Freezing and training submodules of model #17444
Unanswered
anupsingh15
asked this question in
code help: CV
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Hey folks!
I have a model which consists of 3 submodules( A, B, and C) stacked sequentially. The pipeline of the model looks like A–>B–>C. For the first few epochs, I train the A–>B and compute the loss; thereafter, I freeze the A and B submodules and train only C. The forward pass looks like this:
Note that I also reset the Adam optimizer to train the C component.
I did not get the desired results with this pipeline. However, if I train a model which consists of only A–>B, and store its outputs (z = B(A(batch))) externally as a database after the training, and then use this database to train a model which contains only C, then it works for me. Basically, I have to break the pipeline into two to get results rather than just having a single pipeline. Any suggestions on where I might be doing something wrong?
Beta Was this translation helpful? Give feedback.
All reactions