Hey everyone,
I am Sundaram Gupta, the developer behind TorchQuery. I am currently restructuring the project framework to launch a dedicated notebook-focused ecosystem called torchquery-colab on PyPI, and I am looking for passionate deep learning engineers and open-source contributors to join me as core maintainers!
What is torchquery-colab?
torchquery-colab is an interactive, hardware-accelerated tensor manipulation framework built specifically for Jupyter Notebooks and Google Colab (.ipynb) environments. Its primary goal is Neural Healing—surgically tracking down and repairing corrupted NaN and Inf tensor values across massive pipelines using fast, vectorized operations directly in GPU VRAM without interrupting live training loops.
To protect cloud notebook kernels from crashing, it enforces a strict 1 GB CUDA memory isolation boundary using dynamic byte tracking and custom interactive UI/Widgets embedded inside the cells.
What we need help with right now:
-
CUDA Optimization & Benchmarking: Refining our
tqc.gpuecker()hardware diagnostics profile and optimizing dynamic masking speeds on various NVIDIA architectures (Tesla T4, A100, H100). -
Interactive Frontend Enhancements: Designing custom IPython/ipywidgets interactive UI controls (like the “Install Colab in Torchquery” button sequence).
-
Robust Pipeline Testing: Writing testing suites to benchmark memory-safe element-wise transformations on large-scale PyTorch operations.
Who I am looking for:
If you love playing with PyTorch internals, tracking VRAM utilization, writing clean CUDA-accelerated operations, or building beautiful interactive widgets for notebook users, I’d love to add you as an official Maintainer on PyPI and GitHub.
If you are interested in collaborating, please drop a comment below with your background, or reach out to me directly with your GitHub / PyPI username! Let’s build a safer, faster training safety net for the PyTorch community together.
Best regards,
Sundaram Gupta Creator of TorchQuery