add beam search

app_dialogue.py

@@ -286,12 +286,22 @@ with gr.Blocks(title="IDEFICS", theme=gr.themes.Base()) as demo:
                 decoding_strategy = gr.Radio(
                     [
                         "greedy",
+                        "beam_search",
+                        "beam_sampling",
                         "sampling_top_k",
                         "sampling_top_p",
                     ],
                     value="greedy",
                     label="Decoding strategy",
                 )
+                num_beams = top_k = gr.Slider(
+                    minimum=0,
+                    maximum=20,
+                    value=3.0,
+                    step=1.0,
+                    interactive=True,
+                    label="Number of beams",
+                )
                 temperature = gr.Slider(
                     minimum=0.0,
                     maximum=1.0,
@@ -436,6 +446,7 @@ with gr.Blocks(title="IDEFICS", theme=gr.themes.Base()) as demo:
         user_prompt,
         chat_history,
         decoding_strategy,
+        num_beams,
         temperature,
         no_repeat_ngram_size,
         max_new_tokens,
@@ -455,7 +466,7 @@ with gr.Blocks(title="IDEFICS", theme=gr.themes.Base()) as demo:
         # repetition_penalty = 1.0
         hide_special_tokens = False
         # decoding_strategy = "greedy"
-        num_beams = 3
+        # num_beams = 3
         # length_penalty = 1.0
         # top_k = 50
         # top_p = 0.95
@@ -497,6 +508,7 @@ with gr.Blocks(title="IDEFICS", theme=gr.themes.Base()) as demo:
             textbox,
             chatbot,
             decoding_strategy,
+            num_beams,
             temperature,
             no_repeat_ngram_size,
             max_new_tokens,
@@ -515,6 +527,7 @@ with gr.Blocks(title="IDEFICS", theme=gr.themes.Base()) as demo:
             textbox,
             chatbot,
             decoding_strategy,
+            num_beams,
             temperature,
             no_repeat_ngram_size,
             max_new_tokens,