Spaces:
Runtime error
Runtime error
Commit
Β·
36f9277
1
Parent(s):
4727045
Fix: chatbot message format and layout improvements
Browse files
app.py
CHANGED
|
@@ -56,7 +56,8 @@ def transcribe_audio(audio_filepath):
|
|
| 56 |
|
| 57 |
# Convert output IDs to text
|
| 58 |
transcript = model.tokenizer.ids_to_text(output_ids[0].cpu())
|
| 59 |
-
|
|
|
|
| 60 |
|
| 61 |
|
| 62 |
# Enhanced Q&A function with conversation history
|
|
@@ -66,14 +67,15 @@ def answer_question_with_history(transcript, question, history, qa_count):
|
|
| 66 |
return history, qa_count, "Please transcribe audio first"
|
| 67 |
|
| 68 |
if qa_count >= 5:
|
| 69 |
-
history.append(
|
|
|
|
| 70 |
return history, qa_count, ""
|
| 71 |
|
| 72 |
# Build context from history for better continuity
|
| 73 |
context = ""
|
| 74 |
-
for
|
| 75 |
-
if
|
| 76 |
-
context += f"Previous question: {
|
| 77 |
|
| 78 |
with torch.inference_mode(), model.llm.disable_adapter():
|
| 79 |
prompt = f"{context}Current question: {question}\n\nTranscript:\n{transcript}"
|
|
@@ -88,11 +90,12 @@ def answer_question_with_history(transcript, question, history, qa_count):
|
|
| 88 |
|
| 89 |
# Add follow-up prompt if under 5 questions
|
| 90 |
if qa_count < 4:
|
| 91 |
-
answer += f"\n\
|
| 92 |
else:
|
| 93 |
-
answer += "\n\
|
| 94 |
|
| 95 |
-
history.append(
|
|
|
|
| 96 |
return history, qa_count + 1, ""
|
| 97 |
|
| 98 |
# Build the Gradio interface
|
|
@@ -107,39 +110,40 @@ with gr.Blocks(theme=theme) as demo:
|
|
| 107 |
|
| 108 |
with gr.Row():
|
| 109 |
with gr.Column(scale=1):
|
| 110 |
-
gr.Markdown("###
|
| 111 |
audio_input = gr.Audio(
|
| 112 |
sources=["microphone", "upload"],
|
| 113 |
type="filepath",
|
| 114 |
label="Record/Upload Audio (MP3, WAV, M4A, etc.)"
|
| 115 |
)
|
| 116 |
transcribe_btn = gr.Button("Transcribe Audio", variant="primary", size="lg")
|
| 117 |
-
|
| 118 |
-
gr.Markdown("### π Transcript")
|
| 119 |
-
transcript_output = gr.Textbox(
|
| 120 |
-
label="",
|
| 121 |
-
lines=15,
|
| 122 |
-
placeholder="Transcript will appear here after clicking 'Transcribe Audio'..."
