Update README.md

README.md

@@ -53,8 +53,8 @@ It achieves the following results on the evaluation set:
 ## Model description
 
 This model uses a shared `XLM-RoBERTa` base to encode input text. The resulting text representation is then fed into two separate, independent classification layers (heads):
-*   A **Sentiment Head** with 3 outputs for `positive`, `neutral`, and `negative` classes.
-*   A **Multi-Label Head** with 41 outputs, which are decoded to predict the presence or absence of 37 different disaster-related categories.
+*   A **Sentiment Head (Frozen from pre-trained model)** with 3 outputs for `positive`, `neutral`, and `negative` classes.
+*   A **Multi-Label Head (Newly created and fine-tuned)** with 41 outputs, which are decoded to predict the presence or absence of 37 different disaster-related categories.
 
 This dual-head architecture allows for a nuanced understanding of a message, capturing both its emotional content and its specific, actionable information.