Update app.py

app.py

@@ -431,6 +431,7 @@ def get_beam_search_html(
         return_dict_in_generate=True,
         length_penalty=length_penalty,
         output_scores=True,
+        output_logits=True,
         do_sample=False,
     )
     markdown = "The conclusive sequences are the ones that end in an `<|endoftext|>` token or at the end of generation."
@@ -440,8 +441,10 @@ def get_beam_search_html(
     markdown += "\n#### <span style='color:var(--secondary-500)!important'>Output sequences:</span>"
     # Sequences are padded anyway so you can batch decode them
     decoded_sequences = tokenizer.batch_decode(outputs.sequences)
+
+    sequence_scores = (outputs.sequences_scores is n_beams > 1 else outputs.scores)
     for i, sequence in enumerate(decoded_sequences):
-        markdown += f"\n- Score `{outputs.sequences_scores[i]:.2f}`: `{clean(sequence.replace('<s> ', ''))}`"
+        markdown += f"\n- Score `{sequence_scores[i]:.2f}`: `{clean(sequence.replace('<s> ', ''))}`"
 
     if number_beams > 1:
         original_tree = generate_beams(