Update README.md

README.md

@@ -56,7 +56,6 @@ Standard compression and distillation often requires massive retraining. We prov
 
 * **Training Time:** A few hours on **1x NVIDIA H100**.
 * **Data:** Only **MNLI** + **500k Wikipedia Samples**.
-* **Math over Brute Force:** By replacing all FFNs with **Monarch Matrices** $O(N \log N)$, we reduced the mathematical complexity (GFLOPs) by **66%**.
 * **Trade-off:** This extreme compression comes with a moderate accuracy drop (~5%). *Need higher accuracy? Check out our [Hybrid Version](https://huggingface.co/ykae/monarch-bert-base-mnli-hybrid) (<1% loss).*
 
 ## 🚀 Key Benchmarks