Update src/streamlit_app.py

src/streamlit_app.py

@@ -1,8 +1,8 @@
 #!/usr/bin/env python3
 """
-LLM Compatibility Advisor - Streamlined with Download Sizes
+Enhanced LLM Compatibility Advisor - Complete with Quantization & Advanced Features
 Author: Assistant
-Description: Provides device-based LLM recommendations with popular models and download sizes
+Description: Comprehensive device-based LLM recommendations with quantization, comparison, and download assistance
 Requirements: streamlit, pandas, plotly, openpyxl
 """
 
@@ -13,10 +13,11 @@ import re
 import plotly.express as px
 import plotly.graph_objects as go
 from typing import Optional, Tuple, List, Dict
+import json
 
 # ✅ MUST be the first Streamlit command
 st.set_page_config(
-    page_title="LLM Compatibility Advisor", 
+    page_title="Enhanced LLM Compatibility Advisor", 
     layout="wide",
     page_icon="🧠",
     initial_sidebar_state="expanded"
@@ -41,7 +42,6 @@ def load_data():
         except Exception as e:
             return None, f"Error loading '{path}': {str(e)}"
     
-    # Return success case - this was missing!
     if combined_df.empty:
         return None, "No data found in Excel files."
     else:
@@ -71,161 +71,221 @@ 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"}
+# Quantization options and size calculations
+QUANTIZATION_FORMATS = {
+    "FP16": {"multiplier": 1.0, "description": "Full precision, best quality", "icon": "🔥"},
+    "8-bit": {"multiplier": 0.5, "description": "50% smaller, good quality", "icon": "⚡"},
+    "4-bit": {"multiplier": 0.25, "description": "75% smaller, acceptable quality", "icon": "💎"},
+    "2-bit": {"multiplier": 0.125, "description": "87.5% smaller, experimental", "icon": "🧪"}
 }
 
-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:
+def calculate_quantized_size(base_size_str, quant_format):
+    """Calculate quantized model size"""
+    size_match = re.search(r'(\d+\.?\d*)', base_size_str)
+    if not size_match:
         return base_size_str
+    
+    base_size = float(size_match.group(1))
+    unit = base_size_str.replace(size_match.group(1), "").strip()
+    
+    multiplier = QUANTIZATION_FORMATS[quant_format]["multiplier"]
+    new_size = base_size * multiplier
+    
+    return f"{new_size:.1f}{unit}"
 
