Update OTF_ComplexControl.py

OTF_ComplexControl.py

@@ -4,11 +4,12 @@ import json
 from transformers import BitsAndBytesConfig, AutoModelForCausalLM, AutoTokenizer
 import sys
 sys.path.append(f'../source')
-import models
-import helpers
+import cefr_utils
 import torch
 from huggingface_hub import snapshot_download
 from huggingface_hub import login
+import gradio as gr
+import os
 
 parser = argparse.ArgumentParser(description="Generate responses on a CEFR level")
 parser.add_argument("--n", type=int, default=10, help="Number of dialog contexts. Default: %(default)s")
@@ -77,7 +78,7 @@ def detect_cefr_level(text: str) -> str:
 
 """Response Generation Script"""
 def get_response(prompt):
-    response_list = models.generate(llm_model, llm_tokenizer, [prompt])
+    response_list = cefr_utils.generate(llm_model, llm_tokenizer, [prompt])
     response_str = "".join(response_list) if isinstance(response_list, list) else str(response_list)
     return parse_response(response_str)
 
@@ -101,23 +102,22 @@ llm_model = AutoModelForCausalLM.from_pretrained(
 responses = []
 conversation_history = []
 MAX_TURNS = 5  # Limit the number of turns to keep context manageable
-while True:
-    # 1) Get user input
-    user_input = input("User: ").strip()
-    if user_input.lower() in ["quit", "exit"]:
-        break
+def chat(user_input):
+    global conversation_history
+
+    if not user_input.strip():
+        return conversation_history, "Please enter a message."
 
     # 2) Detect CEFR from input context
     detected_level = detect_cefr_level(user_input)
     print(f"[DEBUG] Detected CEFR = {detected_level} for context: {user_input}")
 
     # 3) Build prompt using detected CEFR
-    
     conversation_history.append({"role": "user", "text": user_input, "CEFR": detected_level})
     recent_turns = conversation_history[-MAX_TURNS*2:]  # *2 because each turn has user+model
 
     item = {"context": recent_turns, "CEFR": detected_level, "response": ""}
-    item = helpers.get_CEFR_prompt(item, apply_chat_template=llm_tokenizer.apply_chat_template)
+    item = cefr_utils.get_CEFR_prompt(item, apply_chat_template=llm_tokenizer.apply_chat_template)
     print(f"[DEBUG] Prompt for response generation: {item['prompt']}")
 
     # 4) Generate response
@@ -125,4 +125,20 @@ while True:
     print(f"[{detected_level}] {response}")
 
     # 5) Update conversation history
-    conversation_history.append({"role": "model", "text": response, "CEFR": detected_level})
+    conversation_history.append({"role": "model", "text": response, "CEFR": detected_level})
+
+    gradio_history = []
+    for turn in conversation_history:
+        if turn["role"] == "user":
+            gradio_history.append((turn['text'], None))
+        else:
+            gradio_history[-1] = (gradio_history[-1][0], turn["text"])
+            
+        return gradio_history, ""
+
+with gr.Blocks() as demo:
+    chatbot = gr.Chatbot(label="Adaptive CEFR chatbot")
+    msg = gr.Textbox(placeholder="Type your message here...")
+    msg.submit(chat, inputs=msg, outputs=[chatbot, msg], clear_on_submit=True)
+
+demo.launch(server_name="0.0.0.0", server_port=7860, share=False, ssr_mode=False)