Louati
Louati
@NOT-HAL9000 For the moment we don't support ByT5 models but it should not be very different from T5 models. I think it can work with some minor manipulations and modifications...
@pngmafia Hello, can you please post the full script that you are trying to test ?
@brevity2021 we are working on adding support for t5 conversion using the convert script. I think it should cover the precision for different T5 models (including t5-large).
@pommedeterresautee I'm still testing triton but i need your review for the other parts (conversion onnx and tensorRt) and i'd like to see if those conversions can run on your...
@tanmayb123 Currently, we are not planning to open those parameters, you can try either to add parameters with Triton or to try to pass the wanted parameters in a json...
@pommedeterresautee all tests (backward benchmark) passed on cloud-dev (without cuda-graphs):  All tests (foraward and backward benchmakrs) passed: 
@pommedeterresautee can you please add your `kernel123` so that we do the measures on the same one ?
hello @WissamAntoun, Thanks for your feedback, can you please share the issues you're facing with TensorRT ?
@gaetansnl which test please and which comment ?
Hello, Thanks for trying our library, We are actually working on adding T5 officialy in the convert script so that you can do conversion with one line command, It will...