-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"]
-
+# Enhanced LLM database with more models and metadata
 LLM_DATABASE = {
     "ultra_low": {  # ≤2GB
         "general": [
-            {"name": "TinyLlama-1.1B-Chat", "size": "637MB", "description": "Compact chat model"},
-            {"name": "DistilBERT-base", "size": "268MB", "description": "Efficient BERT variant"},
-            {"name": "all-MiniLM-L6-v2", "size": "91MB", "description": "Sentence embeddings"}
+            {"name": "TinyLlama-1.1B-Chat", "size": "637MB", "description": "Compact chat model", "parameters": "1.1B", "context": "2K"},
+            {"name": "DistilBERT-base", "size": "268MB", "description": "Efficient BERT variant", "parameters": "66M", "context": "512"},
+            {"name": "all-MiniLM-L6-v2", "size": "91MB", "description": "Sentence embeddings", "parameters": "22M", "context": "256"},
+            {"name": "OpenELM-270M", "size": "540MB", "description": "Apple's efficient model", "parameters": "270M", "context": "2K"}
         ],
         "code": [
-            {"name": "CodeT5-small", "size": "242MB", "description": "Code generation"},
-            {"name": "Replit-code-v1-3B", "size": "1.2GB", "description": "Code completion"}
+            {"name": "CodeT5-small", "size": "242MB", "description": "Code generation", "parameters": "60M", "context": "512"},
+            {"name": "Replit-code-v1-3B", "size": "1.2GB", "description": "Code completion", "parameters": "3B", "context": "4K"}
         ]
     },
     "low": {  # 3-4GB
         "general": [
-            {"name": "Phi-1.5", "size": "2.8GB", "description": "Microsoft's efficient model"},
-            {"name": "Gemma-2B", "size": "1.4GB", "description": "Google's compact model"},
-            {"name": "OpenLLaMA-3B", "size": "2.1GB", "description": "Open source LLaMA"}
+            {"name": "Phi-1.5", "size": "2.8GB", "description": "Microsoft's efficient model", "parameters": "1.3B", "context": "2K"},
+            {"name": "Gemma-2B", "size": "1.4GB", "description": "Google's compact model", "parameters": "2B", "context": "8K"},
+            {"name": "OpenLLaMA-3B", "size": "2.1GB", "description": "Open source LLaMA", "parameters": "3B", "context": "2K"},
+            {"name": "StableLM-3B", "size": "2.2GB", "description": "Stability AI model", "parameters": "3B", "context": "4K"}
         ],
         "code": [
-            {"name": "CodeGen-2B", "size": "1.8GB", "description": "Salesforce code model"},
-            {"name": "StarCoder-1B", "size": "1.1GB", "description": "BigCode project"}
+            {"name": "CodeGen-2B", "size": "1.8GB", "description": "Salesforce code model", "parameters": "2B", "context": "2K"},
+            {"name": "StarCoder-1B", "size": "1.1GB", "description": "BigCode project", "parameters": "1B", "context": "8K"}
         ],
         "chat": [
-            {"name": "Alpaca-3B", "size": "2.0GB", "description": "Stanford's instruction model"},
-            {"name": "Vicuna-3B", "size": "2.1GB", "description": "ChatGPT-style training"}
+            {"name": "Alpaca-3B", "size": "2.0GB", "description": "Stanford's instruction model", "parameters": "3B", "context": "2K"},
+            {"name": "Vicuna-3B", "size": "2.1GB", "description": "ChatGPT-style training", "parameters": "3B", "context": "2K"}
         ]
     },
     "moderate_low": {  # 5-6GB
         "general": [
-            {"name": "Phi-2", "size": "5.2GB", "description": "Microsoft's 2.7B model"},
-            {"name": "Gemma-7B-it", "size": "4.2GB", "description": "Google instruction tuned"},
-            {"name": "Mistral-7B-v0.1", "size": "4.1GB", "description": "Mistral AI base model"}
+            {"name": "Phi-2", "size": "5.2GB", "description": "Microsoft's 2.7B model", "parameters": "2.7B", "context": "2K"},
+            {"name": "Gemma-7B-it", "size": "4.2GB", "description": "Google instruction tuned", "parameters": "7B", "context": "8K"},
+            {"name": "Mistral-7B-v0.1", "size": "4.1GB", "description": "Mistral AI base model", "parameters": "7B", "context": "8K"},
+            {"name": "Llama-2-7B", "size": "4.0GB", "description": "Meta's foundation model", "parameters": "7B", "context": "4K"}
         ],
         "code": [
-            {"name": "CodeLlama-7B", "size": "3.8GB", "description": "Meta's code specialist"},
-            {"name": "StarCoder-7B", "size": "4.0GB", "description": "Code generation expert"}
+            {"name": "CodeLlama-7B", "size": "3.8GB", "description": "Meta's code specialist", "parameters": "7B", "context": "16K"},
+            {"name": "StarCoder-7B", "size": "4.0GB", "description": "Code generation expert", "parameters": "7B", "context": "8K"}
         ],
         "chat": [
-            {"name": "Zephyr-7B-beta", "size": "4.2GB", "description": "HuggingFace chat model"},
-            {"name": "Neural-Chat-7B", "size": "4.1GB", "description": "Intel optimized"}
+            {"name": "Zephyr-7B-beta", "size": "4.2GB", "description": "HuggingFace chat model", "parameters": "7B", "context": "32K"},
+            {"name": "Neural-Chat-7B", "size": "4.1GB", "description": "Intel optimized", "parameters": "7B", "context": "32K"}
         ]
     },
     "moderate": {  # 7-8GB
         "general": [
-            {"name": "Llama-2-7B-Chat", "size": "3.5GB", "description": "Meta's popular chat model"},
-            {"name": "Mistral-7B-Instruct-v0.2", "size": "4.1GB", "description": "Latest Mistral instruct"},
-            {"name": "Qwen-7B-Chat", "size": "4.0GB", "description": "Alibaba's multilingual"}
+            {"name": "Llama-2-7B-Chat", "size": "3.5GB", "description": "Meta's popular chat model", "parameters": "7B", "context": "4K"},
+            {"name": "Mistral-7B-Instruct-v0.2", "size": "4.1GB", "description": "Latest Mistral instruct", "parameters": "7B", "context": "32K"},
+            {"name": "Qwen-7B-Chat", "size": "4.0GB", "description": "Alibaba's multilingual", "parameters": "7B", "context": "32K"},
+            {"name": "Solar-10.7B-Instruct", "size": "5.8GB", "description": "Upstage's efficient model", "parameters": "10.7B", "context": "4K"}
         ],
         "code": [
-            {"name": "CodeLlama-7B-Instruct", "size": "3.8GB", "description": "Instruction-tuned CodeLlama"},
-            {"name": "WizardCoder-7B", "size": "4.0GB", "description": "Enhanced coding abilities"},
-            {"name": "Phind-CodeLlama-34B-v2", "size": "4.2GB", "description": "4-bit quantized version"}
+            {"name": "CodeLlama-7B-Instruct", "size": "3.8GB", "description": "Instruction-tuned CodeLlama", "parameters": "7B", "context": "16K"},
+            {"name": "WizardCoder-7B", "size": "4.0GB", "description": "Enhanced coding abilities", "parameters": "7B", "context": "16K"},
+            {"name": "Phind-CodeLlama-34B-v2", "size": "4.2GB", "description": "4-bit quantized version", "parameters": "34B", "context": "16K"}
         ],
         "reasoning": [
-            {"name": "WizardMath-7B", "size": "4.0GB", "description": "Mathematical reasoning"},
