def validate_article(generated_article, lesson_words, base_chars, original_word_count):

app.py

@@ -32,12 +32,31 @@ def extract_article_from_content(article_text):
     return article_text
 
 def validate_article(generated_article, lesson_words, base_chars, original_word_count):
-    clean_article = "".join(char for char in generated_article if char not in "，。！？；：「」『』（）《》【】'\n'")
+    clean_article = "".join(char for char in generated_article if char not in "、，。！？；：「」『』（）《》【】'\n'")
     not_every_new_word_is_used = not all(word in clean_article for word in lesson_words.split())
     word_out_of_range = not set(clean_article).issubset(set(lesson_words + base_chars))
     new_word_count = len(clean_article)
     word_count_error = not (0.9 * original_word_count <= new_word_count <= 1.1 * original_word_count)
-    return not_every_new_word_is_used, word_out_of_range, word_count_error
+
+    lesson_words_not_in_new_article = [word for word in [char for char in lesson_words] if word not in clean_article]
+    words_not_in_both = [word for word in lesson_words_not_in_new_article if word not in base_chars]
+    additional_words = set([word for word in [char for char in clean_article] if word not in [char for char in lesson_words] and word not in [char for char in base_chars]])
+    count_of_words_in_new_article = len(clean_article)
+
+    result = {
+        "not_every_new_word_is_used": not_every_new_word_is_used,
+        "word_out_of_range": word_out_of_range,
+        "word_count_error": word_count_error,
+
+        "lesson_words_not_in_new_article": lesson_words_not_in_new_article,
+        "words_not_in_both": words_not_in_both,
+        "additional_words": additional_words,
+        "count_of_words_in_new_article": count_of_words_in_new_article
+    }
+
+
+
+    return result
 
 def generate_new_article(lesson_words, original_article, original_word_count, base_chars, model_name):
     attempt = 0
@@ -81,9 +100,11 @@ def generate_new_article(lesson_words, original_article, original_word_count, ba
         generated_text = response.choices[0].message.content.strip()
         generated_article = extract_article_from_content(generated_text)
 
-        not_every_new_word_is_used, word_out_of_range, word_count_error = validate_article(
-            generated_article, lesson_words, base_chars, original_word_count)
-        
+        validate_article_result = validate_article(generated_article, lesson_words, base_chars, original_word_count)
+        not_every_new_word_is_used = validate_article_result['not_every_new_word_is_used']
+        word_out_of_range = validate_article_result['word_out_of_range']
+        word_count_error = validate_article_result['word_count_error']
+
         print("====validate_article====")
         print(f"not_every_new_word_is_used: {not_every_new_word_is_used}")
         print(f"word_out_of_range: {word_out_of_range}")
@@ -93,8 +114,20 @@ def generate_new_article(lesson_words, original_article, original_word_count, ba
         if not not_every_new_word_is_used and not word_out_of_range and not word_count_error:
             print("Generated article is valid")
             break
+        else:
+            print("Generated article is invalid")
+            error_messages = []
+            if not_every_new_word_is_used:
+                error_messages.append("Not every new word is used in the article.")
+            if word_out_of_range:
+                error_messages.append("The article contains words that are not in the new words or word library.")
+            if word_count_error:
+                error_messages.append(f"The word count of the new article deviates more than 10% from the original ({original_word_count}).")
+            
+            prompt += "\n".join(error_messages) + "\n"  # Append the error messages to the prompt for the next attempt
+
 
-    return generated_article
+    return generated_article, validate_article_result
 
 def load_csv(file):
     try:
@@ -112,7 +145,6 @@ def load_csv(file):
     return "", "", 0, ""
 
 with gr.Blocks() as demo:
-    gr.Markdown("import CSV file to load data.")
     csv_file_input = gr.File(label="Upload CSV file (Columns: lesson_words, original_article, original_word_count, base_chars)")
     load_button = gr.Button("Load from CSV")
     
@@ -129,33 +161,40 @@ with gr.Blocks() as demo:
             generate_button1 = gr.Button("Generate Article - gpt-4-0125-preview")
         with gr.Column(): 
             output_text1 = gr.Textbox(label="Generated Article - gpt-4-0125-preview")
+        with gr.Column(): 
+            validate_article_result_1 = gr.JSON()
     with gr.Row():
         with gr.Column(): 
             model_2 = gr.Dropdown(label="Model 2", choices=model_list, value="gpt-3.5-turbo")
             generate_button2 = gr.Button("Generate Article - gpt-3.5-turbo")
         with gr.Column(): 
             output_text2 = gr.Textbox(label="Generated Article - gpt-3.5-turbo")
+        with gr.Column(): 
+            validate_article_result_2 = gr.JSON()
     with gr.Row():
         with gr.Column():
             model_3 = gr.Dropdown(label="Model 3", choices=model_list, value="gpt-4")
             generate_button3 = gr.Button("Generate Article - gpt-4")
         with gr.Column(): 
             output_text3 = gr.Textbox(label="Generated Article - gpt-4")
+        with gr.Column():
+            # validate_article_result_3 Json format
+            validate_article_result_3 = gr.JSON()
 
     generate_button1.click(
         generate_new_article,
         inputs=[lesson_words_input, original_article_input, original_word_count_input, base_chars_input, model_1],
-        outputs=output_text1
+        outputs=[output_text1, validate_article_result_1]
     )
     generate_button2.click(
         generate_new_article,
         inputs=[lesson_words_input, original_article_input, original_word_count_input, base_chars_input, model_2],
-        outputs=output_text2
+        outputs=[output_text2, validate_article_result_2]
     )
     generate_button3.click(
         generate_new_article,
         inputs=[lesson_words_input, original_article_input, original_word_count_input, base_chars_input, model_3],
-        outputs=output_text3
+        outputs=[output_text3, validate_article_result_3]
     )
 
     # 為其他模型添加點擊事件