Struct DashboardServerExt 

pub struct DashboardServerExt {
    pub config: DashboardConfig,
    pub loss_history: MetricHistory,
    pub lr_history: MetricHistory,
    pub grad_norm_history: MetricHistory,
    pub connected_clients: usize,
    pub messages_sent: u64,
    /* private fields */
}

In-process training dashboard that records metric histories and can generate CSV / JSON summaries suitable for logging or downstream tools.

This is an in-process implementation that does not bind network sockets. For SSE streaming, see super::websocket::DashboardServer.

Example

use trustformers_debug::dashboard_ws::metrics::{DashboardConfig, DashboardServerExt};

let mut server = DashboardServerExt::new(DashboardConfig::default());
server.record_metrics(10, 1.5, 1e-3, 0.9);
server.record_metrics(20, 1.2, 9e-4, 0.85);
let csv = server.format_metric_csv();
assert!(csv.contains("step"));

Fields

config: DashboardConfig

loss_history: MetricHistory

lr_history: MetricHistory

grad_norm_history: MetricHistory

connected_clients: usize

messages_sent: u64

Implementations

impl DashboardServerExt

pub fn new(config: DashboardConfig) -> Self

Creates a new in-process dashboard server with the given configuration.

pub fn record_metrics(&mut self, step: u64, loss: f32, lr: f32, grad_norm: f32)

Records one training step into the internal metric histories.

pub fn broadcast_message( &mut self, msg: &DashboardMessage, ) -> Result<(), DashboardError>

Appends a DashboardMessage to the internal broadcast buffer.

Returns DashboardError::BufferFull when the buffer is at capacity.

pub fn buffered_messages(&self) -> &[DashboardMessage]

Returns all messages currently in the broadcast buffer (in order).

pub fn clear_buffer(&mut self)

Clears the broadcast buffer.

pub fn generate_summary_json(&self) -> String

Returns a JSON summary of all metric histories.

Includes the latest value, a 10-point trend (slope), and EMA (α = 0.1) for each of loss, lr, and grad_norm.

Example

use trustformers_debug::dashboard_ws::metrics::{DashboardConfig, DashboardServerExt};

let mut s = DashboardServerExt::new(DashboardConfig::default());
s.record_metrics(1, 2.0, 1e-3, 0.5);
let json = s.generate_summary_json();
assert!(json.contains("\"loss\""));

pub fn format_metric_csv(&self) -> String

Exports all recorded metrics as a CSV string.

Columns: step,loss,lr,grad_norm Rows are aligned by index — if one history is shorter it writes empty cells.