Update README.md

README.md

@@ -210,7 +210,7 @@ Input: Please [MASK] the door before leaving.
 - **Training Data** 📚: Trained on Wikipedia, BookCorpus, MNLI, and sentence-transformers/all-nli for broad and specialized NLP strength.  
 - **Key Strength** 💪: Combines extreme efficiency with balanced performance, perfect for edge and general NLP tasks.  
 - **Use Cases** 🎯: Versatile across IoT 🌍, wearables ⌚, smart homes 🏠, and moderate hardware, supporting real-time and offline applications.  
-- **Accuracy** ✅: Competitive with larger models, achieving ~90-95% of BERT-base’s performance (task-dependent).  
+- **Accuracy** ✅: Competitive with larger models, achieving ~90-97% of BERT-base’s performance (task-dependent).  
 - **Contextual Understanding** 🔍: Strong bidirectional context, adept at disambiguating meanings in real-world scenarios.  
 - **License** 📜: MIT License (or Apache 2.0 compatible), free to use, modify, and share for all users.  
 - **Release Context** 🆙: v1.1, released April 04, 2025, reflecting cutting-edge lightweight design.
@@ -245,7 +245,7 @@ Input: Please [MASK] the door before leaving.
   - MIT License offers unrestricted freedom to use, modify, and share, slightly more permissive than `bert-mini`’s typical Apache 2.0.  
 
 - **Competitive Accuracy** ✅  
-  - Matches `bert-mini`’s ~90-95% of BERT-base performance, but with a custom design that excels in edge-specific tasks like NLI.  
+  - Matches `bert-mini`’s ~90-97% of BERT-base performance, but with a custom design that excels in edge-specific tasks like NLI.  
 
 - **Future-Ready** 🚀  
   - Built for the next wave of AI—think IoT and real-time NLP—making it more forward-looking than the general-purpose `bert-mini`.