Updated prompt solution, improved solving efficiency

Former-commit-id: 1f607712ccfcb856435dc4fe669c51ed704a29a4

SRT.py

@@ -115,51 +115,62 @@ class SRT_script():
         start_seg_id = id_range[0]
         end_seg_id = id_range[1]
         
-        def inner_func(input_str):
+        def inner_func(target,input_str):
             response = openai.ChatCompletion.create(
-                    model=model,
+                    #model=model,
+                    model = "gpt-3.5-turbo",
                     messages = [
-                        {"role": "system", "content": "You are a helpful assistant that help calibrates English to Chinese subtitle translations in starcraft2."},
-                        {"role": "system", "content": "You are provided with a translated Chinese transcript; you must modify or split the Chinese sentence to match the meaning and the number of the English transcript exactly one by one. You must not merge ANY Chinese lines, you can only split them but the total Chinese lines MUST equals to number of English lines."},
-                        {"role": "system", "content": "There is no need for you to add any comments or notes, and do not modify the English transcript."},
-                        {"role": "user", "content": 'You are given the English transcript and line number, your task is to merge or split the Chinese to match the exact number of lines in English transcript, no more no less. For example, if there are more Chinese lines than English lines, merge some the Chinese lines to match the number of English lines. If Chinese lines is less than English lines, split some Chinese lines to match the english lines: "{}"'.format(input_str)}
+                        #{"role": "system", "content": "You are a helpful assistant that help calibrates English to Chinese subtitle translations in starcraft2."},
+                        #{"role": "system", "content": "You are provided with a translated Chinese transcript; you must modify or split the Chinese sentence to match the meaning and the number of the English transcript exactly one by one. You must not merge ANY Chinese lines, you can only split them but the total Chinese lines MUST equals to number of English lines."},
+                        #{"role": "system", "content": "There is no need for you to add any comments or notes, and do not modify the English transcript."},
+                        #{"role": "user", "content": 'You are given the English transcript and line number, your task is to merge or split the Chinese to match the exact number of lines in English transcript, no more no less. For example, if there are more Chinese lines than English lines, merge some the Chinese lines to match the number of English lines. If Chinese lines is less than English lines, split some Chinese lines to match the english lines: "{}"'.format(input_str)}
+                        {"role": "system", "content": "你的任务是按照要求合并或拆分句子到指定行数，你需要尽可能保证句意，但必要时可以将一句话分为两行输出"},
+                        {"role": "system", "content": "注意：你只需要输出处理过的中文句子，如果你要输出序号，请使用冒号隔开"},
+                        {"role": "user", "content": '请将下面的句子拆分或组合为{}句:\n{}'.format(target,input_str)}
                     ],
-                    temperature=0.7
+                    #temperature=0.7
+                    temperature = 0.15
                 )
             return response['choices'][0]['message']['content'].strip()
         
+        
         lines = translate.split('\n\n')
         if len(lines) < (end_seg_id - start_seg_id + 1):
             count = 0
             solved = False
             while count<5 and len(lines) != (end_seg_id - start_seg_id + 1):
-                
                 count += 1
                 print("Solving Unmatched Lines|iteration {}".format(count))
-                input_str = "\n"
+                #input_str = "\n"
                 #initialize GPT input
-                for i, seg in enumerate(self.segments[start_seg_id-1:end_seg_id]):
-                    input_str += 'Sentence %d: ' %(i+1)+ seg.source_text + '\n'
-                    #Append to prompt string
-                    #Adds sentence index let GPT keep track of sentence breaks
-                input_str += translate    
+                #for i, seg in enumerate(self.segments[start_seg_id-1:end_seg_id]):
+                #    input_str += 'Sentence %d: ' %(i+1)+ seg.source_text + '\n'
+                #    #Append to prompt string
+                #    #Adds sentence index let GPT keep track of sentence breaks
+                #input_str += translate    
                 #append translate to prompt
-                
                 flag = True
                 while flag:
                     flag = False
+                    #print("translate:")
+                    #print(translate)
                     try:
-                        translate = inner_func(input_str)
+                        #print("target")
+                        #print(end_seg_id - start_seg_id + 1)
+                        translate = inner_func(end_seg_id - start_seg_id + 1,translate)
                     except Exception as e:
                         print("An error has occurred during solving unmatched lines:",e)
                         print("Retrying...")
                         flag = True
-                        
-                lines = translate.split('\n\n')
+                lines = translate.split('\n')
+                #print("result")
+                #print(len(lines))
+                
             if len(lines) < (end_seg_id - start_seg_id + 1):
                 print("Failed Solving unmatched lines, Manually parse needed")
             
-            
+            if not os.path.exists("./logs"):
+                os.mkdir("./logs")
             if video_link:
                 log_file = "./logs/log_link.csv"
                 log_exist = os.path.exists(log_file)
@@ -193,7 +204,7 @@ class SRT_script():
                     if i == len(lines) - 1:
                         break
                 try:
-                    seg.translation = lines[i].split("：" or ":")[1]
+                    seg.translation = lines[i].split("：" or ":" or ".")[1]
                 except:
                     seg.translation = lines[i]
                     #print(lines[i])