Update app.py

app.py

@@ -33,119 +33,62 @@ import tempfile
 
 load_dotenv()
 
-# Load CSS from external file
-with open('styles.css', 'r') as file:
-    css = file.read()
+default_lang = "en"
+engines = {default_lang: Model(default_lang)}
 
 # For maintaining user session (to keep track of userID)
 user_state = gr.State(value="")
 
-load_dotenv()
+@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")
 
-client = OpenAI()
 image_path = "picturePerformance.jpg"
 img_html = get_image_html(image_path)
 
-# Create a thread pool executor
 executor = ThreadPoolExecutor()
 
-def transcribe_audio(audio_path):
-    with open(audio_path, "rb") as audio_file:
-        transcript = client.audio.transcriptions.create(file=audio_file, model="whisper-1", language="en")
-    return transcript.text
-
-async def generate_feedback(user_id, question_choice, strategy_choice, message, feedback_level):
-    current_question_index = thinkingframes.questions.index(question_choice)
-    strategy, explanation = thinkingframes.strategy_options[strategy_choice]
+@spaces.GPU(duration=120)
+def generate_feedback(user_id, question_choice, strategy_choice, message, feedback_level):
+    current_question_index = questions.index(question_choice)
+    strategy, explanation = strategy_options[strategy_choice]
 
     conversation = [{
         "role": "system",
-        "content": f"You are an expert Primary 6 English Language Teacher in a Singapore Primary school, "
-                   f"directly guiding a Primary 6 student in Singapore in their oral responses. "
-                   f"Format the feedback in Markdown so that it can be easily read. "
-                   f"Address the student directly in the second person in your feedback. "
-                   f"The student is answering the question: '{thinkingframes.questions[current_question_index]}'. "
-                   f"For Question 1, consider the picture description: '{thinkingframes.description}'. "
-                   f"For Questions 2 and 3, the picture is not relevant, so the student should not refer to it in their response. "
-                   f"Analyze the student's response using the following step-by-step approach: "
-                   f"1. Evaluate the response against the {strategy} thinking frame. "
-                   f"2. Assess how well the student's response addresses each component of the {strategy} thinking frame: "
-                   f"   - Assign emoticon scores based on how well the student comprehensively covered each component: "
-                   f"     - 😊😊😊 (three smiling faces) for a good coverage "
-                   f"     - 😊😊 (two smiling faces) for an average coverage "
-                   f"     - 😊 (one smiling face) for a poor coverage "
-                   f"   - Provide a clear, direct, and concise explanation of how well the answer addresses each component. "
-                   f"   - Identify specific areas for improvement in students responses, and provide targeted suggestions for improvement. "
-                   f"3. Identify overall strengths and areas for improvement in the student's response using the {strategy} to format and provide targeted areas for improvement. "
-                   f"4. Provide specific feedback on grammar, vocabulary, and sentence structure. "
-                   f"   Suggest age-appropriate enhancements that are one level higher than the student's current response. "
-                   f"5. Conclude with follow-up questions for reflection. "
-                   f"If the student's response deviates from the question, provide clear and concise feedback to help them refocus and try again. "
-                   f"Ensure that the vocabulary and sentence structure recommendations are achievable for Primary 6 students in Singapore. "
-                   f"Example Feedback Structure for Each Component: "
-                   f"Component: [Component Name] "
-                   f"Score: [Smiling emoticons] "
-                   f"Explanation: [Clear, direct, and concise explanation of how well the answer addresses the component. Identify specific areas for improvement, and provide targeted suggestions for improvement.] "
-                   f"{thinkingframes.generate_prompt(feedback_level)}"
+        "content": thinkingframes.generate_system_message(current_question_index, feedback_level)
     }, {
         "role": "user",
         "content": message
     }]
 
-    response = client.chat.completions.create(
-        model='gpt-4o-2024-05-13',
-        messages=conversation,
-        temperature=0.6,
-        max_tokens=1000,
-        stream=True
-    )
-
-    chat_history = []  # Initialize chat history outside the loop
-    full_feedback = ""  # Accumulate the entire feedback message
-    try:
-        for chunk in response:
-            if chunk.choices[0].delta and chunk.choices[0].delta.content:
-                feedback_chunk = chunk.choices[0].delta.content
-                full_feedback += feedback_chunk  # Accumulate the feedback
-                await asyncio.sleep(0)
-
-        # Append the complete feedback to the chat history
-        chat_history.append(("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", full_feedback))
-        yield chat_history  # Yield the chat history only once
-
-    except Exception as e:
-        logging.error(f"An error occurred during feedback generation: {str(e)}")
+    feedback = llm(conversation, max_length=1000, num_return_sequences=1)[0]["generated_text"]
 
     questionNo = current_question_index + 1
-    # Save complete feedback after streaming
-    add_submission(user_id, message, full_feedback, int(0), "", questionNo)
-
-async def generate_audio_feedback(feedback_buffer):
-    try:
-        response = client.audio.speech.create(
-            model="tts-1",
-            voice="alloy",
-            input=feedback_buffer,
-            response_format="wav"
-        )
+    add_submission(user_id, message, feedback, int(0), "", questionNo)
 
-        audio_data = np.frombuffer(response.read(), dtype=np.int16)
-        sample_rate = 24000  # Default sample rate for OpenAI's WAV output
+    return feedback
 
