Update app.py

app.py

@@ -1,37 +1,27 @@
-import os
-os.system('sh setup.sh')
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-# Initialize the Hugging Face client
-client = InferenceClient(model="redael/model_udc")
-
-def generate_response(message, history, system_message, max_tokens, temperature, top_p):
-    messages = [{"role": "system", "content": system_message}]
-
-    for user_message, bot_message in history:
-        if user_message:
-            messages.append({"role": "user", "content": user_message})
-        if bot_message:
-            messages.append({"role": "assistant", "content": bot_message})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-    for token_message in client.chat_completion(
-        messages=messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = token_message.choices[0].delta.content
-        response += token
-        yield response
+from transformers import pipeline, set_seed
+
+# Initialize the Hugging Face pipeline
+generator = pipeline('text-generation', model='redael/model_udc')
+
+def generate_response(prompt, max_length=100, num_beams=5, temperature=0.5, top_p=0.9, repetition_penalty=4.0):
+    # Prepare the prompt
+    prompt = f"User: {prompt}\nAssistant:"
+    responses = generator(prompt, max_length=max_length, num_return_sequences=1, num_beams=num_beams, temperature=temperature, top_p=top_p, repetition_penalty=repetition_penalty)
+    response = responses[0]['generated_text']
+    
+    # Post-processing to clean up the response
+    response = response.split("Assistant:")[-1].strip()
+    response_lines = response.split('\n')
+    clean_response = []
+    for line in response_lines:
+        if "User:" not in line and "Assistant:" not in line:
+            clean_response.append(line)
+    response = ' '.join(clean_response)
+    return response.strip()
 
 def chat_interface(user_input, history, system_message, max_tokens, temperature, top_p):
-    response_generator = generate_response(user_input, history, system_message, max_tokens, temperature, top_p)
-    response = "".join([token for token in response_generator])
+    response = generate_response(user_input, max_length=max_tokens, temperature=temperature, top_p=top_p)
     history.append((user_input, response))
     return history, history