Multi-files input supported

app.py

@@ -31,7 +31,7 @@ with gr.Blocks(theme="dark") as demo:
                         label='Enter your OpenAI API key here', 
                         type='password')
                 
-                file = gr.File(label='Upload your .txt file here', file_types=['.txt'])
+                file = gr.File(label='Upload your .txt file here', file_types=['.txt'], file_count = 'multiple')
                 
                 questions = gr.CheckboxGroup(choices = QUESTIONS, value = QUESTIONS, label="Questions", info="Please select the question you want to ask")
 
@@ -42,14 +42,23 @@ with gr.Blocks(theme="dark") as demo:
             with gr.Group():
                 gr.Markdown(f'<center><h1>Output</h1></center>')
                 gr.Markdown(f'<center><p>The answer to your question is :</p></center>')
+                filename_box = gr.Textbox(label = "File")
                 question_box = gr.Textbox(label='Question')
                 answer_box = gr.Textbox(label='Answer')
                 # reference_box = gr.Textbox(label='Reference')
 
                 highlighted_text = gr.outputs.HTML(label="Highlighted Text")
-                with gr.Row():
-                    btn_last_question = gr.Button(value='Last Question')
-                    btn_next_question = gr.Button(value='Next Question')
+                with gr.Group():
+                    gr.Markdown("<center><h4>Please select different questions</h4></center>")
+                    with gr.Row():
+                        btn_last_question = gr.Button(value='Last Question')
+                        btn_next_question = gr.Button(value='Next Question')
+
+                with gr.Group():
+                    gr.Markdown("<center><h4>Please select different passages</h4></center>")
+                    with gr.Row():
+                        btn_last_passage = gr.Button(value='Last Passage')
+                        btn_next_passage = gr.Button(value='Next Passage')
             
             # Correctness
             with gr.Group():
@@ -103,7 +112,7 @@ with gr.Blocks(theme="dark") as demo:
     btn_submit_txt.click(
             backend.process_file,
             inputs=[file, questions, openai_key],
-            outputs=[question_box, answer_box, highlighted_text, correct_answer, correct_reference],
+            outputs=[filename_box, question_box, answer_box, highlighted_text, correct_answer, correct_reference],
         )
     
     btn_submit_correctness.click(   # TODO
@@ -114,15 +123,25 @@ with gr.Blocks(theme="dark") as demo:
     
     # Switch question button
     btn_last_question.click(
-            backend.process_last,
-            outputs=[question_box, answer_box, highlighted_text, correct_answer, correct_reference, save_results, answer_correct, reference_correct],
-        )
+        backend.process_last,
+        outputs=[filename_box, question_box, answer_box, highlighted_text, correct_answer, correct_reference, save_results, answer_correct, reference_correct],
+    )
 
     btn_next_question.click(
             backend.process_next,
-            outputs=[question_box, answer_box, highlighted_text, correct_answer, correct_reference, save_results, answer_correct, reference_correct],
+            outputs=[filename_box, question_box, answer_box, highlighted_text, correct_answer, correct_reference, save_results, answer_correct, reference_correct],
         )
     
+    # Switch passwage button
+    btn_last_passage.click(
+        backend.switch_last_passage,
+        outputs=[filename_box, question_box, answer_box, highlighted_text, correct_answer, correct_reference, save_results, answer_correct, reference_correct],
+    )
+    btn_next_passage.click(
+        backend.switch_next_passage,
+        outputs=[filename_box, question_box, answer_box, highlighted_text, correct_answer, correct_reference, save_results, answer_correct, reference_correct],
+    )
+
     # Download button
     btn_download_answer.click(
             backend.download_answer,

backend.py

@@ -12,7 +12,7 @@ class Backend:
         self.agent = OpenAI()
         self.prompt = Prompt()
 
-    def read_file(self, file):
+    def read_file_single(self, file):
         # read the file
         if file is not None:
             with open(file.name, 'r') as f:
@@ -21,6 +21,19 @@ class Backend:
             raise gr.Error("You need to upload a file first")
         return text
     
