"Hi everyone,
I am currently working on reproducing the training process for this repository and would like to clarify some observations regarding the loss behavior.
-
Training Experience: Has anyone successfully completed the training process from scratch? I would appreciate it if you could share your experience.
-
Optimal Performance: At which stage (epoch/iteration) does the model typically achieve its best results?
-
Loss Plateau: Is it normal for the training loss to remain stagnant for a long period? In my current run, the loss curve seems to have plateaued (horizontal trend) and hasn't shown significant changes for quite a while.
Is this plateauing behavior expected in the early or middle stages of training for this model? Any insights on the typical convergence pattern would be very helpful.
Thanks in advance for your help!"
"Hi everyone,
I am currently working on reproducing the training process for this repository and would like to clarify some observations regarding the loss behavior.
Training Experience: Has anyone successfully completed the training process from scratch? I would appreciate it if you could share your experience.
Optimal Performance: At which stage (epoch/iteration) does the model typically achieve its best results?
Loss Plateau: Is it normal for the training loss to remain stagnant for a long period? In my current run, the loss curve seems to have plateaued (horizontal trend) and hasn't shown significant changes for quite a while.
Is this plateauing behavior expected in the early or middle stages of training for this model? Any insights on the typical convergence pattern would be very helpful.
Thanks in advance for your help!"