Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothing — mentioned by Imnimo

Mon, 24 Jul 2023 19:48:15 GMT

Right, I get the goal of removing the outlier activations, but I just don't understand why outlier activations are a consequence of the model trying to "pass". The story from the linked paper earlier in the post (https://arxiv.org/pdf/2306.12929.pdf) is that the model is doing the following: -Learn a near-zero representation for some otherwise low-importance token, like delimiters or whitespace. -When a head wants to "pass", emit an outlier activation to attend to that token nearly-exclusively. But I'm surprised the model can't just use its existing tools (the post-concat projection layer and…

Paper and discussion history. A mention is not an endorsement.

Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothing · HackerPapers discussions

Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothing — mentioned by Imnimo