Upload app.py

app.py

@@ -18,11 +18,11 @@ base_model = AutoModelForCausalLM.from_pretrained(
 )
 
 # Load fine-tuned LoRA weights
-fine_tuned_model_path = "./phi2-qlora-adapter"
+fine_tuned_model_path = "piyushgrover/phi2-qlora-adapter-s18erav3"
 model = PeftModel.from_pretrained(base_model, fine_tuned_model_path)
 model = model.merge_and_unload()  # Merge LoRA weights
 
-# Load tokenizer
+# ✅ Load tokenizer
 tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
 tokenizer.pad_token = tokenizer.eos_token
 tokenizer.padding_side = "right"
@@ -31,23 +31,35 @@ tokenizer.padding_side = "right"
 generator = pipeline("text-generation", model=model, tokenizer=tokenizer, max_length=500)
 
 
-# ✅ Chatbot Function
+# ✅ Chatbot Function with Conversation History
 def chat(user_input, history=[]):
-    """Generates a response from the fine-tuned Phi-2 model."""
-    prompt = f"\n\n### User:\n{user_input}\n\n### Assistant:\n"
+    """Generates a response from the fine-tuned Phi-2 model with conversation memory."""
+
+    # Format the chat history properly
+    formatted_history = ""
+    for usr, bot in history:
+        formatted_history += f"\n\n### User:\n{usr}\n\n### Assistant:\n{bot}"
+
+    # Append the latest user message
+    prompt = f"{formatted_history}\n\n### User:\n{user_input}\n\n### Assistant:\n"
+
+    # Generate response
     response = generator(prompt, max_length=500, do_sample=True)
+
+    # Extract only the model's generated response
     answer = response[0]["generated_text"].split("### Assistant:\n")[-1].strip()
 
-    # Append conversation history
+    # Update conversation history
     history.append((user_input, answer))
-    return "", history
+
+    return "", history  # Return empty input and updated history
 
 
 # ✅ Create Gradio Chat Interface
 chatbot = gr.ChatInterface(
     fn=chat,
-    title="Fine-Tuned Phi-2 Chat Assistant",
-    description="🚀 Chat with a fine-tuned Phi-2 model. Ask anything!",
+    title="Fine-Tuned Phi-2 Conversational Chat Assistant",
+    description="🚀 Chat with a fine-tuned Phi-2 model. It remembers the conversation!",
     theme="compact",
 )