Alexandre Eichenberger
Alexandre Eichenberger
# Feature Request Right now, the ONNX model zoo consists of one batch of models, which have been accumulated over the years. Some include quite recent models coded with recent...
Looking at the bertwquad12.onnx model (https://github.com/onnx/models/tree/main/text/machine_comprehension/bert-squad/model) When I do shape inference with the `?` dynamic values replace by 4, the final model is fully shaped with constant shapes ``` bash...
Looking at the bertwquad12.onnx model (https://github.com/onnx/models/tree/main/text/machine_comprehension/bert-squad/model) ### STATIC I took a smaller sample and with static shapes, it works really well now. [bert349c.txt](https://github.com/onnx/onnx-mlir/files/9261717/bert349c.txt) results in all of the types being...
According to Andrew/Tin, we can transform NCHW directly to stick and the code should be quite efficient. They don't use yet the C=1 => transpose become reinterpret cast, so we...
Idea: put here a quick comment to claim the operations that you care currently working on, so that we do not replicate work. Can also add a request for new...
We should increase our doc to get info on how to use ONNX-MLIR and then going to MLIR or LLVM There are a couple of users that asked these questions...
There was an issue with numerical tests of matmul and gemm with random number beyond the -1.0 to 1.0 range. Testing with disabling the fused multiply-add succeeded when performing 10,000...
I wonder if the name of our runtime library is not too generic to be integrated with multiple libs in the traditional system wide libs. Maybe we should have something...
I have a few general questions about our Runtime and Compiler interface. As I am working on a benchmarking benchmark, it is making me revisit our interfaces as I am...
Decisions: We need both command line options and environment variable for many of the compiler options. Here is how we can keep sanity 1. Options to ONNX-MLIR are self described...