Add support for API-Key protected /slots endpoint introduced in llama.cpp b3898

[countzero]

Description

Since the llama.cpp release b3898 the /slots API endpoint is secured with an API key if it's set.

This is a documented breaking change: changelog : llama-server REST API

Implications for Paddler

The paddler agent needs access to the /slots API endpoint to determine the load on the llama-server.

Suggestion

Add a new option --llama-server-api-key to configure the paddler agent. Use the API key in requests to the /slots API endpoint.

Context

The API key implementation of llama.cpp mirrors the OpenAI API Authentication. The following HTTP header is expected in requests to a llama-server instance that is secured via either the --api-key or the --api-key-file option:

Authorization: Bearer LLAMA_API_KEY

[mcharytoniuk]

@countzero Thank you for your detailed report! This feature is implemented in 0.9.0 behind --local-llamacpp-api-key in paddler agent.