latest llama.cpp using q5_0, q4_1 error: "is this really a GGML file?"

#5
by Free-Radical - opened

I am using latest llama.cpp (master-66874d4) and it only works for:

  • ggml-vicuna-13b-4bit.bin

but NOT for:

  • stable-vicuna-13B.ggmlv3.q5_0.bin
  • stable-vicuna-13B.ggmlv3.q4_1.bin

Please check the sha256sum for the stable-vicuna-13B.ggmlv3.q5_0 and q4_1 files, or if in doubt download them again.

Those two files definitely work with the latest llama.cpp, so you likely have incomplete/corrupted downloads. (In fact the q5_0 file would work with older llama.cpp as well - only q4_0, q4_1 and q8_0 require the latest llama.cpp)

I just re-downloaded and tested q5_0 with llama.cpp compiled today and confirmed it works OK.

Ok , MY BAD, 🫢 very embarrassed, i forgot to recompile. YES THEY WORK!

Free-Radical changed discussion status to closed

Sign up or log in to comment