Upload 2 files

app.py

@@ -3,30 +3,44 @@ import tensorflow as tf
 import time
 import warnings
 import os
-from transformers import AutoTokenizer, TFAutoModelForSequenceClassification, AutoModelForCausalLM, \
-    TFAutoModelForQuestionAnswering, pipeline, TFT5ForConditionalGeneration
+from transformers import (
+    AutoTokenizer,
+    AutoModelForSeq2SeqLM,
+    pipeline,
+    AutoModelForCausalLM,
+    AutoModelForSequenceClassification,
+)
 
 # Warning Suppression
-os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
-warnings.filterwarnings('ignore', category=DeprecationWarning)
-warnings.filterwarnings('ignore', category=FutureWarning)
-tf.get_logger().setLevel('ERROR')
+# os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
+# warnings.filterwarnings('ignore', category=DeprecationWarning)
+# warnings.filterwarnings('ignore', category=FutureWarning)
+# warnings.filterwarnings('ignore', category=UserWarning)
+# tf.get_logger().setLevel('ERROR')
 
-emotion_tokenizer = AutoTokenizer.from_pretrained("aegrif/CIS6930_DAAGR_Classification")
-emotion_model = TFAutoModelForSequenceClassification.from_pretrained("aegrif/CIS6930_DAAGR_Classification")
+# emotion classification
+emotion_tokenizer = "aegrif/CIS6930_DAAGR_Classification"
+emotion_model = "aegrif/CIS6930_DAAGR_Classification"
+emotion_pipeline = pipeline("text-classification", model=emotion_model, tokenizer=emotion_tokenizer)
 
 # generation models
-# encoder
-distilbert_model = TFAutoModelForQuestionAnswering.from_pretrained("aegrif/CIS6930_DAAGR_DistilBert")
-distilbert_tokenizer = AutoTokenizer.from_pretrained("aegrif/CIS6930_DAAGR_DistilBert")
+
+# no emotion
+gpt2_model_no_emo = "aegrif/CIS6930_DAAGR_GPT2_NoEmo"
+gpt2_tokenizer_no_emo = "aegrif/CIS6930_DAAGR_GPT2_NoEmo"
+chatbot_gpt_no_emo = pipeline(model=gpt2_model_no_emo, tokenizer=gpt2_tokenizer_no_emo, pad_token_id=50256)
+
 
 # decoder
-gpt2_model = AutoModelForCausalLM.from_pretrained("microsoft/DialoGPT-small")
-gpt2_tokenizer = AutoTokenizer.from_pretrained("microsoft/DialoGPT-small")
+gpt2_model_emo = "aegrif/CIS6930_DAAGR_GPT2_Emo"
+gpt2_tokenizer = "aegrif/CIS6930_DAAGR_GPT2_Emo"
+chatbot_gpt_emo = pipeline(model=gpt2_model_emo, tokenizer=gpt2_tokenizer, pad_token_id=50256)
+
 
 # encoder-decoder
-t5_model = TFT5ForConditionalGeneration.from_pretrained("t5-small")
-t5_tokenizer = AutoTokenizer.from_pretrained("t5-small")
+t5_model_emo = "aegrif/CIS6930_DAAGR_T5_Emo"
+t5_tokenizer = "t5-small"
+chatbot_t5_emo = pipeline(model=t5_model_emo, tokenizer=t5_tokenizer)
 
 emotion_dict = {'disappointed': 0, 'annoyed': 1, 'excited': 2, 'afraid': 3, 'disgusted': 4, 'grateful': 5,
                 'impressed': 6, 'prepared': 7}
@@ -34,23 +48,31 @@ inverted_emotion_dict = {v: k for k, v in emotion_dict.items()}
 
 
 def get_context(user_input):
-    new_user_input_ids = emotion_tokenizer.encode(user_input, return_tensors='tf')
-    output = emotion_model.predict(new_user_input_ids)[0]
-    prediction = tf.argmax(output, axis=1).numpy()[0]
-    context = inverted_emotion_dict.get(prediction)
+
+    output = emotion_pipeline(user_input)[0]['label']
+
+    context = inverted_emotion_dict.get(int(output[-1]))
 
