microsoft/tensorwatch
Debugging, monitoring and visualization for Python Machine Learning and Data Science
Supports interactive "Lazy Logging Mode" where clients execute arbitrary Python queries against live training processes via ZeroMQ, returning filtered streams for visualization—enabling real-time inspection without pre-logging. Built on a composable stream architecture where files, sockets, visualizers, and custom processors are unified as streams that can chain together into dataflow graphs. Integrates with PyTorch (0.4-1.x), TensorFlow eager tensors, and Jupyter Notebooks for in-notebook real-time dashboards, with graphviz support for network architecture diagrams.
3,466 stars and 453 monthly downloads. Used by 1 other package. Actively maintained with 6 commits in the last 30 days. Available on PyPI.
Stars
3,466
Forks
360
Language
Jupyter Notebook
License
MIT
Category
Last pushed
Mar 06, 2026
Monthly downloads
453
Commits (30d)
6
Reverse dependents
1
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/microsoft/tensorwatch"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related frameworks
tensorflow/tensorflow
An Open Source Machine Learning Framework for Everyone
KomputeProject/kompute
General purpose GPU compute framework built on Vulkan to support 1000s of cross vendor graphics...
hidet-org/hidet
An open-source efficient deep learning framework/compiler, written in python.
NVIDIA/nvshmem
NVIDIA NVSHMEM is a parallel programming interface for NVIDIA GPUs based on OpenSHMEM. NVSHMEM...
NVIDIA/DIGITS
Deep Learning GPU Training System