Open jeffmaury opened 1 month ago
@MichaelClifford @Gregory-Pereira Is that something you can look at? It would also make sense to implement a CI/Testing Framework test. WDYT?
I will take a look at this tonight
@Gregory-Pereira do you want me to take this one over?
The temporary fix is to add back the chat_format
to the makefile for only runs or to the run.sh
but this invalidates the work that @jeffmaury wanted around check the chat_format from the model first. The answer to this is that we cannot leverage the native llamacpp functionality to do this, we could add something to the run.sh
to do this in a smarter way (ie read the .gguf
model metadata from the $MODEL_PATH
we already have and propgate it as the value to the run.sh script
... Ill need to think on this. For the meantime il put a temp fix in
Maybe this might be usable in this smarter implementation: https://github.com/ggerganov/llama.cpp/blob/master/gguf-py/scripts/gguf-new-metadata.py
Reopened to keep this here for the long term fix
Got this while running from main branch in Podman AI Lab: