Update README.md

README.md

@@ -41,7 +41,7 @@ tokenizer = AutoTokenizer.from_pretrained("gpt2")
 print(generate(model, tokenizer, "Once upon a time"))
 
 Performance
-FineWeb, 187M params: 3.73 val loss / 41.6 PPL (75k steps)
+FineWeb, 187M params: 3.73 val loss / 41.6 PPL (75k steps•32 batch•1024 seq)
 Architecture: 21 layers, 768d, 12 heads, 16 slots
 Links
 Code: https://github.com/DigitalDaimyo/AddressedStateAttention