Update app.py

app.py

@@ -1,45 +1,39 @@
 import streamlit as st
-import transformers
-import gradio as gr
-import git
-	
-	#Load arabert preprocessor
-import git
-git.Git("arabert").clone("https://github.com/aub-mind/arabert")
-from arabert.preprocess import ArabertPreprocessor
-arabert_prep = ArabertPreprocessor(model_name="bert-base-arabert", keep_emojis=False)
-	
-	
-	#Load Model
-from transformers import EncoderDecoderModel, AutoTokenizer
-tokenizer = AutoTokenizer.from_pretrained("tareknaous/bert2bert-empathetic-response-msa")
-model = EncoderDecoderModel.from_pretrained("tareknaous/bert2bert-empathetic-response-msa")
-model.eval()
-	
-def generate_response(text, minimum_length, k, p, temperature):
-    text_clean = arabert_prep.preprocess(text)
-    inputs = tokenizer.encode_plus(text_clean,return_tensors='pt')
-    outputs = model.generate(input_ids = inputs.input_ids,
-                     attention_mask = inputs.attention_mask,
-                 	 do_sample = True,
-	                 min_length=minimum_length,
-	                 top_k = k,
-	                 top_p = p,
-	                temperature = temperature)
-    preds = tokenizer.batch_decode(outputs) 
-    response = str(preds)
-    response = response.replace("\'", '')
-    response = response.replace("[[CLS]", '')
-    response = response.replace("[SEP]]", '')
-    response = str(arabert_prep.desegment(response))
-    return response
-	
-gr.Interface(fn=generate_response,
-	              inputs=[
-	          gr.inputs.Textbox(),
-	          gr.inputs.Slider(5, 20, step=1, label='Minimum Output Length'),	       
-	          gr.inputs.Slider(0, 1000, step=10, label='Top-K'),
-	          gr.inputs.Slider(0, 1, step=0.1, label='Top-P'),
-	          gr.inputs.Slider(0, 3, step=0.1, label='Temperature'),
-	          ],
-	             outputs="text").launch()
+from streamlit_chat import message as st_message
+from transformers import BlenderbotTokenizer
+from transformers import BlenderbotForConditionalGeneration
+
+
+@st.experimental_singleton
+def get_models():
+    # it may be necessary for other frameworks to cache the model
+    # seems pytorch keeps an internal state of the conversation
+    model_name = "facebook/blenderbot-400M-distill"
+    tokenizer = BlenderbotTokenizer.from_pretrained(model_name)
+    model = BlenderbotForConditionalGeneration.from_pretrained(model_name)
+    return tokenizer, model
+
+
+if "history" not in st.session_state:
+    st.session_state.history = []
+
+st.title("Hello Chatbot")
+
+
+def generate_answer():
+    tokenizer, model = get_models()
+    user_message = st.session_state.input_text
+    inputs = tokenizer(st.session_state.input_text, return_tensors="pt")
+    result = model.generate(**inputs)
+    message_bot = tokenizer.decode(
+        result[0], skip_special_tokens=True
+    )  # .replace("<s>", "").replace("</s>", "")
+
+    st.session_state.history.append({"message": user_message, "is_user": True})
+    st.session_state.history.append({"message": message_bot, "is_user": False})
+
+
+st.text_input("Talk to the bot", key="input_text", on_change=generate_answer)
+
+for chat in st.session_state.history:
+    st_message(**chat)  # unpacking