-
Notifications
You must be signed in to change notification settings - Fork 3
Description
Hi,
This project is beneficial for me as a beginner of generative modelling. However, I cannot reproduce the results (action-matching/notebooks/visualize.ipynb) using the ssm method while the am method worked well. Could you provide the config (config_path_ssm = '../checkpoint/8724464/ssm_hydrogen.config') and the checkpoints (state = torch.load(config.model.checkpoints[2])). Thanks.
In my experiments, I only change am to ssm without any other modifications of the config (hydrogen.py), but it did not work.
Using the AM method and the model ....70000.cpt (Only show the results of the first variable)
Using the SSM method and the model ....20000.cpt (Only show the results of the first variable)
The loss cannot be reduced based on the provided config (only changing the method am to ssm).
Best,
Qunxi


