-
Notifications
You must be signed in to change notification settings - Fork 6
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Run from run_ID #68
Run from run_ID #68
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Beyound my last comment, it seem to me that some changes are missing for this pull request to be completed
alonet/common/pl_helpers.py
Outdated
@@ -71,6 +71,7 @@ def add_argparse_args(parent_parser, add_pl_args=True, mode="training"): | |||
parser.add_argument("--project_run_id", type=str, help="Project related with the run ID to load") | |||
parser.add_argument("--expe_name", type=str, default=None, help="expe_name to be logged in wandb") | |||
parser.add_argument("--no_suffix", action="store_true", help="do not add date suffix to expe_name") | |||
parser.add_argument("--nostrict", action="store_false", help="load from checkpoint to run a model with different weights (default True)") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Usually if one use --nostrict the expected behavior is : Do not strictly load the weights (store true on nostrict)
But here --nostrict set nostrict to false. It seem to me to be counter intuitive.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
now if --nostrict is specified (store True and False by default) it is possible to load a checkpoint with different keys in the load_state_dict. BUT it is still not possible for a checkpoint with same keys but different values (weights dimensions).
when launching a train after having loaded a checkpoint, parameters of the initial model were not kept and default parameters replaced them. It's fixed.
new parse argument "--nostrict" used if we want to run a model from a checkpoint, with different coefficients between the model and the model that generated the checkpoint.