Update app.py

app.py

@@ -324,15 +324,15 @@ def complete(sample, k, current_example):
     output = prefix
     for token in client.text_generation(prefix, do_sample=False, max_new_tokens=512, stream=True):
         if token == "<|endoftext|>":
-            bleu_score = {"BLEU": bleu.compute(predictions=[output],
+            bleu_score = {"Memorization score (BLEU)": bleu.compute(predictions=[output],
                                 references=[current_example])["bleu"]}
             return diff(output, current_example), gr.Label.update(value=bleu_score), current_example
         output += token
-        bleu_score = {"BLEU": bleu.compute(predictions=[output],
+        bleu_score = {"Memorization score (BLEU)": bleu.compute(predictions=[output],
                                 references=[current_example])["bleu"]}
         yield diff(output, current_example), gr.Label.update(value=bleu_score), current_example
         # yield output, diff_texts(output, sample), gr.Label.update(value=bleu_score)
-    bleu_score = {"BLEU": bleu.compute(predictions=[output],
+    bleu_score = {"Memorization score (BLEU)": bleu.compute(predictions=[output],
                                 references=[current_example])["bleu"]}
     # return output, diff_texts(output, sample), gr.Label.update(value=bleu_score)
     return diff(output, current_example), gr.Label.update(value=bleu_score), current_example
@@ -386,7 +386,7 @@ with gr.Blocks(theme=style) as demo:
                 )
                 
             with gr.Column():
-                label = gr.Label(value={"Memorization score (BLEU)": 0},label="BLEU")
+                label = gr.Label(value={"Memorization score (BLEU)": 0},label="Memorization")
                 with gr.Accordion("What is BLEU?", open=False): # NOTE - THIS WEIRDLY BREAKS EVERYTHING IF I UNCOMMENT
                     gr.Markdown("""[BLEU](https://huggingface.co/spaces/evaluate-metric/bleu) score is a metric that can be used to measure the similarity of two sentences.
                             Here, the higher the BLEU score, the more likely the model will learn the example by heart.