Thanks for the comments - they have helped to shed light on things - :bulb:!
I agree with the removal of the `output_map_index`, though my suggestion was to
actually use this as a redirect to the same entries for the user supplied
names. I.e. It would have entries for the user specified names (from
`_get_relay_input_vars`) and then you add some re-directs to these same
`outputs` list entries for the pytorch names.
But as you say, we could just add some extra output entries if we turn
`outputs` into a dictionary, or just use the setDebugName to change the graph -
I will have a look into both.
FYI, my initial method (just as a trial):
* Read in user input_shapes and create simple conversion map from pytorch to
user `{ 'pytorch input.1': 'user input 1' }
* `_get_relay_input_vars` is still used to construct outputs from the user
input_shapes, and gets added to outputs - so the user names are already there
* in `_get_op_inputs` I use the new conversion map to convert the pytorch names
(from `_get_input_names`) into the user names before looking them up in the
output_map_index - heres where I could have just used the output_map_index
instead of an extra map - I was being a bit overly cautious.
---
[Visit
Topic](https://discuss.tvm.ai/t/pytorch-frontend-graph-input-names-can-change-using-loaded-torchscript/6055/16)
to respond.
You are receiving this because you enabled mailing list mode.
To unsubscribe from these emails, [click
here](https://discuss.tvm.ai/email/unsubscribe/ca902c6797b01f5c0a02474c49ce6ce5659989ff0f19b559382dc8b0c132bb74).
