Update app.py

app.py

@@ -1,42 +1,55 @@
 import tensorflow as tf
 from tensorflow import keras
+from tensorflow.keras.preprocessing.text import Tokenizer
+from tensorflow.keras.preprocessing.sequence import pad_sequences
+
 import gradio as gr
 from gradio import mix
 import numpy as np
 import torch
 from keras.utils.data_utils import pad_sequences
-import pickle
+
 
 from huggingface_hub import from_pretrained_keras
 
 model = from_pretrained_keras("keras-io/text-generation-miniature-gpt")
 
-with open('tokenizer.pickle', 'rb') as handle:
-    tokenizer = pickle.load(handle)
-    
-#def tokenize_data(text):
-# Tokenize the review body
-#    input_ =  str(text) + ' </s>'
-#    max_len = 80
-    # tokenize inputs
-#    tokenized_inputs = tokenizer(input_, padding='max_length', truncation=True, max_length=max_len, return_attention_mask=True, return_tensors='pt')
-
- #   inputs={"input_ids": tokenized_inputs['input_ids'],
-  #      "attention_mask": tokenized_inputs['attention_mask']}
-   # return inputs
-
-def generate_answers(text):
-    sequence_test = tokenizer.texts_to_sequences([text])
-    padded_test = pad_sequences(sequence_test, maxlen= 80, padding='post')
-    predictions,_ = model.predict(padded_test)
-    results = np.argmax(predictions, axis=1)[0]
-    answer = tokenizer.sequences_to_texts([results] )
-    answertoString = ' '.join([str(elem) for elem in answer])
-    return answertoString
+def text_process_pipeline(text): #pipeline
+  tokenizer = Tokenizer(num_words=80, split=' ')
+  tokenizer.fit_on_texts(word_to_index.values())
+  processed_text = tokenizer.texts_to_sequences(start_prompt)
+  processed_text = pad_sequences(processed_text, maxlen=80, padding='post')
+  return processed_text
+  
+def sample_from(logits):
+  l, i = tf.math.top_k(logits, k=10, sorted=True)
+  indices = np.asarray(i).astype("int32")
+  preds = keras.activations.softmax(tf.expand_dims(l, 0))[0]
+  preds = np.asarray(preds).astype("float32")
+  return np.random.choice(i, p=preds)
+
+def generate_answers(start_prompt):
+  num_tokens_generated = 0 
+  sample_index = len(start_prompt) - 1
+  start_tokens = [word_to_index.get(_, 1) for _ in start_prompt]
+  tokens_generated= []
+
+  text_out = text_process_pipeline(start_prompt)
+  predictions,_ = mode.predict(text_out)
+  results = np.argmax(predictions, axis=1)[0]
+
+  while num_tokens_generated <= 40:
+    sample_token = sample_from(predictions[0][sample_index])
+    tokens_generated.append(sample_token)
+    start_tokens.append(sample_token)
+    num_tokens_generated = len(tokens_generated)
+
+  text_out = tokenizer.sequences_to_texts([tokens_generated])  
+  return text_out[0]
     
 examples = [["The movie was nice, "], ["It was showing nothing special to "]]
 title = "Text Generation with Miniature GPT"
 description = "Gradio Demo for a miniature with GPT. To use it, simply add your text, or click one of the examples to load them. Read more at the links below."
 
 iface = gr.Interface(fn=generate_answers, title = title, description=description, inputs=['text'], outputs=["text"], examples=examples)
-iface.launch()
+iface.launch(debug=True)