Comment on Grokfast: Accelerated Grokking by Amplifying Slow GradientsparentComments−barfbagginus2yLooks like grokking could give better reasoning and generalization to LLMs, but I'm not sure how practical it would be to overfit a larger LLMSee: https://arxiv.org/abs/2405.15071
Comments
Looks like grokking could give better reasoning and generalization to LLMs, but I'm not sure how practical it would be to overfit a larger LLM
See: https://arxiv.org/abs/2405.15071