-            {"name": "MetaMath-7B", "size": "3.9GB", "description": "Math problem solving"}
+            {"name": "WizardMath-7B", "size": "4.0GB", "description": "Mathematical reasoning", "parameters": "7B", "context": "2K"},
+            {"name": "MetaMath-7B", "size": "3.9GB", "description": "Math problem solving", "parameters": "7B", "context": "2K"}
         ]
     },
     "good": {  # 9-16GB
         "general": [
-            {"name": "Llama-2-13B-Chat", "size": "7.3GB", "description": "Larger Llama variant"},
-            {"name": "Vicuna-13B-v1.5", "size": "7.2GB", "description": "Enhanced Vicuna"},
-            {"name": "OpenChat-3.5", "size": "7.1GB", "description": "High-quality chat model"}
+            {"name": "Llama-2-13B-Chat", "size": "7.3GB", "description": "Larger Llama variant", "parameters": "13B", "context": "4K"},
+            {"name": "Vicuna-13B-v1.5", "size": "7.2GB", "description": "Enhanced Vicuna", "parameters": "13B", "context": "16K"},
+            {"name": "OpenChat-3.5", "size": "7.1GB", "description": "High-quality chat model", "parameters": "7B", "context": "8K"},
+            {"name": "Nous-Hermes-2-Mixtral-8x7B-DPO", "size": "12.9GB", "description": "4-bit quantized MoE", "parameters": "47B", "context": "32K"}
         ],
         "code": [
-            {"name": "CodeLlama-13B-Instruct", "size": "7.3GB", "description": "Larger code model"},
-            {"name": "WizardCoder-15B", "size": "8.2GB", "description": "Advanced coding"},
-            {"name": "StarCoder-15B", "size": "8.5GB", "description": "Large code model"}
+            {"name": "CodeLlama-13B-Instruct", "size": "7.3GB", "description": "Larger code model", "parameters": "13B", "context": "16K"},
+            {"name": "WizardCoder-15B", "size": "8.2GB", "description": "Advanced coding", "parameters": "15B", "context": "16K"},
+            {"name": "StarCoder-15B", "size": "8.5GB", "description": "Large code model", "parameters": "15B", "context": "8K"}
         ],
         "multimodal": [
-            {"name": "LLaVA-7B", "size": "7.0GB", "description": "Vision + language"},
-            {"name": "MiniGPT-4-7B", "size": "6.8GB", "description": "Multimodal chat"}
+            {"name": "LLaVA-7B", "size": "7.0GB", "description": "Vision + language", "parameters": "7B", "context": "2K"},
+            {"name": "MiniGPT-4-7B", "size": "6.8GB", "description": "Multimodal chat", "parameters": "7B", "context": "2K"},
+            {"name": "Instructblip-7B", "size": "7.2GB", "description": "Instruction-tuned VLM", "parameters": "7B", "context": "2K"}
         ],
         "reasoning": [
-            {"name": "WizardMath-13B", "size": "7.3GB", "description": "Advanced math"},
-            {"name": "Orca-2-13B", "size": "7.4GB", "description": "Microsoft reasoning"}
+            {"name": "WizardMath-13B", "size": "7.3GB", "description": "Advanced math", "parameters": "13B", "context": "2K"},
+            {"name": "Orca-2-13B", "size": "7.4GB", "description": "Microsoft reasoning", "parameters": "13B", "context": "4K"}
         ]
     },
     "high": {  # 17-32GB
         "general": [
-            {"name": "Mixtral-8x7B-Instruct-v0.1", "size": "26.9GB", "description": "Mixture of experts"},
-            {"name": "Llama-2-70B-Chat", "size": "38.0GB", "description": "8-bit quantized"},
-            {"name": "Yi-34B-Chat", "size": "19.5GB", "description": "01.AI's large model"}
+            {"name": "Mixtral-8x7B-Instruct-v0.1", "size": "26.9GB", "description": "Mixture of experts", "parameters": "47B", "context": "32K"},
+            {"name": "Llama-2-70B-Chat", "size": "38.0GB", "description": "8-bit quantized", "parameters": "70B", "context": "4K"},
+            {"name": "Yi-34B-Chat", "size": "19.5GB", "description": "01.AI's large model", "parameters": "34B", "context": "200K"},
+            {"name": "Nous-Hermes-2-Yi-34B", "size": "19.2GB", "description": "Enhanced Yi variant", "parameters": "34B", "context": "200K"}
         ],
         "code": [
-            {"name": "CodeLlama-34B-Instruct", "size": "19.0GB", "description": "Large code specialist"},
-            {"name": "DeepSeek-Coder-33B", "size": "18.5GB", "description": "DeepSeek's coder"},
-            {"name": "WizardCoder-34B", "size": "19.2GB", "description": "Enterprise coding"}
+            {"name": "CodeLlama-34B-Instruct", "size": "19.0GB", "description": "Large code specialist", "parameters": "34B", "context": "16K"},
+            {"name": "DeepSeek-Coder-33B", "size": "18.5GB", "description": "DeepSeek's coder", "parameters": "33B", "context": "16K"},
+            {"name": "WizardCoder-34B", "size": "19.2GB", "description": "Enterprise coding", "parameters": "34B", "context": "16K"}
         ],
         "reasoning": [
-            {"name": "WizardMath-70B", "size": "38.5GB", "description": "8-bit quantized math"},
-            {"name": "MetaMath-70B", "size": "38.0GB", "description": "8-bit math reasoning"}
+            {"name": "WizardMath-70B", "size": "38.5GB", "description": "8-bit quantized math", "parameters": "70B", "context": "2K"},
+            {"name": "MetaMath-70B", "size": "38.0GB", "description": "8-bit math reasoning", "parameters": "70B", "context": "2K"}
         ]
     },
     "ultra_high": {  # >32GB
         "general": [
-            {"name": "Llama-2-70B", "size": "130GB", "description": "Full precision"},
-            {"name": "Mixtral-8x22B", "size": "176GB", "description": "Latest mixture model"},
-            {"name": "Qwen-72B", "size": "145GB", "description": "Alibaba's flagship"}
+            {"name": "Llama-2-70B", "size": "130GB", "description": "Full precision", "parameters": "70B", "context": "4K"},
+            {"name": "Mixtral-8x22B", "size": "176GB", "description": "Latest mixture model", "parameters": "141B", "context": "64K"},
+            {"name": "Qwen-72B", "size": "145GB", "description": "Alibaba's flagship", "parameters": "72B", "context": "32K"},
+            {"name": "Llama-3-70B", "size": "140GB", "description": "Meta's latest", "parameters": "70B", "context": "8K"}
         ],
         "code": [
-            {"name": "CodeLlama-34B", "size": "68GB", "description": "Full precision code"},
-            {"name": "DeepSeek-Coder-33B", "size": "66GB", "description": "Full precision coding"}
+            {"name": "CodeLlama-34B", "size": "68GB", "description": "Full precision code", "parameters": "34B", "context": "16K"},
+            {"name": "DeepSeek-Coder-33B", "size": "66GB", "description": "Full precision coding", "parameters": "33B", "context": "16K"}
         ],
         "reasoning": [
-            {"name": "WizardMath-70B", "size": "130GB", "description": "Full precision math"},
-            {"name": "Goat-70B", "size": "132GB", "description": "Arithmetic reasoning"}
+            {"name": "WizardMath-70B", "size": "130GB", "description": "Full precision math", "parameters": "70B", "context": "2K"},
+            {"name": "Goat-70B", "size": "132GB", "description": "Arithmetic reasoning", "parameters": "70B", "context": "2K"}
         ]
     }
 }
 
