Feature request: diversify error mesages when loading a model fails

[flatsiedatsie]

Currently a model can fail to load for a number of different reasons. However, the error raised seems to always be a general "failed to load" error. It would be useful if different errors could be raised for:

• 404 model file not found ( -> inform user to check the URL they provided)
• Server isn't responding (HuggingFace is down again.. -> load from backup server)
• Couldn't load model because there is no internet connection (-> suggest loading another model that is cached)
• Failed to load from cache ( -> clear cache and retry downloading)
• Couldn't load model because the file doesn't actually seem to be a valid .gguf file ( -> inform user to check the URL they provided)
• Couldnt load model because it doesn't fit into memory ( -> recommend trying a smaller one)
• Etc

This is because in my project users can enter their own URL to a .gguf file (or provide a list of shards), so failure can come in many forms.

I can see in the debug console that the worker has precise information (e.g. GET failed), that that precision isn't passed on yet.

[flatsiedatsie]

Another exampleI just ran into while trying to load the new version of Phi 3 128K (Q2). I suspect this error is because of a Llama.cpp version mismatch?