     return context
 
 
-def predict_distilbert(user_input, history):
-    # Get the context from the user input
-    context = get_context(user_input)
+def predict_gpt2_no_emo(user_input, history):
+    user_input = "<|user|>" + user_input + " <|bot|>"
+
+    output = chatbot_gpt_no_emo(
+        user_input,
+        max_new_tokens=40,
+        num_return_sequences=1,
+        do_sample=True,
+        temperature=0.5,
+        renormalize_logits=True,
+        exponential_decay_length_penalty=(5, 1.1),
+        no_repeat_ngram_size=3,
+        repetition_penalty=1.5
+    )
 
-    # Generate a response using the DistilBert model
-    question_answerer = pipeline("question-answering", model=distilbert_model, tokenizer=distilbert_tokenizer)
-    output = question_answerer(question=user_input, context=context)
     # Decode the generated response
-    bot_response = output['answer']
+    bot_response = output[0]['generated_text'].split("<|bot|>")[1].strip()
 
     return bot_response
 
@@ -59,11 +81,23 @@ def predict_gpt2(user_input, history):
     # Get the context from the user input
     context = get_context(user_input)
 
+    user_input = f"<|context|>{context} <|user|>{user_input} <|bot|>"
     # Generate a response using the DialoGPT model
-    chatbot = pipeline("text-generation", model=gpt2_model, tokenizer=gpt2_tokenizer, pad_token_id=50256)
-    output = chatbot(user_input + context, max_length=100, num_return_sequences=1)
+
+    output = chatbot_gpt_emo(
+        user_input,
+        max_new_tokens=40,
+        num_return_sequences=1,
+        do_sample=True,
+        temperature=0.5,
+        renormalize_logits=True,
+        exponential_decay_length_penalty=(5, 1.1),
+        no_repeat_ngram_size=3,
+        repetition_penalty=1.5
+    )
+
     # Decode the generated response
-    bot_response = output[0]['generated_text']
+    bot_response = output[0]['generated_text'].split("<|bot|>")[1].strip()
 
     return bot_response
 
@@ -72,9 +106,19 @@ def predict_t5(user_input, history):
     # Get the context from the user input
     context = get_context(user_input)
 
+    user_input = f"question: {user_input} context: {context} </s>"
     # Generate a response using the T5 model
-    chatbot = pipeline("text2text-generation", model=t5_model, tokenizer=t5_tokenizer, pad_token_id=50256)
-    bot_response = chatbot(user_input + context, max_length=100, num_return_sequences=1)[0]['generated_text']
+    bot_response = chatbot_t5_emo(
+        user_input,
+        max_new_tokens=40,
+        num_return_sequences=1,
+        do_sample=True,
+        temperature=0.5,
+        renormalize_logits=True,
+        exponential_decay_length_penalty=(5, 1),
+        no_repeat_ngram_size=3,
+        repetition_penalty=1.5
+    )[0]['generated_text']
 
     return bot_response
 
@@ -83,14 +127,13 @@ def user(user_message, history):
     return "", history + [[user_message, None]]
 
 
-def distil_bot(history):
+def gpt2_bot_no_emo(history):
     user_message = history[-1][0]
-    bot_message = predict_distilbert(user_message, history)
+    bot_message = predict_gpt2_no_emo(user_message, history)
     history[-1][1] = bot_message
     time.sleep(1)
     return history
 
-
 def gpt2_bot(history):
     user_message = history[-1][0]
     bot_message = predict_gpt2(user_message, history)
@@ -120,7 +163,7 @@ with gr.Blocks() as demo:
             msg3 = gr.Textbox(show_label=False, placeholder="Enter text and press enter").style(container=False)
 
     msg1.submit(user, [msg1, chatbot1], [msg1, chatbot1], queue=False).then(
-        distil_bot, chatbot1, chatbot1
+        gpt2_bot_no_emo, chatbot1, chatbot1
     )
     msg2.submit(user, [msg2, chatbot2], [msg2, chatbot2], queue=False).then(
         gpt2_bot, chatbot2, chatbot2