Pulse

A live ML training debugger - GUI or CLI, any backend.

from pulse import auto_track
if __name__ == "__main__":
    auto_track()
    #training function

GUI mode opens a matrix picker with live shapes, then a live dashboard: a heatmap grid on the left (click a tile to enlarge, right-click to reconfigure axes) and an AI chat panel on the right that's briefed on its role and can optionally see your training code. Scalars (loss, accuracy, lr - anything shape ()) render as a live line chart instead of a heatmap. Loss-like scalars are auto-detected and pre-selected so you don't have to hunt for them every run.

CLI mode is for Colab, SSH, or anywhere headless. No heatmaps are shown on screen - matrices/tensors are "tagged" as text (stats printed each step) and scalars get a live ASCII chart. You can optionally save labeled PDF snapshots per variable per step.

Backends: NumPy, PyTorch, TensorFlow, CuPy, JAX - detected automatically. Pulse never checks for torch/tf/etc. directly; it only ever talks to its own backend-abstraction module.

Install

pip install pulse-ml

tkinter is required for GUI mode and ships with most Python installs. On Debian/Ubuntu, if it's missing:

sudo apt install python3-tk

To also track PyTorch, TensorFlow, CuPy, or JAX tensors, install the matching extra:

pip install "pulse-ml[torch]"
pip install "pulse-ml[tensorflow]"
pip install "pulse-ml[cupy]"
pip install "pulse-ml[jax]"

AI chat / analyst

Set the relevant provider's API key as an environment variable (e.g. ANTHROPIC_API_KEY, OPENAI_API_KEY) to enable the AI chat panel - or leave it unset and Pulse will prompt you for one the first time you send a message.

License

Proprietary. See LICENSE. Use of this software is governed by the terms in that file - copying, redistribution, and reverse engineering are not permitted.