Closed bedapisl closed 2 years ago
Hello @bedapisl,
We are investigating this issue. In the mean time, I would point you at the --fast-math
flag (documented here). This flag allows you to control tradeoff between performance and accuracy for fp32 operators.
We will update this issue as we learn more.
-Taylor
Hello @bedapisl,
We could reproduce and root cause the issue. We are currently working on a fix. Will update once its fixed and tested.
Thanks, Shruthi.
Hello @bedapisl,
We have implemented a fix for the aten::mul operator translation that will resolve the numerical mismatch issues you are seeing. The fix is expected in one of our upcoming releases.
Thanks, Shruthi.
@bedapisl we will update this ticket with the release details that includes the fix
Hi
This has been resolved in the recent 1.17.0 release.
Please find more information here:
Hello, I converted my model to Neuron, but it is now giving different results then the original Pytorch model. Output probabilities are different and final predictions are different in about 10% of cases. Example code:
which gives the following output:
Note how different are the model outputs on the last lines. Neuron gives probabilities
[0.3261, 0.3678, 0.3075]
but Pytorch gives[0.1643, 0.6104, 0.2253]
. The model is DistillBERT with added linear layer: