Crate burn_telemetry 

Training logger plugins for Burn 0.19

This crate provides MetricLogger implementations that export training metrics to various observability platforms.

Features

OpenTelemetry (otel feature)

• Efficient metric export with minimal overhead
• Support for multi-GPU/node training via resource attributes
• Configurable export intervals and batching
• Compatible with OpenTelemetry SDK and OTLP protocol

Rerun.io (rerun feature)

• Rich visualization of training metrics, images, and tensors
• Support for images, videos, text, and other visual formats
• Real-time visualization during training
• Timeline-based metric exploration

Multi-Logger Support

Use MultiLogger to combine multiple loggers and send metrics to multiple backends simultaneously (e.g., OpenTelemetry + Rerun).

Example (OpenTelemetry)

use burn_telemetry::otel::{OpenTelemetryLogger, OpenTelemetryConfig};

let logger = OpenTelemetryLogger::new(
    OpenTelemetryConfig::default(),
    "my-training-job",
    Some("gpu-0"),
).expect("Failed to create logger");

Example (Rerun)

use burn_telemetry::rerun::{RerunLogger, RerunConfig};

let logger = RerunLogger::new(
    RerunConfig::default(),
    "my-training-job",
).expect("Failed to create logger");

Example (Multiple Loggers)

use burn_telemetry::MultiLogger;
use burn_telemetry::otel::{OpenTelemetryLogger, OpenTelemetryConfig};
use burn_telemetry::rerun::{RerunLogger, RerunConfig};

let multi_logger = MultiLogger::new()
    .add_logger(OpenTelemetryLogger::new(
        OpenTelemetryConfig::default(), "job", Some("device-0")
    ).unwrap())
    .add_logger(RerunLogger::new(
        RerunConfig::default(), "job"
    ).unwrap());

Re-exports

pub use multi_logger::MultiLogger;

Modules

factory

Helper factories for constructing metric loggers based on enabled features.

multi_logger

Multi-logger wrapper for combining multiple MetricLogger implementations

otel

OpenTelemetry logger backend

rerun

Rerun.io logger backend