The perplexity of the target string had a large impact on if the model could "memorize" it, and "memorizing" a high-PPL string broke the model.
short.sectorr.dev/llm-memoriza...
The perplexity of the target string had a large impact on if the model could "memorize" it, and "memorizing" a high-PPL string broke the model.
short.sectorr.dev/llm-memoriza...
short.sectorr.dev/wsc
short.sectorr.dev/wsc
short.sectorr.dev/sudo
short.sectorr.dev/sudo