llama-server <embedding> exited with status code -1073741819

[spock]

Describe the bug
Nightly and 0.17.0 fail in all tried modes (rocm, cpu) with the same message - on Windows.

Information about your version

tabby 0.17.0
tabby 0.18.0-dev.0

Information about your GPU
Have AMD GPU; command fails also when run on CPU.

Additional context

Command to start llama-server appears to have a wrong path separator in the model path, ggml/model.gguf which on Windows should probably be ggml\model.gguf. This is the command listed in the logs:

"C:\\Users\\username\\Downloads\\tabby_x86_64-windows-msvc\\dist\\tabby_x86_64-windows-msvc\\llama-server.exe" "-m" "C:\\Users\\username\\.tabby\\models\\TabbyML\\Nomic-Embed-Text\\ggml/model.gguf" "--cont-batching" "--port" "30888" "-np" "1" "--log-disable" "--ctx-size" "4096" "-ngl" "9999" "--embedding" "--ubatch-size" "4096", kill_on_drop: true 

[spock]

the same nightly tabby 0.18.0-dev.0 started fine under WSL2

[wsxiaoys]

fixing in #3152

[ANYMS-A]

I met a similar issue while the ggml model's path seems fine:

Starting...2024-09-25T07:34:39.926298Z  WARN llama_cpp_server::supervisor: crates\llama-cpp-server\src\supervisor.rs:98: llama-server <embedding> exited with status code -1073741819, args: `Command { std: "C:\\Users\\xxx\\Downloads\\dist\\tabby_x86_64-windows-msvc\\llama-server.exe" "-m" "C:\\Users\\xxx\\.tabby\\models\\TabbyML\\Nomic-Embed-Text\\ggml\\model.gguf" "--cont-batching" "--port" "30888" "-np" "1" "--log-disable" "--ctx-size" "4096" "-ngl" "9999" "--embedding" "--ubatch-size" "4096", kill_on_drop: true }`