epochmonitor 1.0.0

A simple and versatile monitor/profiler for model training loops.

EpochMonitor ⏱️

A simple, lightweight, and versatile profiler for monitoring your machine learning training loops. EpochMonitor helps you track training time, CPU/RAM usage, and NVIDIA GPU memory usage with minimal code changes.

Get crucial insights into your model's performance and resource consumption, all presented with a clean, dynamic progress bar.

[Image of a dynamic terminal progress bar]

---

Features

• Easy Integration: Use it as a simple function decorator or a flexible context manager.
• Comprehensive Metrics: Tracks epoch duration, total training time, CPU utilization, and RAM usage (in % and MB).
• NVIDIA GPU Support: Automatically detects NVIDIA GPUs and monitors memory usage per epoch.
• Dynamic Display: Uses tqdm to provide a clean, non-disruptive progress bar that updates in place.
• Flexible Logging: Automatically saves a detailed history of all metrics to a CSV or JSON file for later analysis.
• Multi-GPU Aware: Allows you to specify which GPU to monitor on systems with multiple cards.

---

Installation

You can install EpochMonitor directly from PyPI:

pip install epochmonitor

Usage

As a Decorator (Simplest Method)

Just add @EpochMonitor() on top of your training function.
The monitor object will be injected into your function as a keyword argument.

import time
from epochmonitor import EpochMonitor
from tqdm import tqdm

@EpochMonitor(log_file_prefix="my_model_log", file_format="json")
def train_my_model(epochs, learning_rate, monitor=None):
    for epoch in tqdm(range(epochs), desc="Training Model"):
        monitor.start_epoch()
        print(f"-> Training with lr={learning_rate}...")
        time.sleep(2)  # Simulating work
        monitor.end_epoch()

train_my_model(epochs=5, learning_rate=0.01)

As a Context Manager (More Control)

Use a with statement for explicit control.

import time
from epochmonitor import EpochMonitor
from tqdm import tqdm

def another_training_run(epochs):
    with EpochMonitor(log_file_prefix="context_run_log") as monitor:
        for epoch in tqdm(range(epochs), desc="Training Model"):
            monitor.start_epoch()
            time.sleep(1.5)  # Simulating work
            monitor.end_epoch()

another_training_run(epochs=3)

Listing Available GPUs

If you have multiple GPUs, list them first:

from epochmonitor import EpochMonitor

# Prints all detected NVIDIA GPUs and indices
EpochMonitor.list_gpus()

# Example: Monitor GPU at index 1
# @EpochMonitor(gpu_index=1)
# def train_on_second_gpu(...):
#     ...

Contributing

Contributions are welcome! Whether it's:

Reporting a bug 🐛

Suggesting a feature 💡

Submitting a pull request 📥

Please read the Contributing Guidelines before starting.

License

This project is licensed under the MIT License. See the LICENSE file for details.