Attempted to read or write protected memory with llama.dll from the llama.cpp project - b2072 release

[AshD]

I wanted to try the new Qwen model, so decided to update my setup

Steps

1. Get the latest LlamaSharp
2. LlamaSharp loads the Mistral Instruct v0.2 fine
3. Get the llama.dll from llama-b2072-bin-win-cublas-cu11.7.1-x64.zip
4. Loading the same Mistral Instruct v0.2 throws System.AccessViolationException in llama_load_model_from_file(modelPath, lparams)

System.AccessViolationException
  HResult=0x80004003
  Message=Attempted to read or write protected memory. This is often an indication that other memory is corrupt.

Wanted to check if I am upgrading it incorrectly or there is an issue with the latest llama.cpp?

Thanks,
Ash

[martindevans]

Get the llama.dll...

You can't do this. There's zero compatibility guarantee from one version of llama.cpp to another.

For example this is the set of change required to update from 9fb13f9 to d71ac90

[AshD]

Thanks Martin. I did not realize that the llama.cpp API was changing so much between releases.