Pian Pawakapan
Pian Pawakapan
Hi, I'm interested in working on/supporting the Text-to-speech issue, please let me know if anyone else is :)
Same here, this would be extremely helpful!
It might be possible to generate an empty graph dataset that has the same format as COCO-SGG.h5, but which has no nodes or edges for each image. I haven't tried...
Very strangely I'm able to resolve this by adding `import habitat` before everything else: `import habitat; import torch; import quaternion` now runs completely fine.
> Trusting the tests and the summary, the changes are great! Mostly skimmed over the code. Thank you!
@borisfom we just merged this [PR](https://github.com/pytorch/pytorch/pull/127129) in, which allows those complex guards to be expressed as runtime asserts in the graph, allowing compilation to proceed. Could you retry it? It...
@borisfom I see, it looks like the runtime assertions use a sym_not call_function node for the NotEquals guard being generated, and ONNX doesn't support that. A PR recently got merged...
What's the error log you're seeing on the NeMo test case?
@borisfom I see, I think in this case these are meaningful suggested fixes? Export is doing model tracing, and has figured out that the computation graph somehow ties together b1...
@borisfom Thanks for the feedback! I'm now able to reproduce the test cases you've posted. On the export side, after rebasing on this [PR](https://github.com/pytorch/pytorch/pull/127554), and applying the suggested fixes, I...