edit apps

app.py

@@ -137,8 +137,8 @@ Just respond to the instruction directly. DO NOT add additional explanations or
 def split_draft(draft, split_char = '\n\n'):
     # 将draft切分为多个段落
     # split_char: '\n\n'
-    draft_paragraphs = draft.split(split_char)
-    draft_paragraphs = [d for d in draft_paragraphs if d]
+    paragraphs = draft.split(split_char)
+    draft_paragraphs = [para for para in paragraphs if len(para)>5]
     # print(f"The draft answer has {len(draft_paragraphs)}")
     return draft_paragraphs
 
@@ -197,7 +197,7 @@ If you find the answer is right and do not need to add more details, just output
 **IMPORTANT**
 Try to keep the structure (multiple paragraphs with its subtitles) in the revised answer and make it more structual for understanding.
 Add more details from retrieved text to the answer.
-Split the paragraphs with `\n\n` characters.
+Split the paragraphs with \n\n characters.
 Just output the revised answer directly. DO NOT add additional explanations or annoucement in the revised answer unless you are asked to.
 '''
     # openai_client = OpenAI(api_key = openai.api_key)
@@ -218,6 +218,31 @@ Just output the revised answer directly. DO NOT add additional explanations or a
     ).choices[0].message.content
     return revised_answer
 
+def get_reflect_answer(question, answer):
+    reflect_prompt = '''
+Add a title to each paragraph in the answer and output the final answer using markdown format. This will make the answer to this question look more structured for better understanding.
+**IMPORTANT**
+Try to keep the structure (multiple paragraphs with its subtitles) in the response and make it more structual for understanding.
+Split the paragraphs with \n\n characters.
+Just output the revised answer directly. DO NOT add additional explanations or annoucement in the revised answer unless you are asked to.
+'''
+    openai_client = OpenAI(api_key = os.getenv('OPENAI_API_KEY'))
+    reflected_answer = openai_client.chat.completions.create(
+        model="gpt-3.5-turbo",
+        messages=[
+                {
+                    "role": "system",
+                    "content": chatgpt_system_prompt
+                },
+                {
+                    "role": "user",
+                    "content": f"##Question:\n{question}\n\n##Answer:\n{answer}\n\n##Instruction:\n{reflect_prompt}"
+                }
+            ],
+            temperature = 1.0
+    ).choices[0].message.content
+    return reflected_answer
+
 def get_query_wrapper(q, question, answer):
     result = get_query(question, answer)
     q.put(result)  # 将结果放入队列
@@ -230,6 +255,10 @@ def get_revise_answer_wrapper(q, question, answer, content):
     result = get_revise_answer(question, answer, content)
     q.put(result)
 
+def get_reflect_answer_wrapper(q, question, answer):
+    result = get_reflect_answer(question, answer)
+    q.put(result)
+
 from multiprocessing import Process, Queue
 def run_with_timeout(func, timeout, *args, **kwargs):
     q = Queue()  # 创建一个Queue对象用于进程间通信
@@ -239,12 +268,12 @@ def run_with_timeout(func, timeout, *args, **kwargs):
     # 等待进程完成或超时
     p.join(timeout)
     if p.is_alive():
-        print(f"{datetime.now()} [INFO] 函数{str(func)}执行已超时({timeout}s)，正在终止进程...")
+        print(f"{datetime.now()} [INFO] Function {str(func)} running timeout ({timeout}s), terminating...")
         p.terminate()  # 终止进程
         p.join()  # 确保进程已经终止
         result = None  # 超时情况下，我们没有结果
     else:
-        print(f"{datetime.now()} [INFO] 函数{str(func)}执行成功完成")
+        print(f"{datetime.now()} [INFO] Function {str(func)} executed successfully.")
         result = q.get()  # 从队列中获取结果
     return result
 