-        return (sample_rate, audio_data)
-
-    except Exception as e:
-        logging.error(f"An error occurred during speech generation: {str(e)}")
-        return None  # Return None in case of an error
+@spaces.GPU(duration=60)
+async def generate_audio_feedback(feedback_buffer):
+    communicate = edge_tts.Communicate(feedback_buffer)
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
+        tmp_path = tmp_file.name
+        await communicate.save(tmp_path)
+    return tmp_path
 
 async def predict(question_choice, strategy_choice, feedback_level, audio):
-    current_audio_output = None  # Initialize current_audio_output to None
-    final_feedback = ""  # Store only the assistant's feedback
+    current_audio_output = None
 
     if audio is None:
         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:
@@ -161,58 +104,60 @@ async def predict(question_choice, strategy_choice, feedback_level, audio):
     yield chat_history, current_audio_output
 
     try:
-        transcription_future = executor.submit(transcribe_audio, audio_path)
+        transcription_future = executor.submit(transcribe, audio_path)
         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
             return
 
-        chat_history.append(("Student", student_response))  # Add student's transcript
+        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 ..."))
         yield chat_history, current_audio_output
 
-        moderation_response = client.moderations.create(input=student_response)
-        flagged = any(result.flagged for result in moderation_response.results)
-        if flagged:
-            moderated_message = "The message has been flagged. Please see your teacher to clarify."
-            questionNo = thinkingframes.questions.index(question_choice) + 1
-            add_submission(int(user_state.value), moderated_message, "", int(0), "", questionNo)
-            yield chat_history, current_audio_output
-            return
+        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)
 
-        async for chat_update in generate_feedback(int(user_state.value), question_choice, strategy_choice, student_response, feedback_level):
-                # Append the assistant's feedback to the existing chat_history
-                chat_history.extend(chat_update)  
-                final_feedback = chat_history[-1][1]  # Update final_feedback with the latest chunk
-                yield chat_history, current_audio_output # Yield audio output
+        chat_history.append(("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", feedback))
+        yield chat_history, current_audio_output
 
-        feedback_buffer = final_feedback  # Use final_feedback for TTS
-        audio_task = asyncio.create_task(generate_audio_feedback(feedback_buffer))
-        current_audio_output = await audio_task  # Store audio output
-        yield chat_history, current_audio_output  # Yield audio output
+        audio_future = executor.submit(generate_audio_feedback, feedback)
+        audio_output_path = await asyncio.wrap_future(audio_future)
+
+        current_audio_output = (24000, audio_output_path)
+        yield chat_history, current_audio_output
 
     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
 
-with gr.Blocks(title="Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", theme=theme, css=css) as app:
-    with gr.Tab("Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡"):
+# 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):
+    if not policy_checked:
+        return "Please agree to the Things to Note When using the Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡ before submitting.", gr.update(visible=False)
+    user_id, message = add_user_privacy(class_name, index_no)
+    if "Error" in message:
+        return message, gr.update(visible=False)
+    user_state.value = user_id
+    return message, gr.update(visible=True)
+
+with gr.Blocks(title="Oral Coach powered by Hugging Face", theme=theme) as app:
+    with gr.Tab("Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡"):
         gr.Markdown("## Student Information")
         class_name = gr.Dropdown(label="Class", choices=CLASS_OPTIONS)
         index_no = gr.Dropdown(label="Index No", choices=[f"{i:02}" for i in range(1, 46)])
 
         policy_text = gr.Markdown(user_acceptance_policy)
-        policy_checkbox = gr.Checkbox(label="I have read and agree to the Things to Note When using the Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", value=False)
+        policy_checkbox = gr.Checkbox(label="I have read and agree to the Things to Note When using the Oral Coach ⚡ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", value=False)
 
         submit_info_btn = gr.Button("Submit Info")
         info_output = gr.Text()
-  
+
         with gr.Column(visible=False) as oral_coach_content:
-            gr.Markdown("## English Language Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡")
-            gr.Markdown(img_html)  # Display the image
+            gr.Markdown("## Powered by Hugging Face")
+            gr.Markdown(img_html)
             with gr.Row():
                 with gr.Column(scale=1):
                     gr.Markdown("### Step 1: Choose a Question")
@@ -229,31 +174,14 @@ with gr.Blocks(title="Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", theme=t
                     submit_answer_btn = gr.Button("Submit Oral Response")
 
                     gr.Markdown("### Step 5: Review your personalised feedback")
-                    feedback_output = gr.Chatbot(
-                        label="Feedback",
-                        scale=4,
-                        height=700,
-                        show_label=True
-                    )
-                    #audio
-            #submit_answer_here        
+                    feedback_output = gr.Chatbot(label="Feedback", scale=4, height=700, show_label=True)
                     audio_output = gr.Audio(type="numpy", label="Audio Playback", format="wav", autoplay="True")
 
-            submit_answer_btn.click(
-                predict,
-                inputs=[question_choice, strategy_choice, feedback_level, audio_input],
-                outputs=[feedback_output, audio_output],
-                api_name="predict"
-            )
-
-        def toggle_oral_coach_visibility(class_name, index_no, policy_checked):
-            if not policy_checked:
-                return "Please agree to the Things to Note When using the Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡ before submitting.", gr.update(visible=False)
-            validation_passed, message, userid = collect_student_info(class_name, index_no)
-            if not validation_passed:
-                return message, gr.update(visible=False)
-            user_state.value = userid
-            return message, gr.update(visible=True)
+                    submit_answer_btn.click(
+                        predict,
+                        inputs=[question_choice, strategy_choice, feedback_level, audio_input],
+                        outputs=[feedback_output, audio_output]
+                    )
 
         submit_info_btn.click(
             toggle_oral_coach_visibility,
@@ -261,7 +189,6 @@ with gr.Blocks(title="Oral Coach  ⚡ ϞϞ(๑⚈ ․̫ ⚈๑)∩ ⚡", theme=t
             outputs=[info_output, oral_coach_content]
         )
 
-    # Define other tabs like Teacher's Dashboard
     create_teachers_dashboard_tab()
 
 app.queue(max_size=20).launch(