Update app.py

app.py

@@ -13,12 +13,12 @@ untethered_model = GPT2LMHeadModel.from_pretrained('zmbfeng/untethered_20240225_
 untethered_paraphrased_model = GPT2LMHeadModel.from_pretrained('zmbfeng/untethered_20240227_epochs_350')
 
 def create_response_untethered_paraphrased(input_str, 
-                             num_beams, 
+                             # num_beams, 
                              num_return_sequences,
                              temperature,
                              repetition_penalty,
                              top_p,
-                             top_k,
+                             # top_k,
                              do_sample):
   print("input_str="+input_str)
   num_beams = int(num_beams)
@@ -62,12 +62,12 @@ def create_response_untethered_paraphrased(input_str,
 
                                  
 def create_response_untethered(input_str, 
-                             num_beams, 
+                             # num_beams, 
                              num_return_sequences,
                              temperature,
                              repetition_penalty,
                              top_p,
-                             top_k,
+                             # top_k,
                              do_sample):
   print("input_str="+input_str)
   num_beams = int(num_beams)
@@ -110,12 +110,12 @@ def create_response_untethered(input_str,
   return outputs
                                  
 def create_response_original(input_str, 
-                             num_beams, 
+                             # num_beams, 
                              num_return_sequences,
                              temperature,
                              repetition_penalty,
                              top_p,
-                             top_k,
+                             # top_k,
                              do_sample):
   print("input_str="+input_str)
   num_beams = int(num_beams)
@@ -171,15 +171,15 @@ interface1 = gr.Interface(fn=create_response_original,
     title="original",
     description="original language model, no fine tuning",
     examples=[ 
-      ["What is death?",7,5,0.2,1.5,0.9,50,True], # The first example
-      ["One of the best teachers in all of life turns out to be what?",7,5,0.2,1.5,0.9,50,True], # The second example
-      ["what is your most meaningful relationship?",7,5,0.2,1.5,0.9,50,True], # The third example
-      ["What actually gives life meaning?",7,5,0.2,1.5,0.9,50,True]
+      ["What is death?",5,0.2,1.5,0.9,True], # The first example
+      ["One of the best teachers in all of life turns out to be what?",5,0.2,1.5,0.9,True], # The second example
+      ["what is your most meaningful relationship?",5,0.2,1.5,0.9,True], # The third example
+      ["What actually gives life meaning?",5,0.2,1.5,0.9,True]
     ],
     inputs=[
     gr.Textbox(label="input text here", lines=3),
-    gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
-                "likely ones (specified in num_beams)", value=7),
+    # gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
+    #             "likely ones (specified in num_beams)", value=7),
     gr.Number(label="num_return_sequences (integer)  the number of outputs selected from num_beams possible output",
                   value=5),
         gr.Number(
@@ -194,11 +194,11 @@ interface1 = gr.Interface(fn=create_response_original,
         gr.Number(label="top_p (decimal) the model will only consider the words that have a high enough probability" +
                         " to reach a certain threshold",
                   value=0.9),
-        gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
-                        "This means that only the tokens with the highest probabilities are considered for sampling" +
-                        "This reduces the diversity of the generated sequences, "+
-                        "but also makes them more likely to be coherent and fluent.",
-                  value=50),
+        # gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
+        #                 "This means that only the tokens with the highest probabilities are considered for sampling" +
+        #                 "This reduces the diversity of the generated sequences, "+
+        #                 "but also makes them more likely to be coherent and fluent.",
+        #           value=50),
         gr.Checkbox(label="do_sample. If is set to False, num_return_sequences must be 1 because the generate function will use greedy decoding, " +
                           "which means that it will select the word with the highest probability at each step. " +
                           "This results in a deterministic and fluent output, but it might also lack diversity and creativity" +
@@ -212,15 +212,15 @@ interface2 = gr.Interface(fn=create_response_untethered,
     title="untethered",
     description="untethered fine tuning",
     examples=[ 
-      ["What is death?",7,5,0.2,1.5,0.9,50,True], # The first example
-      ["One of the best teachers in all of life turns out to be what?",7,5,0.2,1.5,0.9,50,True], # The second example
-      ["what is your most meaningful relationship?",7,5,0.2,1.5,0.9,50,True], # The third example
-      ["What actually gives life meaning?",7,5,0.2,1.5,0.9,50,True]