@@ -271,38 +300,38 @@ def generate_diff_html(text1, text2):
 newline_char = '\n'
 
 def rat(question):
-    print(f"{datetime.now()} [INFO] 生成草稿中...")
+    print(f"{datetime.now()} [INFO] Generating draft...")
     draft = get_draft(question)
-    print(f"{datetime.now()} [INFO] 获得草稿")
+    print(f"{datetime.now()} [INFO] Return draft.")
     # print(f"##################### DRAFT #######################")
     # print(draft)
     # print(f"#####################  END  #######################")
 
-    print(f"{datetime.now()} [INFO] 处理草稿...")
+    print(f"{datetime.now()} [INFO] Processing draft ...")
     draft_paragraphs = split_draft(draft)
-    print(f"{datetime.now()} [INFO] 草稿被切分为{len(draft_paragraphs)}部分")
+    print(f"{datetime.now()} [INFO] Draft is splitted into {len(draft_paragraphs)} sections.")
     answer = ""
     for i, p in enumerate(draft_paragraphs):
         # print(str(i)*80)
-        print(f"{datetime.now()} [INFO] 修改第{i+1}/{len(draft_paragraphs)}部分...")
+        print(f"{datetime.now()} [INFO] Revising {i+1}/{len(draft_paragraphs)} sections ...")
         answer = answer + '\n\n' + p
         # print(f"[{i}/{len(draft_paragraphs)}] Original Answer:\n{answer.replace(newline_char, ' ')}")
 
         # query = get_query(question, answer)
-        print(f"{datetime.now()} [INFO] 生成对应Query...")
+        print(f"{datetime.now()} [INFO] Generating query ...")
         res = run_with_timeout(get_query_wrapper, 10, question, answer)
         if not res:
-            print(f"{datetime.now()} [INFO] 生成检索词超时，跳过后续步骤...")
+            print(f"{datetime.now()} [INFO] Generating query timeout, skipping...")
             continue
         else:
             query = res
         print(f">>> {i}/{len(draft_paragraphs)} Query: {query.replace(newline_char, ' ')}")
 
-        print(f"{datetime.now()} [INFO] 获取网页内容...")
+        print(f"{datetime.now()} [INFO] Crawling network pages ...")
         # content = get_content(query)
         res = run_with_timeout(get_content_wrapper, 10, query)
         if not res:
-            print(f"{datetime.now()} [INFO] 获取网页内容超时，跳过后续步骤...")
+            print(f"{datetime.now()} [INFO] Parsing network pages timeout, skipping ...")
             continue
         else:
             content = res
@@ -310,21 +339,25 @@ def rat(question):
         for j, c in enumerate(content):
             if  j > 2:
                 break
-            print(f"{datetime.now()} [INFO] 根据网页内容修改对应答案...[{j}/{min(len(content),3)}]")
+            print(f"{datetime.now()} [INFO] Revising answers with retrieved network pages...[{j}/{min(len(content),3)}]")
             # answer = get_revise_answer(question, answer, c)
             res = run_with_timeout(get_revise_answer_wrapper, 15, question, answer, c)
             if not res:
-                print(f"{datetime.now()} [INFO] 修改答案超时，跳过后续步骤...")
+                print(f"{datetime.now()} [INFO] Revising answers timeout, skipping ...")
                 continue
             else:
                 diff_html = generate_diff_html(answer, res)
                 display(HTML(diff_html))
                 answer = res
-            print(f"{datetime.now()} [INFO] 答案修改完成[{j}/{min(len(content),3)}]")
+            print(f"{datetime.now()} [INFO] Answer revised [{j}/{min(len(content),3)}]")
         # print(f"[{i}/{len(draft_paragraphs)}] REVISED ANSWER:\n {answer.replace(newline_char, ' ')}")
         # print()
+    res = run_with_timeout(get_reflect_answer_wrapper, 30, question, answer)
+    if not res:
+        print(f"{datetime.now()} [INFO] Reflecting answers timeout, skipping next steps...")
+    else:
+        answer = res
     return draft, answer
-    # return answer
 
 page_title = "RAT: Retrieval Augmented Thoughts Elicit Context-Aware Reasoning in Long-Horizon Generation"
 page_md = """