zhiics commented on a change in pull request #5770:
URL: https://github.com/apache/incubator-tvm/pull/5770#discussion_r441302269
##########
File path: python/tvm/contrib/graph_runtime.py
##########
@@ -162,7 +162,11 @@ def set_input(self, key=None, value=None, **params):
keys = list(params.keys())
keys.sort(key=lambda x: -np.prod(params[x].shape))
for k in keys:
- self._get_input(k).copyfrom(params[k])
+ # TODO(zhiics) Skip the weights for submodule in a better way.
+ # We could get all inputs required by graphruntime first,
+ # we should use MetadataModule for initialization.
+ if "_const_" not in k:
Review comment:
Yes, the constants for graph runtime all start with "p"
https://github.com/apache/incubator-tvm/blob/3e72be58f362a21dbcc1de36f9dbed216e854baf/src/relay/backend/graph_runtime_codegen.cc#L314
However, as the TODO says, we should have a better way to handle this. But
graph runtime current doesn't really distinguish input/weights. I think we can
use this simple workaround from the frontend for now.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
