Update app.py

app.py

@@ -2,8 +2,8 @@ import gradio as gr
 from transformers import AutoTokenizer, AutoModel, GPT2LMHeadModel, GPT2Tokenizer
 import torch
 
-# Load the bi-encoder model and tokenizer
-bi_encoder_model_name = "sentence-transformers/all-MiniLM-L6-v2"
+# Load the NASA-specific bi-encoder model and tokenizer
+bi_encoder_model_name = "nasa-impact/nasa-smd-ibm-st-v2"
 bi_tokenizer = AutoTokenizer.from_pretrained(bi_encoder_model_name)
 bi_model = AutoModel.from_pretrained(bi_encoder_model_name)
 
@@ -19,8 +19,8 @@ def encode_text(text):
     return outputs.last_hidden_state.mean(dim=1).detach().numpy()
 
 def generate_response(user_input, context_embedding):
-    # Combine user input with context embedding for GPT-2 input
-    combined_input = user_input + " " + context_embedding
+    # Create a structured prompt for GPT-2
+    combined_input = f"Question: {user_input}\nContext: {context_embedding}\nAnswer:"
     
     # Generate a response using GPT-2 with adjusted parameters
     gpt2_inputs = gpt2_tokenizer.encode(combined_input, return_tensors='pt')
@@ -28,7 +28,7 @@ def generate_response(user_input, context_embedding):
         gpt2_inputs,
         max_length=150,
         num_return_sequences=1,
-        temperature=0.5,
+        temperature=0.7,
         top_p=0.9,
         repetition_penalty=1.2
     )
@@ -47,7 +47,7 @@ iface = gr.Interface(
     inputs=[gr.Textbox(lines=2, placeholder="Enter your message here..."), gr.Textbox(lines=2, placeholder="Enter context here (optional)...")],
     outputs="text",
     title="Context-Aware Dynamic Response Chatbot",
-    description="A chatbot using a bi-encoder model to understand the input context and GPT-2 to generate dynamic responses."
+    description="A chatbot using a NASA-specific bi-encoder model to understand the input context and GPT-2 to generate dynamic responses."
 )
 
 # Launch the interface
@@ -56,3 +56,4 @@ iface.launch()
 
 
 
+