Update app.py

app.py

@@ -7,27 +7,36 @@ def run_eval(hf_token, subset, task, n_samples):
         model="ibm-granite/granite-3.3-8b-instruct",
         token=hf_token
     )
-    ds = load_dataset("RMT-team/babilong", subset)[task]
     
+    ds = load_dataset("RMT-team/babilong", subset)
+    samples = ds[task]
+    
+    results = []
     correct = 0
-    for i in range(int(n_samples)):
-        s = ds[i]
+    n = int(n_samples)
+    
+    for i in range(n):
+        s = samples[i]
         prompt = f"{s['input']}\n\nQuestion: {s['question']}\nAnswer:"
         pred = client.text_generation(prompt, max_new_tokens=10).strip().lower()
-        if s['target'].lower() in pred:
+        gold = s['target'].lower()
+        match = gold in pred
+        if match:
             correct += 1
+        results.append(f"[{i+1}] Q: {s['question']} | Gold: {gold} | Pred: {pred} | {'OK' if match else 'WRONG'}")
     
-    return f"Accuracy: {correct}/{n_samples} = {correct/int(n_samples):.2%}"
+    summary = f"Accuracy: {correct}/{n} = {correct/n:.2%}\n\n" + "\n".join(results)
+    return summary
 
 demo = gr.Interface(
     fn=run_eval,
     inputs=[
         gr.Textbox(label="HF Token", type="password"),
-        gr.Dropdown(["0k","1k","2k","4k","8k"], label="Subset", value="0k"),
-        gr.Dropdown(["qa1","qa2","qa3"], label="Task", value="qa1"),
-        gr.Slider(5, 100, value=20, step=5, label="Samples"),
+        gr.Dropdown(["0k","1k","2k","4k","8k"], label="Context Length", value="0k"),
+        gr.Dropdown(["qa1","qa2","qa3","qa4","qa5"], label="Task", value="qa1"),
+        gr.Slider(5, 50, value=10, step=5, label="Number of Samples"),
     ],
-    outputs=gr.Textbox(label="Result")
+    outputs=gr.Textbox(label="Results", lines=30)
 )
 
 demo.launch()