pulse / ml debugging
pulse / live solutions github pypi
pulse / CLI + headless 001 — 005
residual / primary product

pulse

a live ML training debugger

01 / TRACK
see what the model sees.
Pulse tracks numerical state while training is running.
tensors
gradients
activations
loss
01 / 05
automatic progression
scroll to inspect
01 / SOLUTIONS

built around
the run.

Residual builds tools for the parts of AI development that become difficult once models leave the notebook. The first product is Pulse, a live ML training debugger designed for CLI and headless environments.
residual / solution 01
pulse
cli / headless
Pulse is a live ML training debugger for CLI and headless environments. It combines runtime observability, deterministic numerical checks, ML-specific linting, and an AI debugging agent to investigate failures while training is running.
01
Track
02
Diagnose
03
Verify
04
Patch
environments
terminals, SSH sessions, containers, remote servers, cloud GPU machines, Google Colab
backends
NumPy · PyTorch · TensorFlow · CuPy · JAX
runtime
tensors · gradients · activations · weights · loss · metrics · scalar histories
verification
deterministic numerical checks, ML-specific linting, source inspection, verification and minimal patches
tracking
CPU-first by default. GPU inspection is explicitly opt-in.
interface
CLI-first, with no graphical interface required
install pulse
pip install pulseml
dashboard github pypi
pulse / terminal
pulse
live ML training debugger
pip install pulseml
github pypi dashboard