Model Architecture & Training MethodsMIT (Xiao et al.)MIT

StreamingLLM

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

StreamingLLM — discovery of the attention sink

Why it matters

The finding that initial tokens act as attention sinks; 2 years later gpt-oss adopted learnable sinks

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2309.17453

Added in Week 35, 2026 · See that week

On the map — ⑪ Model Architecture line

  1. Previous stopSolidGoldMagikarp
  2. This record · stopStreamingLLM
  3. Next stopDeepSeek-V2
See it on the map

Same week

Promptbreeder publishedPaperMistral 7BOfficialOpenAI publishes GPT-4V(ision) System CardOfficial

Nearby on the same axis

YaRNPaperMambaPaperBeyond Chinchilla-OptimalPaperGQAPaper