Model Load Fails with "invalid ggml type 39 (NONE)"

#1
by danchev - opened

Attempting to load a GGUF model using llama-cpp fails during initialization with the following error:

gguf_init_from_file_impl: tensor 'blk.0.ffn_down_exps.weight' has invalid ggml type 39 (NONE)
gguf_init_from_file_impl: failed to read tensor info

https://github.com/ggml-org/llama.cpp/pull/15091

You need a version of llama-cpp that was released with this pr merged.

@Green-Sky The PR appears to have just been merged into master. As suggested, rebuilding llama-server using the latest commit resolved the issue. Thank you!

danchev changed discussion status to closed

Sign up or log in to comment