Exception generating exporting a model with prior
blazejkotowski opened this issue · comments
-
I have preprocessed the data following way:
rave preprocess --channels 1 --input_path <input_path> --output_path <db_path>
-
Then I trained the model with following command
rave train --batch 2 --channels 1 --db_path <db_path> --name <run_name> --out_path <run_out> --val_every 2500 --config v2.gin
-
Then I trained the prior
rave train_prior --db_path <db_path> --name <prior_name> --out_path <prior_out_path> --val_every 2500 --gpu 0
-
Finally, on the export with the following command
rave export --name organizmo_fire --streaming true --run <run_out_path> --prior <prior_out_path> --output ts_models/
I get the following exception
Traceback (most recent call last):
File "/miniconda/bin/rave", line 8, in <module>
sys.exit(main())
^^^^^^
File "/miniconda/lib/python3.11/site-packages/scripts/main_cli.py", line 38, in main
app.run(export.main)
File "/miniconda/lib/python3.11/site-packages/absl/app.py", line 308, in run
_run_main(main, args)
File "/miniconda/lib/python3.11/site-packages/absl/app.py", line 254, in _run_main
sys.exit(main(argv))
^^^^^^^^^^
File "/miniconda/lib/python3.11/site-packages/scripts/export.py", line 566, in main
scripted_rave = script_class(
^^^^^^^^^^^^^
File "/miniconda/lib/python3.11/site-packages/scripts/export.py", line 197, in __init__
self.register_method(
File "/miniconda/lib/python3.11/site-packages/nn_tilde/__init__.py", line 102, in register_method
y = getattr(self, method_name)(x)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/miniconda/lib/python3.11/site-packages/scripts/export.py", line 345, in prior
return self.prior_module.forward(temp)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/miniconda/lib/python3.11/site-packages/scripts/export.py", line 464, in forward
x[..., i:i + 1] = self.step_forward(temp)
^^^^^^^^^^^^^^^^^^^^^^^
File "/miniconda/lib/python3.11/site-packages/scripts/export.py", line 445, in step_forward
self.previous_step.copy_(x.clone())
RuntimeError: output with shape [1, 128, 1] doesn't match the broadcast shape [4, 128, 1]
At the same time, I am able to export the model without prior. What is the issue?
I have the same problem, I think @ballerburg9005 suggests to train with msprior here: #300
But I'm not sure it works, and how.
Same error here. Has anyone managed to train and export a prior (not msprior) model that can be used with nn~
, using a version of this repo later than v1?