search

trzy / llava-cpp-server

LLaVA server (llama.cpp).
MIT License
173 stars 9 forks source link

whe I give non-existant file names, it segfaults #3

Closed dataf3l closed 11 months ago

dataf3l commented 11 months ago

these files do not exist:

models/*

I get this error

➜  llava-cpp-server git:(main) ./bin/llava-server -m ./models/ggml-model-q5_k.gguf --mmproj ./models/mmproj-model-f16.gguf
[1]    57986 segmentation fault  ./bin/llava-server -m ./models/ggml-model-q5_k.gguf --mmproj

what I kinda expected: "sorry this file doesn't exist"

thanks in advance

trzy commented 11 months ago

Great catch but this is actually a problem with llama.cpp itself, which handles command line parsing and model file loading. That is, if you go into llama.cpp/ (or clone that repo independently and build it), and run ./llava with nonexistant filenames, the same segfault occurs. I want to refrain from reaching inside of llama.cpp but this would be a good issue to file with them directly.

dataf3l commented 11 months ago

@trzy https://github.com/ggerganov/llama.cpp/issues/3663