Memory Leakage Problem?

When I tried to fine-tune the model with a large annotation set (about 2M), memory usage gradually increased during the training process. Ultimately, the process was exploded.

The report names no file, test, or entry point. Start by reproducing fine-tuning with an annotation set of about 2M and monitor memory over the training process. Done means identifying why memory grows and confirming that the process completes without being terminated.

Written by the indexing model from the issue text.