search

tensorflow / mesh

Mesh TensorFlow: Model Parallelism Made Easier
Apache License 2.0
1.59k stars 255 forks source link

Mesh-tf model conversion to onnx? #368

Open b-analyst opened 2 years ago

b-analyst commented 2 years ago

Hi, I've been trying to deploy an mtf model to the NVIDIA Triton Inference Server by converting the SavedModel (output of model.export()) to an onnx file with no luck. I've been receiving several errors, but the main recurring one seems to be regarding the lack of a registered SentencePieceOps. Is there a tutorial for an mtf model deployment available anywhere?

WingsOfPanda commented 2 years ago

hi @b-analyst , I am just wondering that are you working on image models like u-net or language models like Bert? thx!

b-analyst commented 2 years ago

Hi! Thank you for replying. I am using a language model - T5 to be specific. I've been using the mtf model implementation from Google-research's GitHub. I managed to convert it to a Pytorch model, and I've been successful at converting to onnx.