llama.cpp
llama.cpp copied to clipboard
bad command line parsing behaviour with some filenames
quantize (only command I have tested this with) gets confused by some filenames of some models, e.g.
quantize 08x7bhf.gguf 08x7bhfGGUF~ Q4_K_S
... should quantize one gguf file into another, but instead, it fails with a weird error message:
main: invalid nthread 'Q4_K_S' (stoi)
My guess is that somehow it interprets the 08x name as a number, but that's clearly not the whole story. Might even be a security issue if commands can be tricked into misinterinterpreting filenames as something else (this behaviuour cannot be suppressed with "--" either).