Update README.md

README.md

@@ -21,7 +21,7 @@ MixtureofRecursionwithRouter is tailored for technical domains, combining:
 ->Custom Tokenizer: Byte-pair encoding (BPE) with special tokens for code, math, and conversation roles (e.g., <user>, <assistant>).
 ->Adaptive Embeddings: Token embeddings with configurable positional encodings (learned, sinusoidal, or RoPE).
 ->Recursive Transformer: Multi-layered architecture with a RecursionRouter to dynamically adjust computation steps based on input complexity.
-->Ultra-Fast Training: Optimized for low loss (<2.0) and perplexity (<12) in 4-5 hours using mixed precision and cosine scheduling.
+->Ultra-Fast Training: Optimized for low loss (<2.0) and perplexity (<12) using mixed precision and cosine scheduling.
 
 ## Model Details