Allow passing additional config files (e.g. chat templates) into the model container
#1.783 aberto em 1 de ago. de 2025
Métricas do repositório
- Stars
- (3.010 estrelas)
- Métricas de merge de PR
- (Mesclagem média 6d 7h) (32 fundiu PRs em 30d)
Description
Feature request description
I was poking at ramalama today, and wound up trying to run the unsloth version of the new Qwen3 coder model. There's a bug in it, though. They sent out a tentative fix - an updated chat template file - with instructions to just run with --chat-template-file <path/to/thatfile>, no need to redownload the whole model!
So, OK, I try with --runtime-args="--jinja --chat-template-file <path/to/thatfile>"...and it complains about not being able to read the file. I assume this is because the model container doesn't have my home directory mounted into it or anything like that.
Is there any way to handle this kind of situation? I poked about in the docs but couldn't find anything. If I'm understanding it all right, it seems like it would be useful to have an option to mount some arbitrary file or directory into the model container for cases like this. If I'm not understanding it all right, please let me know. :D
Suggest potential solution
No response
Have you considered any alternatives?
No response
Additional context
No response