+# GPU compatibility database
+GPU_DATABASE = {
+    "RTX 3060": {"vram": 8, "performance": "mid", "architecture": "Ampere"},
+    "RTX 3070": {"vram": 8, "performance": "high", "architecture": "Ampere"},
+    "RTX 3080": {"vram": 10, "performance": "high", "architecture": "Ampere"},
+    "RTX 3090": {"vram": 24, "performance": "ultra", "architecture": "Ampere"},
+    "RTX 4060": {"vram": 8, "performance": "mid", "architecture": "Ada Lovelace"},
+    "RTX 4070": {"vram": 12, "performance": "high", "architecture": "Ada Lovelace"},
+    "RTX 4080": {"vram": 16, "performance": "ultra", "architecture": "Ada Lovelace"},
+    "RTX 4090": {"vram": 24, "performance": "ultra", "architecture": "Ada Lovelace"},
+    "Apple M1": {"vram": 8, "performance": "mid", "architecture": "Apple Silicon"},
+    "Apple M2": {"vram": 16, "performance": "high", "architecture": "Apple Silicon"},
+    "Apple M3": {"vram": 24, "performance": "ultra", "architecture": "Apple Silicon"},
+    "RX 6700 XT": {"vram": 12, "performance": "mid", "architecture": "RDNA 2"},
+    "RX 7900 XTX": {"vram": 24, "performance": "ultra", "architecture": "RDNA 3"},
+}
+
+def get_gpu_recommendations(gpu_name, ram_gb):
+    """Get GPU-specific model recommendations"""
+    if gpu_name == "No GPU":
+        return "CPU-only models recommended", "Use 4-bit quantization for better performance"
+    
+    gpu_info = GPU_DATABASE.get(gpu_name.split(" (")[0], {"vram": 0, "performance": "low"})
+    vram = gpu_info["vram"]
+    
+    if vram <= 8:
+        return f"7B models with 4-bit quantization", f"Estimated VRAM usage: ~{vram-1}GB"
+    elif vram <= 12:
+        return f"13B models with 8-bit quantization", f"Estimated VRAM usage: ~{vram-1}GB"
+    elif vram <= 16:
+        return f"13B models at FP16 or 30B with 4-bit", f"Estimated VRAM usage: ~{vram-1}GB"
+    else:
+        return f"70B models with 4-bit quantization", f"Estimated VRAM usage: ~{vram-2}GB"
+
+def predict_inference_speed(model_size_gb, ram_gb, has_gpu=False, gpu_name=""):
+    """Predict approximate inference speed"""
+    if model_size_gb > ram_gb:
+        return "❌ Insufficient RAM", "Consider smaller model or quantization"
+    
+    if has_gpu and gpu_name != "No GPU":
+        gpu_info = GPU_DATABASE.get(gpu_name.split(" (")[0], {"performance": "low"})
+        perf = gpu_info["performance"]
+        
+        if perf == "ultra":
+            if model_size_gb <= 4:
+                return "⚡ Blazing Fast", "~50-100 tokens/sec"
+            elif model_size_gb <= 8:
+                return "🚀 Very Fast", "~30-60 tokens/sec"
+            elif model_size_gb <= 16:
+                return "🏃 Fast", "~15-30 tokens/sec"
+            else:
+                return "🐌 Moderate", "~5-15 tokens/sec"
+        elif perf == "high":
+            if model_size_gb <= 4:
+                return "⚡ Very Fast", "~30-50 tokens/sec"
+            elif model_size_gb <= 8:
+                return "🚀 Fast", "~15-30 tokens/sec"
+            else:
+                return "🐌 Moderate", "~5-15 tokens/sec"
+        else:  # mid performance
+            if model_size_gb <= 4:
+                return "⚡ Fast", "~15-30 tokens/sec"
+            else:
+                return "🐌 Slow", "~3-10 tokens/sec"
+    else:
+        # CPU inference
+        if model_size_gb <= 2:
+            return "⚡ Acceptable", "~5-15 tokens/sec"
+        elif model_size_gb <= 4:
+            return "🐌 Slow", "~1-5 tokens/sec"
+        else:
+            return "🐌 Very Slow", "~0.5-2 tokens/sec"
+
 # Enhanced LLM recommendation with performance tiers
 def recommend_llm(ram_str) -> Tuple[str, str, str, Dict[str, List[Dict]]]:
     """Returns (recommendation, performance_tier, additional_info, detailed_models)"""
@@ -300,37 +360,33 @@ def get_os_info(os_name) -> Tuple[str, str]:
     else:
         return "💻", os_name
 
