jiangjiajun edited a comment on pull request #8645: URL: https://github.com/apache/tvm/pull/8645#issuecomment-898309823
@leandron PaddlePaddle supports exporting model to disk, the whole model contains 2 files, the `model_name` is user defined - `model_name.pdmodel `: stores the model structure information - `model_name.pdiparams` : stores the model weights here is a PaddlePaddle resnet50 model: https://bj.bcebos.com/x2paddle/models/paddle_resnet50.tar For PaddlePaddle, we load the exported model using the following codes ``` import paddle model = paddle.jit.load('dirname/model_name') ``` we pass the model path but not include the suffix. I think it's necessary to implement a `tvmc` frontend for PaddlePaddle, which will bring better experience for users. We can use `tvmc` to compile PaddlePaddle model like ``` tvmc compile \ --target "llvm" \ --output resnet50-v2-7-tvm.tar \ dirname/model_name ``` How do you think about this solution? And is it better that I put the `tvmc` frontend implementation in this pull request, or send another pull request? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