+    def read_file(self, files):
+        # read the file
+        text_list = []
+        self.filename_list = []
+        if files is not None:
+            for file in files:
+                with open(file.name, 'r', encoding='utf-8') as f:
+                    text_list.append(f.read())
+                    self.filename_list.append(file.name.split('\\')[-1])
+        else:
+            raise gr.Error("You need to upload a file first")
+        return text_list
+    
     def highlight_text(self, text, highlight_list):
         # Find the original sentences
         # Split the passage into sentences
@@ -40,42 +53,61 @@ class Backend:
         text = text.replace('\n', f" <br /> ")
 
         # add scroll bar
-        text = f'<div style="height: 500px; overflow: auto;">{text}</div>'
+        text = f'<div style="height: 300px; overflow: auto;">{text}</div>'
 
         return text
     
-    def process_file(self, file, questions, openai_key):
+    def process_file(self, file, questions, openai_key, progress = gr.Progress()):
         # record the questions
         self.questions = questions
 
-        # get the text
-        self.text = self.read_file(file)
+        # get the text_list
+        self.text_list = self.read_file(file)
 
         # make the prompt
-        prompt = self.prompt.get(self.text, questions, 'v3')
+        prompt_list = [self.prompt.get(text, questions, 'v3') for text in self.text_list]
 
         # interact with openai
-        res = self.agent(prompt, with_history = False, temperature = 0.1, model = 'gpt-3.5-turbo-16k', api_key = openai_key)
-        res = self.prompt.process_result(res, 'v3')
+        self.res_list = []
+        for prompt in progress.tqdm(prompt_list, desc = 'Generating answers...'):
+            res = self.agent(prompt, with_history = False, temperature = 0.1, model = 'gpt-3.5-turbo-16k', api_key = openai_key)
+            res = self.prompt.process_result(res, 'v3')
+            self.res_list.append(res)
+
+        # Use the first file as default
+        # Use the first question for multiple questions
+        gpt_res = self.res_list[0]
+        self.gpt_result = gpt_res
 
-        # for multiple questions
-        self.gpt_result = res
-        self.curret_question = 0
+        self.current_question = 0
         self.totel_question = len(res.keys())
+        self.current_passage = 0
+        self.total_passages = len(self.res_list)
 
         # make a dataframe to record everything
-        self.ori_answer_df = pd.DataFrame(res).T
-        self.answer_df = pd.DataFrame(res).T
+        self.ori_answer_df = pd.DataFrame()
+        self.answer_df = pd.DataFrame()
+        for i, res in enumerate(self.res_list):
+            tmp = pd.DataFrame(res).T
+            tmp = tmp.reset_index()
+            tmp = tmp.rename(columns={"index":"question_id"})
+            tmp['filename'] = self.filename_list[i]
+            tmp['question'] = self.questions
+            self.ori_answer_df = pd.concat([tmp, self.ori_answer_df])
+            self.answer_df = pd.concat([tmp, self.answer_df])
 
         # default fist question
         res = res['Question 1']
-        question = self.questions[self.curret_question]
+        question = self.questions[self.current_question]
         self.answer = res['answer']
+        self.text = self.text_list[0]
         self.highlighted_out = res['original sentences']
         highlighted_out_html = self.highlight_text(self.text, self.highlighted_out)
         self.highlighted_out = '\n'.join(self.highlighted_out)
+
+        file_name = self.filename_list[self.current_passage]
         
-        return question, self.answer, highlighted_out_html, self.answer, self.highlighted_out
+        return file_name, question, self.answer, highlighted_out_html, self.answer, self.highlighted_out
     
     def process_results(self, answer_correct, correct_answer, reference_correct, correct_reference):
         if not hasattr(self, 'clicked_correct_answer'):
@@ -87,75 +119,138 @@ class Backend:
         if not hasattr(self, 'answer_df'):
             raise gr.Error("You need to submit the document first")
         
-        if self.curret_question >= self.totel_question or self.curret_question < 0:
+        if self.current_question >= self.totel_question or self.current_question < 0:
             raise gr.Error("No more questions, please return back")
                 
