AI Model Debugging: Find the First Divergence Across AI Accelerators
From IBM Technology · · 14 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.
If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.
Description
Learn more about AI Debugging here → https://ibm.biz/~i26jCJsKx
The same AI model can produce different results across accelerators. Kaoutar El Maghraoui explains AI model debugging by finding the first observed divergence between executions. Learn how tensor comparisons, tolerances, and minimal tests help isolate failures.
00:00 – Debug Different AI Model Results Across Accelerators 00:53 – Trace Errors Through the AI Software Stack 02:40 – Separate Expected Variation from Correctness Bugs 03:05 – See How Small Differences Propagate Through Models 04:45 – Localize the Difference Before Diagnosing It 05:52 – Build a Validated Reference Execution 06:50 – Find the First Observed Divergence 08:42 – Compare Tensors with Numerical Tolerances 10:23 – Isolate the Operation with a Minimal Test 12:32 – Turn the Fix into a Regression Test
Continue your education in the course Debugging Across the Dataflow Stack on Coursera here → https://ibm.biz/~NdKCGTusI
AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM → https://ibm.biz/~Tsdh7IC6m
AI was used in the creation of the transcript and metadata for this video.
#aimodel #aiaccelerator #aiengineering #machinelearning --------------------------------------------------------------------------------------------------------- Find us on YouTube: 🔵 IBM Technology: https://www.youtube.com/@IBMTechnology 🔵 IBM: https://www.youtube.com/@IBM 🔵 IBM Developer: https://youtube.com/@IBMDeveloperAdvocates