@gu_xiangming@OpenAI It seems that the initial <start> token still exhibit large activation outliers, even with learnable key bias. Do you observe the same?
ref: https://t.co/VB3YZ6qGPs
@Guangxuan_Xiao Do you know why the very first <start> token still has massive activation norms?
I'd always assumed this was related to the first token being an attention sink, but maybe there's something else going on
In this report we describe the 2025 Apple Foundation Models ("AFM"). We also introduce the new Foundation Models framework, which gives app developers direct access to the on-device AFM model.
https://t.co/nEbtxuGrjD
At WWDC we introduce a new generation of LLMs developed to enhance the Apple Intelligence features. We also introduce the new Foundation Models framework, which gives app developers direct access to the on-device foundation language model.
https://t.co/SnjCXrIyYj
@mathena To some extent I think there was still some truth in your statement.
What people usually think of Robinhood:
Product: Commision-free trading
Customer: Retail traders
What Robinhood really is:
Product: Retail traders
Customer: HFTs