Update app.py

app.py

@@ -5,9 +5,9 @@ import pandas as pd
 import torch
 
 # ✅ Get Hugging Face token securely from Space Secrets
-HUGGINGFACE_TOKEN = os.getenv("HF_TOKEN")  # Make sure you set this in HF Secrets!
+HUGGINGFACE_TOKEN = os.getenv("HF_TOKEN")  # Ensure this is set in HF Secrets
 
-# ✅ Load Llama-2 Model & Tokenizer (without login())
+# ✅ Load Llama-2 Model & Tokenizer (optimized)
 model_name = "meta-llama/Llama-2-7b-chat-hf"
 tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=HUGGINGFACE_TOKEN)
 model = AutoModelForCausalLM.from_pretrained(
@@ -15,25 +15,38 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
-# ✅ Function to analyze CSV data
+# ✅ Function to analyze CSV data based on accuracy
 def analyze_csv(file):
     df = pd.read_csv(file.name)  # Read uploaded CSV
-    benchmark_text = df.to_string()  # Convert DataFrame to text
     
-    prompt = f"Analyze the following benchmark scores and tell the best test result:\n\n{benchmark_text}"
+    # Identify the dataset with the highest accuracy
+    best_model = df.loc[df["Accuracy (%)"].idxmax()]  # Find row with max accuracy
     
-    # Generate response using Llama-2
-    output = pipe(prompt, max_length=500, do_sample=True, temperature=0.7)
-    analysis_result = output[0]['generated_text']
+    # Optimized prompt for accuracy analysis
+    prompt = f"""
+    Given the following test benchmark results, determine which dataset performed best in terms of accuracy.
     
-    return analysis_result  # Return the analysis
+    Data Summary:
+    {df[["Dataset", "Accuracy (%)", "Latency (ms)", "Throughput (samples/sec)", "Memory Usage (GB)", "GPU Utilization (%)"]].to_string(index=False)}
+    
+    - Which test achieved the highest accuracy?
+    - What were its corresponding latency, throughput, and resource utilization?
+    - Provide a brief recommendation based on the findings.
+    
+    Highlight the most accurate test.
+    """
+
+    # Generate response using Llama-2 (optimized settings)
+    output = pipe(prompt, max_length=300, do_sample=True, temperature=0.5)
+    
+    return output[0]['generated_text']  # Return optimized analysis
 
 # ✅ Gradio Interface
 iface = gr.Interface(
     fn=analyze_csv,
     inputs=gr.File(label="Upload CSV File"),
     outputs="text",
-    title="Llama-2 CSV Benchmark Analyzer"
+    title="Llama-2 Accuracy-Based CSV Analyzer"
 )
 
 iface.launch()