attention_sinks

tomaarsen

Extend existing LLMs way beyond the original training length with constant memory usage, without retraining

nlp

View on GitHub Website

736 Stars

44 Forks

736 Watchers

Python Language

apache-2.0 License

100 SrcLog Score

Cost to Build

$393.9K

Market Value

$911.2K

How is this calculated?

Growth over time

2 data points · 2026-04-09 → 2026-04-17

Stars Forks Watchers

💬

How do you feel about this project?

Ask AI about attention_sinks

Question copied to clipboard

What is the tomaarsen/attention_sinks GitHub project? Description: "Extend existing LLMs way beyond the original training length with constant memory usage, without retraining". Written in Python. Explain what it does, its main use cases, key features, and who would benefit from using it.

Question is copied to clipboard — paste it after the AI opens.

How to clone attention_sinks

Clone via HTTPS

git clone https://github.com/tomaarsen/attention_sinks.git

Clone via SSH

[email protected]:tomaarsen/attention_sinks.git

Download ZIP

Download master.zip

Found an issue?

Report bugs or request features on the attention_sinks issue tracker:

Open GitHub Issues

Similar to attention_sinks

lectures spaCy HanLP compromise gensim stanford-tensorflow-tutorials nltk awesome-nlp TextBlob ailearning CoreNLP ansj_seg rasa tensorflow_cookbook allennlp flashtext TagUI franc mycroft-core practical-pytorch text_classification nlp_tasks DeepPavlov snips-nlu Awesome-pytorch-list kcws Awesome-Chinese-NLP sentiment prose DeepLearn