Update README.md

README.md

@@ -71,9 +71,9 @@ Retained ~106M documents / ~125B tokens ( Llama 2 tokenizer)
 
 ## 🚀 Performance Impact
 
-When used for continued pretraining of TinyLlama-1.1B (1T EN tokens), ClassiCC-PT improved Portuguese benchmark performance (Poeta v2) significantly, outperforming mC4-PT and matching ClueWeb-22-PT. The model trained with ClassiCC-PT is called Curió 1.1B and is available at huggingface.
+When used for continued pretraining of TinyLlama-1.1B (1T EN tokens), ClassiCC-PT improved Portuguese benchmark performance (Poeta v1) significantly, outperforming mC4-PT and matching ClueWeb-22-PT. The model trained with ClassiCC-PT is called Curió 1.1B and is available at huggingface.
 
-| Model | Training Regimen | Poeta v2 NPM |
+| Model | Training Regimen | Poeta v1 NPM |
 | ---------------------------- | ----------------- | ------------ |
 | TinyLlama-1T (EN) | – | 17.4 |
 | mC4-PT | cont. pretraining | \~20 |