Support Constant nodes in pytorch parser #1123
Conversation
hls4ml/converters/pytorch_to_hls.py
Outdated
| input_layers.append(input_layer['name']) | ||
| n_inputs += 1 | ||
| if 'const' in node.name: | ||
| pytorch_class = "Constant" |
There was a problem hiding this comment.
We use single quotes wherever possible, let's not mix the two styles.
| def parse_constant_layer(operation, layer_name, node): | ||
| assert 'Constant' in operation | ||
|
|
||
| layer = {} |
There was a problem hiding this comment.
Should we also add empty list as inputs here (and do the same in onnx parser)? The Constant node in the IR will override anyway, but it feels like that feature should be a check and an error, otherwise it will silently changing things making it harder to debug.
There was a problem hiding this comment.
So the suggested change here would just be layer['inputs'] = [], and then we'd add a check to the Constant node to throw an error if the input list is not empty?
There was a problem hiding this comment.
Here and in qonnx parser. Then the self.inputs = [] in Constant layer could be a check and a warning before setting. This could be better for futureproofing for constant nodes coming from various sources (like optimizers). @jmitrevs what do you think?
There was a problem hiding this comment.
I have added it here.
JanFSchulte
added
please test
Fixes #1082.
Constants, for example for multiplications, are recognized and added to the hls4ml model using the
Constantlayer introduced in the QONNX parser. Thanks to @sei-jgwohlbier for finding the issue and designing the fix.Type of change
For a new feature or function, please create an issue first to discuss it
with us before submitting a pull request.
Note: Please delete options that are not relevant.
Tests
Checklist
pre-commiton the files I edited or added.