support custom test case

app.py

@@ -21,15 +21,13 @@ from multiprocessing import Process, Queue
 import io
 import sys
 
-def unsafe_execute(problem, completion, timeout=3):
+def unsafe_execute(prompt, completion, suffix, test_case, timeout=3):
     check_program = (
-        problem["prompt"]
+        prompt
         + completion
-        + problem["suffix"]
-        + "\n"
-        + problem["test"]
-        + "\n"
-        + f"check({problem['entry_point']})"
+        + suffix
+        + "\n\n"
+        + test_case
     )
 
     # 重定向标准输出和标准错误
@@ -433,16 +431,11 @@ def get_example_input():
     test_case = problem['test']
     test_case = test_case.replace('def check(candidate):', 'def run_test():')
     test_case = test_case.replace('candidate', problem['entry_point'])
+    test_case = test_case + '\n\n\nrun_test()'
     return prefix, '', suffix, test_case, task_id, ''
 
-def check_result(task_id, completion):
-    # 从数据集中读取问题
-    problem = problems[task_id]
-    print('run check')
-    print(task_id)
-    print(completion)
-    # 这里假设 `unsafe_execute` 是一个可以执行代码并返回结果的函数
-    result = unsafe_execute(problem, completion)
+def check_result(prompt, completion, suffix, test_case):
+    result = unsafe_execute(prompt, completion, suffix, test_case)
     return result
 
 
@@ -463,8 +456,8 @@ def create_chatbot_demo():
         with gr.Row():
             sample_btn = gr.Button("Example Prompt")
             generate_btn = gr.Button("Generate", variant="primary")
-            clear_btn = gr.Button("Clear")
             check_btn = gr.Button("Run test case")
+            clear_btn = gr.Button("Clear")
 
         with gr.Row():
             with gr.Column():
@@ -620,7 +613,7 @@ def create_chatbot_demo():
         check_btn.click(
             fn=check_result,
             inputs=[task_id_input, output_vis],
-            outputs=[result_output],
+            outputs=[prefix_input, output_vis, suffix_input, test_case_input]
             queue=False
         )