Is it possible to support flatbuffer 2.0?

[littleday]

Hi team, we just started to use onnx models to speed up our model performance. However, the infrastructure requires flatbuffer 2.0. Is it possible to support it in the next release?

[hwangdeyu]

Was the flatbuffer 2.0 blocked you? I can use tf2onnx successfully in flatbuffer 2.0.

[ahoereth]

setup.py seems to pin flat buffers to 1.12 , as of #1511.

[bryce-evans]

The latest tensorflow-onnx throws errors when used with the latest tensorflow. For many projects, this is a non-starter.

It's been 18 months since the offending code was committed and 6+ months with no response from the team. Is this going to be handled (i.e. modifying the pin at setup.py) at some point?

[ahoereth]

Looks tensorflow-onnx now installs flatbuffers 2.0, pinned to flatbuffers>=1.12 since #2050 and to flatbuffers<3.0,>=1.12 on current main as of #2057.