Closed jmitrevs closed 4 months ago
My understanding of when the dimensions include a batch dimension and when it didn't seems to not be right. Will investigate. For ONNX can't depend on the batch dimension being None.
This is being replaced with PR #979.
Description
This change updates the ONNX parser and adds support for QONNX. It replaces PR #591. It only supports ONNX that has been cleaned by the qonnx package, including converting convolutions to be channels-last and changing
Gemm
toMatMul
andAdd
.In QONNX
Quant
nodes can act on constants as well as the datapath. To make handling this easier, we explicitly put constants in the initial graph. There are also some helper nodes likeMatMul
andConv
that are introduced to support the explicit constant nodes. After theconvert
flow, no special ONNX nodes remain in the graph, though.Generally Quant nodes that have power-of-2 scales and no zero-offset get converted to fixed data types either by setting the types of constants or adding a linear activation that is usually merged into preceding nodes. Non-power-of-2 scales result in
ApplyAlpha
nodes beings added to scale and unscale, with propagation across some layers. This can be further optimized and has generally been tested less.Binary networks are not yet supported.
Currently some of the automatic type setting depends on QONNX-set attributes. When we introduce auto type values, this should be updated accordingly.
Type of change
Tests
The pytest,
test_qonnx.py
, is the main test, building some models from the QONNX model zooChecklist
pre-commit
on the files I edited or added.