+      ["What is death?",5,0.2,1.5,0.9,True], # The first example
+      ["One of the best teachers in all of life turns out to be what?",5,0.2,1.5,0.9,True], # The second example
+      ["what is your most meaningful relationship?",5,0.2,1.5,0.9,True], # The third example
+      ["What actually gives life meaning?",5,0.2,1.5,0.9,True]
     ],
     inputs=[
     gr.Textbox(label="input text here", lines=3),
-    gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
-                "likely ones (specified in num_beams)", value=7),
+    # gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
+    #             "likely ones (specified in num_beams)", value=7),
     gr.Number(label="num_return_sequences (integer)  the number of outputs selected from num_beams possible output",
                   value=5),
         gr.Number(
@@ -235,11 +235,11 @@ interface2 = gr.Interface(fn=create_response_untethered,
         gr.Number(label="top_p (decimal) the model will only consider the words that have a high enough probability" +
                         " to reach a certain threshold",
                   value=0.9),
-        gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
-                        "This means that only the tokens with the highest probabilities are considered for sampling" +
-                        "This reduces the diversity of the generated sequences, "+
-                        "but also makes them more likely to be coherent and fluent.",
-                  value=50),
+        # gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
+        #                 "This means that only the tokens with the highest probabilities are considered for sampling" +
+        #                 "This reduces the diversity of the generated sequences, "+
+        #                 "but also makes them more likely to be coherent and fluent.",
+        #           value=50),
         gr.Checkbox(label="do_sample. If is set to False, num_return_sequences must be 1 because the generate function will use greedy decoding, " +
                           "which means that it will select the word with the highest probability at each step. " +
                           "This results in a deterministic and fluent output, but it might also lack diversity and creativity" +
@@ -252,15 +252,15 @@ interface3 = gr.Interface(fn=create_response_untethered_paraphrased,
     title="untethered paraphrased",
     description="untethered paraphrased fine tuning",
     examples=[ 
-      ["What is death?",7,5,0.2,1.5,0.9,50,True], # The first example
-      ["One of the best teachers in all of life turns out to be what?",7,5,0.2,1.5,0.9,50,True], # The second example
-      ["what is your most meaningful relationship?",7,5,0.2,1.5,0.9,50,True], # The third example
-      ["What actually gives life meaning?",7,5,0.2,1.5,0.9,50,True]
+      ["What is death?",5,0.2,1.5,0.9,True], # The first example
+      ["One of the best teachers in all of life turns out to be what?",5,0.2,1.5,0.9,True], # The second example
+      ["what is your most meaningful relationship?",5,0.2,1.5,0.9,True], # The third example
+      ["What actually gives life meaning?",5,0.2,1.5,0.9,True]
     ],
     inputs=[
     gr.Textbox(label="input text here", lines=3),
-    gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
-                "likely ones (specified in num_beams)", value=7),
+    # gr.Number(label="num_beams (integer) explores the specified number of possible outputs and selects the most " +
+    #             "likely ones (specified in num_beams)", value=7),
     gr.Number(label="num_return_sequences (integer)  the number of outputs selected from num_beams possible output",
                   value=5),
         gr.Number(
@@ -275,11 +275,11 @@ interface3 = gr.Interface(fn=create_response_untethered_paraphrased,
         gr.Number(label="top_p (decimal) the model will only consider the words that have a high enough probability" +
                         " to reach a certain threshold",
                   value=0.9),
-        gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
-                        "This means that only the tokens with the highest probabilities are considered for sampling" +
-                        "This reduces the diversity of the generated sequences, "+
-                        "but also makes them more likely to be coherent and fluent.",
-                  value=50),
+        # gr.Number(label="top_k (integer) The number of highest probability vocabulary word will be considered" +
+        #                 "This means that only the tokens with the highest probabilities are considered for sampling" +
+        #                 "This reduces the diversity of the generated sequences, "+
+        #                 "but also makes them more likely to be coherent and fluent.",
+        #           value=50),
         gr.Checkbox(label="do_sample. If is set to False, num_return_sequences must be 1 because the generate function will use greedy decoding, " +
                           "which means that it will select the word with the highest probability at each step. " +
                           "This results in a deterministic and fluent output, but it might also lack diversity and creativity" +