD14, with the pip install command and a note (the CUDA
CMAKE_ARGS for llama-cpp-python, onnxruntime-gpu for WD14). Nothing is
installed.
load_model refuses such a model before the loader with
MissingBackendPackages (one WARNING line, no traceback); a
ModuleNotFoundError from a loader becomes the same refusal. The suggest
route, the legacy single and batch routes and both tagging jobs answer
409 model-unavailable with that sentence.
System > Auto-tagging reads "Downloaded. Needs llama-cpp-python: pip
install llama-cpp-python. ..." instead of "Downloaded. Not loaded.",
Review explains instead of reviewing, and the bulk tagging dialog keeps
Start disabled with the sentence. "downloaded" keeps its files-only
meaning, so the legacy pages are unchanged.
Closes#424
Vito
41027aa1b1