|
| 123 |
-
)
|
| 124 |
|
| 125 |
with gr.Column(scale=1):
|
| 126 |
-
gr.Markdown("###
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
)
|
| 133 |
-
|
| 134 |
-
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
|
| 144 |
gr.Markdown("""
|
| 145 |
### Example Questions to Try:
|
|
@@ -170,7 +174,7 @@ with gr.Blocks(theme=theme) as demo:
|
|
| 170 |
)
|
| 171 |
|
| 172 |
clear_chat_btn.click(
|
| 173 |
-
fn=lambda t: ([
|
| 174 |
inputs=[transcript_state],
|
| 175 |
outputs=[chatbot, qa_counter]
|
| 176 |
)
|
|
|
|
| 56 |
|
| 57 |
# Convert output IDs to text
|
| 58 |
transcript = model.tokenizer.ids_to_text(output_ids[0].cpu())
|
| 59 |
+
initial_message = [{"role": "assistant", "content": f"Transcript ready. Ask me questions about it."}]
|
| 60 |
+
return transcript, transcript, initial_message, 0
|
| 61 |
|
| 62 |
|
| 63 |
# Enhanced Q&A function with conversation history
|
|
|
|
| 67 |
return history, qa_count, "Please transcribe audio first"
|
| 68 |
|
| 69 |
if qa_count >= 5:
|
| 70 |
+
history.append({"role": "user", "content": question})
|
| 71 |
+
history.append({"role": "assistant", "content": "You've reached the maximum of 5 questions for this transcript. Please transcribe new audio to continue."})
|
| 72 |
return history, qa_count, ""
|
| 73 |
|
| 74 |
# Build context from history for better continuity
|
| 75 |
context = ""
|
| 76 |
+
for msg in history[-4:]: # Use last 2 exchanges for context
|
| 77 |
+
if msg.get("role") == "user":
|
| 78 |
+
context += f"Previous question: {msg['content']}\n"
|
| 79 |
|
| 80 |
with torch.inference_mode(), model.llm.disable_adapter():
|
| 81 |
prompt = f"{context}Current question: {question}\n\nTranscript:\n{transcript}"
|
|
|
|
| 90 |
|
| 91 |
# Add follow-up prompt if under 5 questions
|
| 92 |
if qa_count < 4:
|
| 93 |
+
answer += f"\n\nQuestion {qa_count + 1}/5 - What else would you like to know?"
|
| 94 |
else:
|
| 95 |
+
answer += "\n\nThis is your final question for this transcript."
|
| 96 |
|
| 97 |
+
history.append({"role": "user", "content": question})
|
| 98 |
+
history.append({"role": "assistant", "content": answer})
|
| 99 |
return history, qa_count + 1, ""
|
| 100 |
|
| 101 |
# Build the Gradio interface
|
|
|
|
| 110 |
|
| 111 |
with gr.Row():
|
| 112 |
with gr.Column(scale=1):
|
| 113 |
+
gr.Markdown("### Audio Input")
|
| 114 |
audio_input = gr.Audio(
|
| 115 |
sources=["microphone", "upload"],
|
| 116 |
type="filepath",
|
| 117 |
label="Record/Upload Audio (MP3, WAV, M4A, etc.)"
|
| 118 |
)
|
| 119 |
transcribe_btn = gr.Button("Transcribe Audio", variant="primary", size="lg")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 120 |
|
| 121 |
with gr.Column(scale=1):
|
| 122 |
+
gr.Markdown("### Transcript")
|
| 123 |
+
transcript_output = gr.Textbox(
|
| 124 |
+
label="",
|
| 125 |
+
lines=10,
|
| 126 |
+
placeholder="Transcript will appear here after clicking 'Transcribe Audio'...",
|
| 127 |
+
max_lines=10
|
| 128 |
)
|
| 129 |
+
|
| 130 |
+
gr.Markdown("### Interactive Q&A")
|
| 131 |
+
chatbot = gr.Chatbot(
|
| 132 |
+
type="messages",
|
| 133 |
+
height=400,
|
| 134 |
+
label="Conversation History",
|
| 135 |
+
bubble_full_width=False
|
| 136 |
+
)
|
| 137 |
+
|
| 138 |
+
with gr.Row():
|
| 139 |
+
question_input = gr.Textbox(
|
| 140 |
+
label="Your Question",
|
| 141 |
+
placeholder="e.g., What was the main topic? Why did they say that?",
|
| 142 |
+
scale=4
|
| 143 |
+
)
|
| 144 |
+
ask_btn = gr.Button("Ask", variant="primary", scale=1)
|
| 145 |
+
|
| 146 |
+
clear_chat_btn = gr.Button("Clear Chat", variant="secondary")
|
| 147 |
|
| 148 |
gr.Markdown("""
|
| 149 |
### Example Questions to Try:
|
|
|
|
| 174 |
)
|
| 175 |
|
| 176 |
clear_chat_btn.click(
|
| 177 |
+
fn=lambda t: ([{"role": "assistant", "content": "Chat cleared. Ask me questions about the transcript."}] if t else [], 1 if t else 0),
|
| 178 |
inputs=[transcript_state],
|
| 179 |
outputs=[chatbot, qa_counter]
|
| 180 |
)
|