Nonstop answer from HTTP OpenAI-compatible model

[hxt365]

Describe the bug
I ran Tabby with an HTTP OpenAI-compatible chat model running remotely with vLLM. In the chat, Tabby kept sending a nonstop answer, like the following image.

Information about your version
0.18.0

Additional context
Log from Tabby:

tabby-1  | The application panicked (crashed).
tabby-1  | Message:  index out of bounds: the len is 0 but the index is 0
tabby-1  | Location: ee/tabby-webserver/src/service/answer.rs:173
tabby-1  | 
tabby-1  | Backtrace omitted. Run with RUST_BACKTRACE=1 environment variable to display it.