-# Performance visualization
-def create_performance_chart(df):
-    """Create a performance distribution chart"""
-    laptop_rams = df["Laptop RAM"].apply(extract_numeric_ram).dropna()
-    mobile_rams = df["Mobile RAM"].apply(extract_numeric_ram).dropna()
-    
-    fig = go.Figure()
-    
-    fig.add_trace(go.Histogram(
-        x=laptop_rams,
-        name="Laptop RAM",
-        opacity=0.7,
-        nbinsx=10
-    ))
-    
-    fig.add_trace(go.Histogram(
-        x=mobile_rams,
-        name="Mobile RAM",
-        opacity=0.7,
-        nbinsx=10
-    ))
-    
-    fig.update_layout(
-        title="RAM Distribution Across Devices",
-        xaxis_title="RAM (GB)",
-        yaxis_title="Number of Students",
-        barmode='overlay',
-        height=400
-    )
+# Model comparison function
+def create_model_comparison_table(selected_models, quantization_type="FP16"):
+    """Create a comparison table for selected models"""
+    comparison_data = []
     
-    return fig
+    for model_info in selected_models:
+        quant_size = calculate_quantized_size(model_info['size'], quantization_type)
+        
+        # Extract numeric size for VRAM calculation
+        size_match = re.search(r'(\d+\.?\d*)', quant_size)
+        if size_match:
+            size_num = float(size_match.group(1))
+            estimated_vram = f"{size_num * 1.2:.1f}GB"
+        else:
+            estimated_vram = "Unknown"
+        
+        comparison_data.append({
+            'Model': model_info['name'],
+            'Parameters': model_info.get('parameters', 'Unknown'),
+            'Context': model_info.get('context', 'Unknown'),
+            'Original Size': model_info['size'],
+            f'{quantization_type} Size': quant_size,
+            'Est. VRAM': estimated_vram,
+            'Description': model_info['description']
+        })
+    
+    return pd.DataFrame(comparison_data)
 