         # record the answer
-        self.answer_df.loc[f'Question {self.curret_question + 1}', 'answer_correct'] = answer_correct
-        self.answer_df.loc[f'Question {self.curret_question + 1}', 'reference_correct'] = reference_correct
+        condition = (self.answer_df['question_id'] == f'Question {self.current_question + 1}' ) & \
+            (self.answer_df['filename'] == self.filename_list[self.current_passage]) 
+        self.answer_df.loc[condition, 'answer_correct'] = answer_correct
+        self.answer_df.loc[condition, 'reference_correct'] = reference_correct
+
+        # self.answer_df.loc[f'Question {self.current_question + 1}', 'answer_correct'] = answer_correct
+        # self.answer_df.loc[f'Question {self.current_question + 1}', 'reference_correct'] = reference_correct
         
         if self.clicked_correct_answer == True:
             if hasattr(self, 'answer'):
-                self.answer_df.loc[f'Question {self.curret_question + 1}', 'correct_answer'] = self.answer
+                self.answer_df.loc[condition, 'correct_answer'] = self.answer
             else:
                 raise gr.Error("You need to submit the document first")
         else:
-            self.answer_df.loc[f'Question {self.curret_question + 1}', 'correct_answer'] = correct_answer
+            # self.answer_df.loc[f'Question {self.current_question + 1}', 'correct_answer'] = correct_answer
+            self.answer_df.loc[condition, 'correct_answer'] = correct_answer
         
         if self.clicked_correct_reference == True:
             if hasattr(self, 'highlighted_out'):
-                self.answer_df.loc[f'Question {self.curret_question + 1}', 'correct_reference'] = self.highlighted_out
+                self.answer_df.loc[condition, 'correct_reference'] = self.highlighted_out
             else:
                 raise gr.Error("You need to submit the document first")
         else:
-            self.answer_df.loc[f'Question {self.curret_question + 1}', 'correct_reference'] = correct_reference
+            self.answer_df.loc[condition, 'correct_reference'] = correct_reference
         
         gr.Info('Results saved!')
         return "Results saved!"
     
     def process_next(self):
-        self.curret_question += 1
+        self.current_question += 1
         if hasattr(self, 'clicked_correct_answer'):
             del self.clicked_correct_answer
         if hasattr(self, 'clicked_correct_reference'):
             del self.clicked_correct_reference
 
-        if self.curret_question >= self.totel_question:
-            # self.curret_question -= 1
-            return "No more questions!", "No more questions!", "No more questions!", 'No more questions!', 'No more questions!', 'Still need to click the button above to save the results', None, None
+        if self.current_question >= self.totel_question:
+            # self.current_question -= 1
+            return "No more questions!", "No more questions!", "No more questions!", "No more questions!", 'No more questions!', 'No more questions!', 'Still need to click the button above to save the results', None, None
         else:
-            res = self.gpt_result[f'Question {self.curret_question + 1}']
-            question = self.questions[self.curret_question]
+            res = self.gpt_result[f'Question {self.current_question + 1}']
+            question = self.questions[self.current_question]
             self.answer = res['answer']
             self.highlighted_out = res['original sentences']
             highlighted_out_html = self.highlight_text(self.text, self.highlighted_out)
             self.highlighted_out = '\n'.join(self.highlighted_out)
-            return question, self.answer, highlighted_out_html, 'Please judge on the generated answer', 'Please judge on the generated answer', 'Still need to click the button above to save the results', None, None
+            file_name = self.filename_list[self.current_passage]
+
+            return file_name, question, self.answer, highlighted_out_html, 'Please judge on the generated answer', 'Please judge on the generated answer', 'Still need to click the button above to save the results', None, None
 
     def process_last(self):
-        self.curret_question -= 1
+        self.current_question -= 1
 
+        # To make sure to correct the answer first
         if hasattr(self, 'clicked_correct_answer'):
             del self.clicked_correct_answer
         if hasattr(self, 'clicked_correct_reference'):
             del self.clicked_correct_reference
         
