Xida Ren

sorry. I don't know shit about vim. I meant to say I would like vimsplain to explain keypress sequences like `:wq` `:%s/a/b/gc` Things that start with a colon.

aw okay. I'm looking forward to that bugfix. btw, awesome project. I really appreciate it. I get difficult-to-understand vimcommands from stackoverflow all the time, and vimsplain has already helped me...

I'm now writing a module to implement this, similar to the linux_perf module. But if there's a better way to do so (e.g. by forking and editing a benchmark configuration...

Was going to file an issue, but found this. I often need to kill tabs en-masse. Would love to have this.

the prhase "Multidimensional array modeling Torch's Tensor type" comes from `include/torch-mlir/Dialect/Torch/IR`: ```cpp class AnyTorchTensorType : Torch_Type { let summary = "Multi-dimensional array modeling Torch's Tensor type"; ```

https://github.com/llvm/torch-mlir/pull/2904 here's my earlier fix for the unsqueeze issue. That one should fully lower to linalg with no problem. feel free to cannibalize it to make this work. from what...

@vivekkhandelwal1 oh consider adding support for the lower versions of unsqueeze. Here's an IREE issue where somebody ran into a wall with an opset7 onnx file https://github.com/openxla/iree/issues/16339

Test case draft: lstm.onnx.mlir ```mlir module { func.func @lstm(%arg0: !torch.vtensor, %arg1: !torch.vtensor, %arg2: !torch.vtensor, %arg3: !torch.vtensor) -> (!torch.vtensor, !torch.vtensor, !torch.vtensor) attributes {torch.onnx_meta.ir_version = 9 : si64, torch.onnx_meta.opset_version = 20 :...

Currently testing with: ```bash /home/azureuser/torch-mlir/build/bin/torch-mlir-opt -pass-pipeline='builtin.module(func.func(convert-torch-onnx-to-torch))' lstm.onnx.mlir -o lstm.torch.mlir ~/torch-mlir/build/bin/torch-mlir-opt --mlir-print-debuginfo --mlir-elide-elementsattrs-if-larger=16 --mlir-print-stacktrace-on-diagnostic --mlir-disable-threading --mlir-print-ir-after-failure --mlir-print-ir-module-scope -pass-pipeline='builtin.module(torch-backend-to-linalg-on-tensors-backend-pipeline)' ./lstm.torch.mlir -o ./lstm.linalg.mlir 2>&1 | tee torchtolinalg.log ```

encountered a bunch of issues earlier, fixed thanks to Rob. Currently stuck on: ./lstm.torch.mlir:61:13: error: 'torch_c.to_builtin_tensor' op operand #0 must be Multi-dimensional array modeling Torch's Tensor type, but got 'tensor'...