Justin Chu

icon company @microsoft icon location Code for the better. Microsoft AI Frameworks. Previously @ X, the Moonshot Factory (Alphabet); CMU

Results 444 comments of Justin Chu

The PixelUnshuffle op Cannot be converted to SpaceToDepth

You may use the rewriter to rewrite the graph as you wish: https://onnxscript.ai/tutorial/rewriter/rewrite_patterns.html

Any operation to convert "Tile" to "Slice"

You may consider using the rewriter for this purpose: https://onnxscript.ai/tutorial/rewriter/rewrite_patterns.html

Please specify the behavior of min > max in Clip

cc @gramalingam

Test using bfloat16 and model input/output

Thanks. We can enable bfloat16 support with the ml_dtypes package in the reference evaluator.

onnx.shape_inference.infer_shapes creates a blank graph when passed a >2GB model

It's most likely that the model proto provided is not serialized correctly because it's larger than 2GB. I talked with @carzh. We suggest loading the model without weights to run...

Could you make a release?

Bump

Could you make a release?

@suo Just checking again - thanks!

Could you make a release?

Friendly bump on this @suo

Feature request: Implement GroupNormalization-21

GroupNormalization - 21 Version name: [GroupNormalization (GitHub)](https://github.com/onnx/onnx/blob/main/docs/Operators.md#GroupNormalization) domain: main since_version: 21 function: True support_level: SupportType.COMMON shape inference: False This version of the operator has been available since version 21. Summary...

Why does SplitToSequence not allow zeros in `split` input?

@gramalingam