https://github.com/tomaarsen/attention_sinks

Extend existing LLMs way beyond the original training length with constant memory usage, without retraining
https://github.com/tomaarsen/attention_sinks

llm llms nlp python transformers

Last synced: about 2 months ago
JSON representation

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

• Host: GitHub
• URL: https://github.com/tomaarsen/attention_sinks
• Owner: tomaarsen
• License: apache-2.0
• Created: 2023-10-02T14:51:53.000Z (12 months ago)
• Default Branch: main
• Last Pushed: 2024-04-10T22:54:31.000Z (5 months ago)
• Last Synced: 2024-07-05T00:42:26.989Z (2 months ago)
• Topics: llm, llms, nlp, python, transformers
• Language: Python
• Homepage: https://huggingface.co/blog/tomaarsen/attention-sinks
• Size: 8.66 MB
• Stars: 648
• Watchers: 12
• Forks: 41
• Open Issues: 19
• Metadata Files:
  • Readme: README.md
  • Changelog: CHANGELOG.md
  • License: LICENSE

Awesome Lists containing this project