Update app.py

app.py

@@ -1,14 +1,3 @@
-import subprocess
-
-# Install flash attention
-subprocess.run(
-    "pip install flash-attn --no-build-isolation",
-    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
-    shell=True,
-    check=True  # This will raise an exception if the command fails
-)
-
-# Rest of your app.py code
 import gradio as gr
 import asyncio
 import os
@@ -16,7 +5,6 @@ import thinkingframes
 import soundfile as sf
 import numpy as np
 import logging
-from transformers import pipeline
 from dotenv import load_dotenv
 from policy import user_acceptance_policy
 from styles import theme
@@ -39,21 +27,21 @@ engines = {default_lang: Model(default_lang)}
 # For maintaining user session (to keep track of userID)
 user_state = gr.State(value="")
 
-@spaces.GPU(duration=120)
-def transcribe(audio):
-    lang = "en"
-    model = engines[lang]
-    text = model.stt_file(audio)[0]
-    return text
-
 # Load the Meta-Llama-3-8B model from Hugging Face
-llm = pipeline("text-generation", model="models/meta-llama/Meta-Llama-3-8B")
+llm = gr.load("meta-llama/Meta-Llama-3-8B", src="models")
 
 image_path = "picturePerformance.jpg"
 img_html = get_image_html(image_path)
 
 executor = ThreadPoolExecutor()
 
+@spaces.GPU(duration=120)
+def transcribe(audio):
+    lang = "en"
+    model = engines[lang]
+    text = model.stt_file(audio)[0]
+    return text
+
 @spaces.GPU(duration=120)
 def generate_feedback(user_id, question_choice, strategy_choice, message, feedback_level):
     current_question_index = questions.index(question_choice)
@@ -67,7 +55,8 @@ def generate_feedback(user_id, question_choice, strategy_choice, message, feedba
         "content": message
     }]
 
-    feedback = llm(conversation, max_length=1000, num_return_sequences=1)[0]["generated_text"]
+    # Use the loaded model for generating feedback
+    feedback = llm(conversation)[0]["generated_text"]
 
     questionNo = current_question_index + 1
     add_submission(user_id, message, feedback, int(0), "", questionNo)
@@ -86,13 +75,13 @@ async def predict(question_choice, strategy_choice, feedback_level, audio):
     current_audio_output = None
 
     if audio is None:
-        yield [("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "No audio data received. Please try again.")], current_audio_output
+        yield [("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "No audio data received. Please try again.")], current_audio_output
         return
 
     sample_rate, audio_data = audio
 
     if audio_data is None or len(audio_data) == 0:
-        yield [("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "No audio data received. Please try again.")], current_audio_output
+        yield [("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "No audio data received. Please try again.")], current_audio_output
         return
 
     audio_path = "audio.wav"
@@ -100,7 +89,7 @@ async def predict(question_choice, strategy_choice, feedback_level, audio):
         raise ValueError("audio_data must be a numpy array")
     sf.write(audio_path, audio_data, sample_rate)
 
-    chat_history = [("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "Transcribing your audio, please listen to your oral response while waiting ...")]
+    chat_history = [("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "Transcribing your audio, please listen to your oral response while waiting ...")]
     yield chat_history, current_audio_output
 
     try:
@@ -108,19 +97,19 @@ async def predict(question_choice, strategy_choice, feedback_level, audio):
         student_response = await asyncio.wrap_future(transcription_future)
 
         if not student_response.strip():
-            yield [("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "Transcription failed. Please try again or seek assistance.")], current_audio_output
+            yield [("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "Transcription failed. Please try again or seek assistance.")], current_audio_output
             return
 
         chat_history.append(("Student", student_response))
         yield chat_history, current_audio_output
 
-        chat_history.append(("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "Transcription complete. Generating feedback. Please continue listening to your oral response while waiting ..."))
+        chat_history.append(("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "Transcription complete. Generating feedback. Please continue listening to your oral response while waiting ..."))
         yield chat_history, current_audio_output
 
         feedback_future = executor.submit(generate_feedback, int(user_state.value), question_choice, strategy_choice, student_response, feedback_level)
         feedback = await asyncio.wrap_future(feedback_future)
 
-        chat_history.append(("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", feedback))
+        chat_history.append(("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", feedback))
         yield chat_history, current_audio_output
 
         audio_future = executor.submit(generate_audio_feedback, feedback)
@@ -131,7 +120,7 @@ async def predict(question_choice, strategy_choice, feedback_level, audio):
 
     except Exception as e:
         logging.error(f"An error occurred: {str(e)}", exc_info=True)
-        yield [("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "An error occurred. Please try again or seek assistance.")], current_audio_output
+        yield [("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", "An error occurred. Please try again or seek assistance.")], current_audio_output
 
 # Modify the toggle_oral_coach_visibility function to call add_user_privacy and store the returned user_id in user_state.value
 def toggle_oral_coach_visibility(class_name, index_no, policy_checked):