-        if self.curret_question < 0:
-            # self.curret_question += 1
-            return "No more questions!", "No more questions!", "No more questions!", 'No more questions!', 'No more questions!', 'Still need to click the button above to save the results', None, None
+        # check question boundary
+        if self.current_question < 0:
+            # self.current_question += 1
+            return "No more questions!", "No more questions!", "No more questions!", "No more questions!", 'No more questions!', 'No more questions!', 'Still need to click the button above to save the results', None, None
         else:
-            res = self.gpt_result[f'Question {self.curret_question + 1}']
-            question = self.questions[self.curret_question]
+            res = self.gpt_result[f'Question {self.current_question + 1}']
+            question = self.questions[self.current_question]
             self.answer = res['answer']
             self.highlighted_out = res['original sentences']
             highlighted_out_html = self.highlight_text(self.text, self.highlighted_out)
             self.highlighted_out = '\n'.join(self.highlighted_out)
-            return question, self.answer, highlighted_out_html, 'Please judge on the generated answer', 'Please judge on the generated answer', 'Still need to click the button above to save the results', None, None
-        
+            file_name = self.filename_list[self.current_passage]
+            return file_name, question, self.answer, highlighted_out_html, 'Please judge on the generated answer', 'Please judge on the generated answer', 'Still need to click the button above to save the results', None, None
+
+    def switch_next_passage(self):
+        self.current_question = 0
+
+        # To make sure to correct the answer first
+        if hasattr(self, 'clicked_correct_answer'):
+            del self.clicked_correct_answer
+        if hasattr(self, 'clicked_correct_reference'):
+            del self.clicked_correct_reference
+
+        self.current_passage += 1
+
+        if self.current_passage >= self.total_passages:
+            # self.current_passage -= 1
+            return "No more passages!", "No more passages!", "No more passages!", "No more passages!", 'No more passages!', 'No more passages!', 'Still need to click the button above to save the results', None, None
+        else:
+            gpt_res = self.res_list[self.current_passage]
+            self.gpt_result = gpt_res
+            res = self.gpt_result[f'Question {self.current_question + 1}']
+            question = self.questions[self.current_question]
+            self.answer = res['answer']
+            self.highlighted_out = res['original sentences']
+            highlighted_out_html = self.highlight_text(self.text, self.highlighted_out)
+            self.highlighted_out = '\n'.join(self.highlighted_out)
+            file_name = self.filename_list[self.current_passage]
+            return file_name, question, self.answer, highlighted_out_html, 'Please judge on the generated answer', 'Please judge on the generated answer', 'Still need to click the button above to save the results', None, None
+
+    def switch_last_passage(self):
+        self.current_question = 0
+
+        # To make sure to correct the answer first
+        if hasattr(self, 'clicked_correct_answer'):
+            del self.clicked_correct_answer
+        if hasattr(self, 'clicked_correct_reference'):
+            del self.clicked_correct_reference
+
+        self.current_passage -= 1
+
+        if self.current_passage < 0:
+            # self.current_passage += 1
+            return "No more passages!", "No more passages!", "No more passages!", "No more passages!", 'No more passages!', 'No more passages!', 'Still need to click the button above to save the results', None, None
+        else:
+            gpt_res = self.res_list[self.current_passage]
+            self.gpt_result = gpt_res
+            res = self.gpt_result[f'Question {self.current_question + 1}']
+            question = self.questions[self.current_question]
+            self.answer = res['answer']
+            self.highlighted_out = res['original sentences']
+            highlighted_out_html = self.highlight_text(self.text, self.highlighted_out)
+            self.highlighted_out = '\n'.join(self.highlighted_out)
+            file_name = self.filename_list[self.current_passage]
+            return file_name, question, self.answer, highlighted_out_html, 'Please judge on the generated answer', 'Please judge on the generated answer', 'Still need to click the button above to save the results', None, None
+
     def download_answer(self, path = './tmp', name = 'answer.xlsx'):
         os.makedirs(path, exist_ok = True)
         path = os.path.join(path, name)
-        self.ori_answer_df['questions'] = self.questions
+        # self.ori_answer_df['questions'] = self.questions
         self.ori_answer_df.to_excel(path, index = False)
 
         return path
@@ -163,7 +258,7 @@ class Backend:
     def download_corrected(self, path = './tmp', name = 'corrected_answer.xlsx'):
         os.makedirs(path, exist_ok = True)
         path = os.path.join(path, name)
-        self.answer_df['questions'] = self.questions
+        # self.answer_df['questions'] = self.questions
         self.answer_df.to_excel(path, index = False)
 
         return path

openai.py

@@ -32,7 +32,14 @@ class OpenAI:
         }, headers={
             'Authorization': f"Bearer {api_key}"
         })
-        # print(resp.json())
+        print(resp)
         self.history.append(resp.json()['choices'][0]['message'])
+        res = resp.json()['choices'][0]['message']['content']
 
-        return resp.json()['choices'][0]['message']['content']
+        # with open("tmp_res.txt", 'w') as f:
+        #     f.write(res)            
+
+        # with open("tmp_res.txt", 'r') as f:
+        #     res = f.read()   
+
+        return res