omnitrack 0.1.0

Unified logging and experiment tracking for research workflows

Omnitrack 🚀

Unified logging and experiment tracking for ML research.
Log once → sync everywhere (console, JSONL, W&B, …).

Features

• 📊 Multi-sink logging (ConsoleSink, JSONLSink, WandbSink)
• 🌲 Hierarchical config & tags
• 🚀 Beautiful Rich-based console
• 🔌 Extensible sink interface
• ⚡ Efficient: batch logging, manual refresh

Quickstart

pip install omnitrack

from omnitrack import LogSession, record, step, push_config, set_tags
from omnitrack.sinks.console import ConsoleSink

with LogSession(sinks=[ConsoleSink()]):
    push_config({"lr": 1e-3, "batch_size": 64})
    set_tags(env="demo")

    for e in range(3):
        step("epoch")
        for b in range(10):
            step("batch")
            record(step_name="batch", loss=0.1, acc=0.9)

Demo

Run the built-in demo:

omnitrack demo --jsonl logs/demo.jsonl