Homarr

llama.cpp

System Monitoring

Shows the health, loaded model and generation speed of a local llama.cpp llama-server.

This widget shows the live status of your local llama.cpp llama-server: whether it is online, which model is loaded (with quantization, context size, file size and parameter count), the generation speed in tokens per second, and how many requests are being processed. When the server is started with --metrics, it also reports total token counts and the prompt cache hit rate.

Generation speed

The widget picks the most meaningful speed for the current state (hover the speed for an explanation of which one is shown):

  • While a request is being generated — the average speed of that request since it started, derived from the token count reported by /slots.
  • While the server is busy but the per-request data is unavailable — the live speed from the /metrics rate gauge.
  • While the server is idle — the average generation speed since the server started (cumulative tokens ÷ seconds), so the value stays meaningful instead of dropping to zero.

Supported Integrations

llama.cpp icon
llama.cppConnect to a local llama.cpp llama-server to show model, throughput and request status on your dashboard.Displays health, model details, generation speed and request status
View guide

Adding the widget

Add the widget from the board's Add board content menu in edit mode.

Configuration

NameDescriptionValuesDefault value
Show titleDisplays the llama.cpp label and status badge at the top of the widgetyes / noyes
Show model infoDisplays the loaded model name, quantization, context size and file sizeyes / noyes
Show cache hit rateDisplays the prompt cache hit rate as a percentage baryes / noyes
Show context usageDisplays the context (KV cache) token usage as a progress baryes / noyes

On this page