Update src/streamlit_app.py

src/streamlit_app.py

@@ -72,6 +72,49 @@ def extract_numeric_ram(ram) -> Optional[int]:
     return None
 
 # Streamlined LLM database with popular models and download sizes
+QUANTIZATION_INFO = {
+    "fp16": {"multiplier": 1.0, "label": "FP16 (Full)", "description": "Best quality, largest size"},
+    "8bit": {"multiplier": 0.5, "label": "8-bit", "description": "Good quality, 50% smaller"},
+    "4bit": {"multiplier": 0.25, "label": "4-bit", "description": "Decent quality, 75% smaller"}
+}
+
+def calculate_quantized_size(base_size_str: str, quantization: str) -> str:
+    """Calculate quantized model size based on base size and quantization type"""
+    try:
+        import re
+        match = re.match(r'(\d+(?:\.\d+)?)\s*(GB|MB)', base_size_str.upper())
+        if not match:
+            return base_size_str
+        
+        value, unit = float(match.group(1)), match.group(2)
+        multiplier = QUANTIZATION_INFO[quantization]["multiplier"]
+        
+        new_value = value * multiplier
+        
+        if unit == "MB" and new_value >= 1024:
+            new_value = new_value / 1024
+            unit = "GB"
+        elif unit == "GB" and new_value < 1:
+            new_value = new_value * 1024
+            unit = "MB"
+        
+        return f"{new_value:.0f}{unit}" if new_value >= 10 else f"{new_value:.1f}{unit}"
+    except:
+        return base_size_str
+
+def get_quantization_recommendations(ram_gb: int) -> List[str]:
+    """Recommend best quantization options based on available RAM"""
+    if ram_gb <= 2:
+        return ["4bit"]
+    elif ram_gb <= 4:
+        return ["4bit", "8bit"]
+    elif ram_gb <= 8:
+        return ["4bit", "8bit"]
+    elif ram_gb <= 16:
+        return ["8bit", "fp16"]
+    else:
+        return ["fp16", "8bit", "4bit"]
+
 LLM_DATABASE = {
     "ultra_low": {  # ≤2GB
         "general": [