 # Enhanced model details display function
 def display_model_categories(models_dict: Dict[str, List[Dict]], ram_gb: int, show_quantization=True):
@@ -343,8 +399,18 @@ def display_model_categories(models_dict: Dict[str, List[Dict]], ram_gb: int, sh
     for category, model_list in models_dict.items():
         if model_list:
             with st.expander(f"📂 {category.replace('_', ' ').title()} Models"):
-                for model in model_list[:6]:  # Reduced to show quantization options
+                for model in model_list[:6]:  # Show top 6 models per category
                     st.markdown(f"**{model['name']}**")
+                    
+                    # Model details
+                    detail_col1, detail_col2, detail_col3 = st.columns(3)
+                    with detail_col1:
+                        st.caption(f"📊 {model.get('parameters', 'Unknown')} params")
+                    with detail_col2:
+                        st.caption(f"🔍 {model.get('context', 'Unknown')} context")
+                    with detail_col3:
+                        st.caption(f"💾 {model['size']} original")
+                    
                     st.markdown(f"*{model['description']}*")
                     
                     if show_quantization:
@@ -354,32 +420,68 @@ def display_model_categories(models_dict: Dict[str, List[Dict]], ram_gb: int, sh
                             with quant_cols[i]:
                                 quant_size = calculate_quantized_size(model['size'], quant_type)
                                 st.metric(
-                                    label=quant_type,
+                                    label=f"{quant_info['icon']} {quant_type}",
                                     value=quant_size,
                                     help=quant_info['description']
                                 )
-                    else:
-                        st.markdown(f"**Original Size:** {model['size']}")
                     
                     st.markdown("---")
 
