controlling-llm-memorization
controlling-llm-memorization copied to clipboard
request about training log
Hello, I have a LLM which continuing pretrained on a domain-specific dataset. My goal is to use this LLM as an index, where given a prefix or a query similar to the prefix, the model can generate text similar to the original suffix.
However, I've noticed an issue during training: the aligned CLM loss is increasing, which confuse me.
I would greatly appreciate if you could provide me with the training logs. I believe having access to the logs will help me better understand and address the increasing loss problem.