Jonathan Hayase
@jon.jon.ke
5th year PhD student at UW CSE, working on Security and Privacy for ML
Tokenizers govern the allocation of computation. It's a waste to spend a whole token of compute predicting the "way" in "By the way". SuperBPE redirects that compute to predict more difficult tokens, leading to wins on downstream tasks!
We created SuperBPE🚀, a *superword* tokenizer that includes tokens spanning multiple words.
When pretraining at 8B scale, SuperBPE models consistently outperform the BPE baseline on 30 downstream tasks (+8% MMLU), while also being 27% more efficient at inference time.🧵
When pretraining at 8B scale, SuperBPE models consistently outperform the BPE baseline on 30 downstream tasks (+8% MMLU), while also being 27% more efficient at inference time.🧵
March 21, 2025 at 6:31 PM
Tokenizers govern the allocation of computation. It's a waste to spend a whole token of compute predicting the "way" in "By the way". SuperBPE redirects that compute to predict more difficult tokens, leading to wins on downstream tasks!
Reposted by Jonathan Hayase
excited to be at #NeurIPS2024! I'll be presenting our data mixture inference attack 🗓️ Thu 4:30pm w/ @jon.jon.ke — stop by to learn what trained tokenizers reveal about LLM development (‼️) and chat about all things tokenizers.
🔗 arxiv.org/abs/2407.16607
🔗 arxiv.org/abs/2407.16607
December 11, 2024 at 10:08 PM
excited to be at #NeurIPS2024! I'll be presenting our data mixture inference attack 🗓️ Thu 4:30pm w/ @jon.jon.ke — stop by to learn what trained tokenizers reveal about LLM development (‼️) and chat about all things tokenizers.
🔗 arxiv.org/abs/2407.16607
🔗 arxiv.org/abs/2407.16607