Initial commit

app.py

@@ -1,14 +1,48 @@
-import os
 import gradio as gr
 import requests
 from typing import List, Dict, Tuple
+from flask import Flask, request, jsonify
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import threading
 
-# Define the API URL (adjust according to your server address)
-API_URL = "http://127.0.0.1:5000/chat"
+# Define the API URL to use the internal server
+API_URL = "http://localhost:5000/chat"
 
 History = List[Tuple[str, str]]
 Messages = List[Dict[str, str]]
 
+app = Flask(__name__)
+
+# Load the model and tokenizer
+model_name = "path/to/your/dictalm2.0"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(model_name)
+
+@app.route('/chat', methods=['POST'])
+def chat():
+    data = request.json
+    messages = data.get('messages', [])
+    
+    if not messages:
+        return jsonify({"response": "No messages provided"}), 400
+    
+    # Concatenate all user inputs into a single string
+    user_input = " ".join([msg['content'] for msg in messages if msg['role'] == 'user'])
+    
+    inputs = tokenizer.encode(user_input, return_tensors='pt')
+    outputs = model.generate(inputs)
+    response_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    return jsonify({"response": response_text})
+
+# Function to run the Flask app
+def run_flask():
+    app.run(host='0.0.0.0', port=5000)
+
+# Start the Flask app in a separate thread
+threading.Thread(target=run_flask).start()
+
+# Gradio interface functions
 def clear_session() -> History:
     return []
 
@@ -32,16 +66,28 @@ def model_chat(query: str, history: History) -> Tuple[str, History]:
     messages = history_to_messages(history)
     messages.append({'role': 'user', 'content': query.strip()})
 
-    response = requests.post(API_URL, json={"messages": messages})
-    if response.status_code != 200:
-        return "Error: Failed to get response from the API", history
+    try:
+        response = requests.post(API_URL, json={"messages": messages})
+        response.raise_for_status()  # This will raise an HTTPError if the HTTP request returned an unsuccessful status code
+        response_json = response.json()
+        response_text = response_json.get("response", "Error: Response format is incorrect")
+    except requests.exceptions.HTTPError as e:
+        response_text = f"HTTPError: {str(e)}"
+        print(f"HTTPError: {e.response.text}")  # Detailed error message
+    except requests.exceptions.RequestException as e:
+        response_text = f"RequestException: {str(e)}"
+        print(f"RequestException: {e}")  # Debug print statement
+    except ValueError as e:
+        response_text = "ValueError: Invalid JSON response"
+        print(f"ValueError: {e}")  # Debug print statement
+    except Exception as e:
+        response_text = f"Exception: {str(e)}"
+        print(f"General Exception: {e}")  # Debug print statement
     
-    response_json = response.json()
-    response_text = response_json["response"]
-
     history.append((query.strip(), response_text.strip()))
     return response_text.strip(), history
 
+# Gradio interface setup
 with gr.Blocks(css='''
     .gr-group {direction: rtl;}
     .chatbot{text-align:right;}
@@ -90,7 +136,7 @@ with gr.Blocks(css='''
   </a>  
   <div class="dicta-intro-text">
     <h1>צ'אט מערכי - הדגמה ראשונית</h1>
-    <span dir='rtl'>ברוכים הבאים לדמו האינטראקטיבי הראשון. חקרו את יכולות המודל וראו כיצד הוא יכול לסייע לכן במשימותיכם</span><br/>
+    <span dir='rtl'>ברוכים הבאים לדמו האינטראקטיבי הראשון. חקרו את יכולות המודל וראו כיצד הוא יכול לסייע לכם במשימותיכם</span><br/>
     <span dir='rtl'>הדמו נכתב על ידי סרן רועי רתם תוך שימוש במודל שפה דיקטה שפותח על ידי מפא"ת</span><br/>
   </div>
 </div>
@@ -101,7 +147,9 @@ with gr.Blocks(css='''
     clear_btn = gr.Button("נקה שיחה")
 
     def respond(query, history):
+        print(f"Query: {query}")  # Debug print statement
         response, history = model_chat(query, history)
+        print(f"Response: {response}")  # Debug print statement
         return history, gr.update(value="", interactive=True)
 
     demo_state = gr.State([])
@@ -109,4 +157,4 @@ with gr.Blocks(css='''
     query.submit(respond, [query, demo_state], [chatbot, query, demo_state])
     clear_btn.click(clear_session, [], demo_state, chatbot)
 
-demo.launch(share=True)
+demo.launch()