Update to support highlighting differences

app.py

@@ -115,7 +115,6 @@ def split_article(article, trigger):
     paragraphs = article.replace("\\c\\c", "\c\c").split("\\\\c\\\\c")
     format_pars = [par for par in paragraphs]
     pars = [str(par) + " -- " + str(trigger) for par in paragraphs]
-    # pd.DataFrame({"paragraph": pars}).to_csv("./util/experiments/input_paragraphs.csv")
     formatted_input = "\n".join(format_pars)
     return pars, formatted_input 
 
@@ -123,20 +122,16 @@ def config():
     load_dotenv()
 
 def call_gpt(paragraph, trigger):
-    # openai.api_key = os.environ.get("GPT_API")
-    openai.api_key = "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"
+    openai.api_key = os.environ.get("GPT_API")
     tokenizer = BartTokenizer.from_pretrained("theQuert/NetKUp-tokenzier")
     inputs_for_gpt = f"""
-As an article writer, your task is to provide an updated paragraph in the length same as non-updated paragraph based on the given non-updated paragraph and a triggered news.
+s an article writer, your task is to provide an updated paragraph in the length same as non-updated paragraph based on the given non-updated paragraph and a triggered news.Remember, the length of updated paragraph is restricted into a single paragraph.
     Non-updated paragraph:
     {paragraph}
 
     Triggered News:
     {trigger}
         """
-        # merged_with_prompts.append(merged.strip())
-        # pd.DataFrame({"paragraph": merged_with_prompts}).to_csv("./experiments/paragraphs_with_prompts.csv")
-
     completion = openai.ChatCompletion.create(
          model = "gpt-3.5-turbo",
          messages = [
@@ -144,6 +139,8 @@ As an article writer, your task is to provide an updated paragraph in the length
          ]
      )
     response = completion.choices[0].message.content
+    if "<"+response.split("<")[-1].strip() == "<"+paragraph.split("<")[-1].strip(): response = response 
+    else: response = response + " <"+paragraph.split("<")[-1].strip()
     return str(response)
 
 def call_vicuna(paragraphs_tirgger):
@@ -165,9 +162,19 @@ As an article writer, your task is to provide an updated paragraph in the length
 
     
 def main(input_article, input_trigger):
-    csv_path = "./util/experiments/input_paragraphs.csv"
-    if os.path.isfile(csv_path):
-        os.remove(csv_path)
+    paths = [".util/experiments/input_paragraphs.csv",
+             "./util.experiments/formatted_input.txt",
+             "./util/experiments/updated_article.txt",
+             "./util/experiments/paragraphs_needed.txt",
+             "./util/experiments/updated_paragraphs.txt",
+             "./util/experiments/paragraphs_with_prompts.csv",
+             "./util/experiments/classification.csv",
+             "./util/experiments/paragraphs_needed.csv",
+             "./util/experiments/par_with_class.csv"]
+    for path in paths: 
+        try:
+            if os.path.isfile(path): os.remove(path)
+        except: pass
     modified = "TRUE"
     # device = "cuda" if torch.cuda.is_available() else "cpu"
     device="cpu"
@@ -238,12 +245,21 @@ def main(input_article, input_trigger):
 
     # feed the positive paragraphs to decoder
     paragraphs_needed = [data_test[idx] for idx in pos_ids]
+    paragraphs_needed = [par.split(" -- ")[0].replace("[ADD]", "") for par in paragraphs_needed]
     pd.DataFrame({"paragraph": paragraphs_needed}).to_csv("./util/experiments/paragraphs_needed.csv", index=False)
+    paragraphs_needed_str = "\n\n".join(paragraphs_needed)
+    # paragraphs_needed_str = paragraphs_needed_str.replace("Updated Paragraph:\n", "")
+    with open("./util/experiments/paragraphs_needed.txt", "w") as f:
+        f.write(paragraphs_needed_str)
 
     # updated_paragraphs = decode(input_paragraph, input_trigger)
-    config()
-    updated_paragraphs = [call_gpt(paragraph.split(" -- ")[0], input_trigger) for paragraph in paragraphs_needed]
     # updated_paragraphs = call_vicuna(paragraphs_needed, input_trigger)
+    config()
+    updated_paragraphs = [call_gpt(paragraph, input_trigger) for paragraph in paragraphs_needed]
+    updated_paragraphs_str = "\n\n".join(updated_paragraphs)
+    updated_paragraphs_str = updated_paragraphs_str.replace("Updated Paragraph:\n", "")
+    with open("./util/experiments/updated_paragraphs.txt", "w") as f:
+        f.write(updated_paragraphs_str)
 
     # merge updated paragraphs with non-updated paragraphs
     paragraphs_merged = data_test.copy()
@@ -266,11 +282,7 @@ def main(input_article, input_trigger):
 
     # combine the predictions and paragraphs into csv format file
     merged_par_pred_df = pd.DataFrame({"paragraphs": data_test, "predictions": predictions}).to_csv("./util/experiments/par_with_class.csv")
-    # return updated_article, modified, merged_par_pred_df
     modified_in_all = str(len(paragraphs_needed)) + " / " + str(len(data_test))
-    
-    os.remove("./util/experiments/classification.csv")
-    os.remove("./util/experiments/paragraphs_needed.csv")
     return updated_article, modified_in_all
 
 def copy_to_clipboard(t):
@@ -279,12 +291,12 @@ def copy_to_clipboard(t):
         pyperclip.copy(t)
 
 def compare_versions():
-    old, new = [], []
-    with open("./util/experiments/formatted_input.txt", "r") as f:
+    with open("./util/experiments/paragraphs_needed.txt", "r") as f:
         old = f.read()
         old = old.replace("[ADD]", "")
-    with open("./util/experiments/updated_article.txt", "r") as f:
+    with open("./util/experiments/updated_paragraphs.txt", "r") as f:
         new = f.read()
+        new = new.replace("[ADD]", "")
     return old, new
 
 with open("./examples/non_update.txt", "r") as f:
@@ -328,8 +340,8 @@ with gr.Blocks() as demo:
             fn=main,
             inputs=[input_1, input_2],
             outputs=[output_1, output_2],
-            cache_examples=True,
-            run_on_click=True,
+            # cache_examples=True,
+            # run_on_click=True,
                 ),
         com_1_value, com_2_value = "Pls finish article updating, then click the button above", "Pls finish article updating, then click the button above."
     with gr.Tab("Compare between versions"):
@@ -338,10 +350,6 @@ with gr.Blocks() as demo:
             com_1 = gr.Textbox(label="Non-update Article", value=com_1_value, lines=15)
             com_2 = gr.Textbox(label="Updated Article", value=com_2_value, lines=15)
         btn_com.click(fn=compare_versions, inputs=[], outputs=[com_1, com_2])
-        formatted_input_path = "./util/experiments/formatted_input.txt"
-        updated_article_path = "./util/experiments/updated_article.txt"
-        if os.path.isfile(formatted_input_path): os.remove(formatted_input_path)
-        if os.path.isfile(updated_article_path): os.remove(updated_article_path)
     gr.HTML("""
             <div align="center">
                 <p>
@@ -357,3 +365,4 @@ with gr.Blocks() as demo:
         )
 
 demo.launch()
+