AI Development & ML EngineeringCurated
Training Failure Diagnostic
For training runs where guessing wastes GPU hours.
The prompt
My model is [describe failure, e.g. 'not converging' / 'overfitting']. Here is the setup: [paste]. Diagnose the likely cause, ordered by probability, and the test for each. Do not suggest all fixes at once.
Anything in square brackets is meant to be replaced. The builder turns those into fields you can type into.
Tired of re-pasting the same context into every new chat? That is what the Unimatrix MCP server is for.