Update README.md

README.md

@@ -257,7 +257,7 @@ Sample sizes: Carbon & GENERator n=500. Evo2-7B n=150 at 16k, n=100 at 32k, n=20
 - **Architecture:** decoder-only Transformer (Llama-style), 30 layers, hidden 3072, FFN 8448, 32 attention heads with GQA (4 KV groups), SwiGLU, RMSNorm.
 - **Tokenizer:** Carbon 6-mer hybrid (vocab ≈ 156 k including DNA tags and metadata tokens and BPE tokens for future English & DNA continual pretraining).
 - **Precision:** bfloat16
-- **Positional embedding:** RoPE, base $\theta = 5 \times 10^{6}$, max position 32 768.
+- **Positional embedding:** RoPE, base θ = 5 × 10^6, max position 32,768.
 
 ### Pre-training