Update app.py

app.py

@@ -5,13 +5,24 @@ from huggingface_hub import InferenceClient
 MODELS = {
     "Phase 2: Stable (Formal)": {
         "id": "st192011/Maltese-EuroLLM-1.7B-Phase2-Stable",
-        "description": "The 'Bureaucrat Bot'. Trained on 200k rows of EU/Government data (TildeMODEL). High fidelity for legal and official documents.",
+        "description": (
+            "The 'Bureaucrat Bot'. Built upon a foundational adaptation phase that mixed "
+            "monolingual Maltese and Italian to bridge morphological roots. This version "
+            "was fine-tuned on high-fidelity EU and governmental parallel corpora, "
+            "optimizing it for extreme formal precision and administrative accuracy."
+        ),
         "chrf": "60.18",
         "comet": "0.6431"
     },
     "Phase 4: Anchored (Native)": {
         "id": "st192011/Maltese-EuroLLM-1.7B-Phase4-Anchored",
-        "description": "The 'Native Speaker'. Uses Anchored Reasoning (CoT) distilled from Llama-70B. Designed for natural phrasing and cultural awareness.",
+        "description": (
+            "The 'Native Speaker'. An evolution of Phase 2 utilizing a curriculum-based "
+            "'Full Circle' approach. It integrates synthesized reasoning chains (CoT) "
+            "that allow the model to process linguistic logic before translating. By mixing "
+            "all previous data types, it anchors factual accuracy to native-level phrasing "
+            "and cultural awareness."
+        ),
         "chrf": "52.68",
         "comet": "0.6567"
     }
@@ -50,7 +61,7 @@ def translate_logic(text, selected_models, temp):
 
 # --- GRADIO UI ---
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
-    gr.Markdown("# 🇲🇹 Maltese-MT Arena")
+    gr.Markdown("# 🇲🇹 Maltese-MT Lab")
     gr.Markdown("Compare different generations of fine-tuned EuroLLM models for English-to-Maltese translation.")
     
     with gr.Row():