+# Performance visualization
+def create_performance_chart(df):
+    """Create a performance distribution chart"""
+    laptop_rams = df["Laptop RAM"].apply(extract_numeric_ram).dropna()
+    mobile_rams = df["Mobile RAM"].apply(extract_numeric_ram).dropna()
+    
+    fig = go.Figure()
+    
+    fig.add_trace(go.Histogram(
+        x=laptop_rams,
+        name="Laptop RAM",
+        opacity=0.7,
+        nbinsx=10,
+        marker_color='#1f77b4'
+    ))
+    
+    fig.add_trace(go.Histogram(
+        x=mobile_rams,
+        name="Mobile RAM",
+        opacity=0.7,
+        nbinsx=10,
+        marker_color='#ff7f0e'
+    ))
+    
+    fig.update_layout(
+        title="RAM Distribution Across Devices",
+        xaxis_title="RAM (GB)",
+        yaxis_title="Number of Students",
+        barmode='overlay',
+        height=400,
+        showlegend=True
+    )
+    
+    return fig
+
 # Demo data generator for when Excel files are not available
 def generate_demo_data():
     """Generate demo data for testing when Excel files are missing"""
     demo_data = {
         "Full Name": [
             "Demo Student 1", "Demo Student 2", "Demo Student 3", "Demo Student 4",
-            "Demo Student 5", "Demo Student 6", "Demo Student 7", "Demo Student 8"
+            "Demo Student 5", "Demo Student 6", "Demo Student 7", "Demo Student 8",
+            "Demo Student 9", "Demo Student 10", "Demo Student 11", "Demo Student 12"
         ],
-        "Laptop RAM": ["8GB", "16GB", "4GB", "32GB", "6GB", "12GB", "2GB", "24GB"],
-        "Mobile RAM": ["4GB", "8GB", "3GB", "12GB", "6GB", "4GB", "2GB", "8GB"],
+        "Laptop RAM": ["8GB", "16GB", "4GB", "32GB", "6GB", "12GB", "2GB", "24GB", "64GB", "3GB", "20GB", "10GB"],
+        "Mobile RAM": ["4GB", "8GB", "3GB", "12GB", "6GB", "4GB", "2GB", "8GB", "16GB", "3GB", "6GB", "8GB"],
         "Laptop Operating System": [
             "Windows 11", "macOS Monterey", "Ubuntu 22.04", "Windows 10",
-            "macOS Big Sur", "Fedora 36", "Windows 11", "macOS Ventura"
+            "macOS Big Sur", "Fedora 36", "Windows 11", "macOS Ventura",
+            "Ubuntu 20.04", "Windows 10", "macOS Sonoma", "Pop!_OS 22.04"
         ],
         "Mobile Operating System": [
             "Android 13", "iOS 16", "Android 12", "iOS 15",
-            "Android 14", "iOS 17", "Android 11", "iOS 16"
+            "Android 14", "iOS 17", "Android 11", "iOS 16",
+            "Android 13", "iOS 15", "Android 14", "iOS 17"
         ]
     }
     return pd.DataFrame(demo_data)
@@ -388,10 +490,8 @@ def generate_demo_data():
 def prepare_user_options(df):
     """Safely prepare user options for selectbox, handling NaN values and mixed types"""
     try:
-        # Get unique names and filter out NaN values
         unique_names = df["Full Name"].dropna().unique()
         
-        # Convert to strings and filter out any remaining non-string values
         valid_names = []
         for name in unique_names:
             try:
@@ -401,7 +501,6 @@ def prepare_user_options(df):
             except:
                 continue
         
-        # Create options list with proper string concatenation
         options = ["Select a student..."] + sorted(valid_names)
         return options
     except Exception as e: