Skip to main content
GPU Profiler TUI Demo A production-ready, open source tool for real-time GPU memory profiling, leak detection, and optimization in PyTorch and TensorFlow deep learning workflows.

​
Why GPU Memory Profiler?

Training deep learning models often leads to out-of-memory crashes and inefficient resource usage. GPU Memory Profiler helps you understand exactly where your memory is going and how to optimize it.

Prevent OOM crashes

Catch memory leaks and inefficiencies before they crash your training jobs

Optimize performance

Get actionable insights and recommendations for memory usage patterns

Unified interface

Works seamlessly with both PyTorch and TensorFlow frameworks

Beautiful visualizations

Timeline plots, heatmaps, and interactive dashboards for deep insights

​
Key features

Track GPU memory usage in real-time with live dashboards and CLI tools. Monitor allocated memory, reserved memory, and memory fragmentation as your model trains.
Automatic detection of memory leaks with configurable warning and critical thresholds. Get alerts when memory usage exceeds expected patterns.
Profile specific functions and code blocks using decorators and context managers. Understand memory usage at a granular level.
Generate timeline plots, heatmaps, and interactive dashboards. Export data to CSV, JSON, PNG (Matplotlib), or HTML (Plotly) formats.
Powerful command-line interface for profiling, monitoring, and diagnostics. Integrate into CI/CD pipelines and automated workflows.
Works on laptops and CI agents without CUDA. Falls back to psutil-powered CPU memory tracking when GPU is not available.

​
Framework support

from gpumemprof import GPUMemoryProfiler

profiler = GPUMemoryProfiler(track_tensors=True)

with profiler.profile_context("training"):
    output = model(data)
    loss = criterion(output, target)
    loss.backward()
    optimizer.step()

summary = profiler.get_summary()
print(f"Peak memory: {summary['peak_memory_usage'] / (1024**3):.2f} GB")

​
Requirements

  • Python 3.10 or later
  • PyTorch 1.8+ (for PyTorch profiling)
  • TensorFlow 2.4+ (for TensorFlow profiling)
  • CUDA-capable GPU (optional - falls back to CPU mode)

​
Get started

Installation

Install via pip or from source with optional dependencies

Quick start

Get up and running with your first memory profile in 5 minutes

CLI usage

Learn about command-line tools and automation

​
Terminal UI

Prefer an interactive dashboard? Install the optional TUI dependencies and launch the Textual interface: 
pip install "gpu-memory-profiler[tui]"
gpu-profiler
The TUI surfaces system info, PyTorch/TensorFlow quick actions, live monitoring with real-time charts, and CLI automation tools.

​
Use cases

Training optimization

Profile your training loops to identify memory bottlenecks and optimize batch sizes

Leak detection

Find memory leaks in long-running training jobs before they cause crashes

Model debugging

Debug out-of-memory errors with detailed snapshots and stack traces

Production monitoring

Monitor GPU memory usage in production deployments and inference services

​
Community

GitHub

View source code, report issues, and contribute

PyPI

Latest releases and package information
Current version: 0.2.0 (launch candidate)License: MIT

Build docs developers (and LLMs) love

Get started for freeTalk to us