attention_sinks
Extend existing LLMs way beyond the original training length with constant memory usage, without retraining
tomaarsen/attention_sinks
View on GitHub ↗Extend existing LLMs way beyond the original training length with constant memory usage, without retraining
attention_sinks grew +0.0% in the last 30 days, ranks 22nd of 101 in Model Optimization, last commit 927 days ago.
Star trajectory
79 days tracked2026-08-07734 stars today2026-10-25
Momentum
+0.0% in 30d, below the Model Optimization median of +0.7%.
Maintenance
Last commit 927 days ago, staler than the median of 53d.
How it compares in Model Optimization
attention_sinks
734
stars
+0.0% 30d
commit 927 days ago
Model Optimization median
266
stars
+0.7% 30d
commit ~53d ago
Leader · kimi-k3-in-c
9.2k
stars
+6.9% 30d
commit 23 days ago
Ranked 22nd of 101 by stars.
ModelTC/LightCompress751-17warpfront/hipfire747-13
tomaarsen/attention_sinks734
openinfer-project/openinfer725+9SqueezeAILab/SqueezeLLM725+9- Language
- Python
- License
- Apache License 2.0
- Contributors
- 4
- Forks
- 43
- Watchers
- 9
- Open issues
- 19
- Releases
- 0
- Created
- 2023-10-02
- Stars this week
- +0
You've seen where attention_sinks stands today.
Get the moves before everyone else: the 5 Model Optimization repos climbing fastest, every Tuesday.
Free weekly digest. Double opt-in, unsubscribe anytime.