Closed iProzd closed 2 months ago
The update involves refining the function replace_model_params_with_pretrained_model
within the finetune.py
module. A previously commented-out TODO note and its related but unused logic, which concerned the retention of certain non-structural parameters, have been removed. This change simplifies the function by eliminating unnecessary comments and inactive code.
File Path | Change Summary |
---|---|
.../tf/utils/finetune.py |
Removed a commented-out TODO and related unused logic. |
Thank you for using CodeRabbit. We offer it for free to the OSS community and would appreciate your support in helping us grow. If you find it useful, would you consider giving us a shout-out on your favorite social media?
I've checked for #3551 , there are several params that need discussed whether to keep from new input script (now we only keep trainable
params and replace others with those in the pre-trained model):
exclude_types
from descriptor
atom_ener
from EnerFitting
Do we need to keep these params from user-defined fine-tuning input script? @njzjz @wanghan-iapcm If not, there's nothing to change for this issue.
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 79.28%. Comparing base (
4b319a0
) to head (69d0973
).
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Do we need to keep these params from user-defined fine-tuning input script?
Why not provide an option to users?
Use this issue to track this need: https://github.com/deepmodeling/deepmd-kit/issues/3869
Summary by CodeRabbit