Ziheng Jiang
Ziheng Jiang
Sorry, this problem is caused by the difference between MXNet API and NumPy API, as you can see, in MXNet, the code below cannot work well too: ``` import mxnet.ndarray...
Agree. BTW, should PreferMXNetPolicy utilize the black list information?
Can we approve and merge this PR? @vivekkhandelwal1 @silvasean
We are working on the transposed convolution on the Torch2MHLO lowering side so checked this PR to see whether we can merge it. It's ok to wait the upstream PR...
The PR on the PyTorch side is accepted. Can we merge this PR now? @vivekkhandelwal1 @silvasean
I also learned that @fortianyou is working on the similar thing. Discussion and collaboration are more than welcome.
Thanks for all the discussion. Looking forward to today's meeting. FYI, here is the POC(https://github.com/llvm/torch-mlir/pull/1025) for this RFC, in which we have implemented BERT and ResNet conversion from Torch to...
Hi @silvasean Thanks for the suggestion! Dynamic shape support is a good point. We are happy to update the PR to support this or collaborate with @fortianyou to distribute those...
According to the offline meeting between Bytedance and Alibaba. We decide to break the proposal into several PRs: - Integrate MHLO into TorchMLIR Repo - Basic conversion pipeline - Operator...
With the ResNet and BERT examples (from Torch to MHLO) has been merged into the main branch, we could say that we have completed the promise we made in the...