Update to pixeltable 0.7.8 / gradio 5.49: UDFs to udfs.py, model_kwargs, pinned deps, python 3.11

README.md

@@ -4,11 +4,15 @@ emoji: 🐠
 colorFrom: green
 colorTo: purple
 sdk: gradio
-sdk_version: 5.1.0
+sdk_version: 5.49.0
 app_file: app.py
 pinned: false
 license: apache-2.0
 short_description: Conversation Intelligence
 ---
 
+Call analysis tool built with [Pixeltable](https://github.com/pixeltable/pixeltable). Docs: https://docs.pixeltable.com
+
+Requires an `OPENAI_API_KEY` secret.
+
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,86 +1,81 @@
+import os
+
 import gradio as gr
 import pixeltable as pxt
-from pixeltable.iterators import FrameIterator, StringSplitter
-from pixeltable.functions.video import extract_audio
-from pixeltable.functions.audio import get_metadata
 from pixeltable.functions import openai
-import os
-import getpass
-import numpy as np
+from pixeltable.functions.audio import get_metadata
 from pixeltable.functions.huggingface import sentence_transformer
+from pixeltable.functions.string import string_splitter
+from pixeltable.functions.video import extract_audio
+
+from udfs import create_chatbot_prompt, generate_insights
 
-# Store OpenAI API Key
-if 'OPENAI_API_KEY' not in os.environ:
-    os.environ['OPENAI_API_KEY'] = getpass.getpass('Enter your OpenAI API key:')
+# The OpenAI API key must be provided via the OPENAI_API_KEY environment variable (a Space secret).
+OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
 
 MAX_VIDEO_SIZE_MB = 35
 
 def process_video(video_file, progress=gr.Progress()):
 
     progress(0, desc="Initializing...")
-    
+
+    if not OPENAI_API_KEY:
+        return '', '', None, 'OPENAI_API_KEY is not set. Add it as a Space secret and restart the Space.'
+
     try:
         # Create a Table, a View, and Computed Columns
-        pxt.drop_dir('gong_demo', force=True)
+        pxt.drop_dir('gong_demo', force=True, if_not_exists='ignore')
         pxt.create_dir('gong_demo')
-        
+
         calls_table = pxt.create_table(
             'gong_demo.calls', {
-            "video": pxt.VideoType(nullable=True),
+            'video': pxt.Video | None,
             }
         )
-        
+
         # Create computed columns to store transformations and persist outputs
-        calls_table['audio'] = extract_audio(calls_table.video, format='mp3')
-        calls_table['metadata'] = get_metadata(calls_table.audio)
-        calls_table['transcription'] = openai.transcriptions(audio=calls_table.audio, model='whisper-1')
-        calls_table['transcription_text'] = calls_table.transcription.text.astype(pxt.StringType())
+        calls_table.add_computed_column(audio=extract_audio(calls_table.video, format='mp3'))
+        calls_table.add_computed_column(metadata=get_metadata(calls_table.audio))
+        calls_table.add_computed_column(transcription=openai.transcriptions(audio=calls_table.audio, model='whisper-1'))
+        calls_table.add_computed_column(transcription_text=calls_table.transcription.text.astype(pxt.String))
 
         sentences_view = pxt.create_view(
             'gong_demo.sentences',
             calls_table,
-            iterator=StringSplitter.create(
+            iterator=string_splitter(
                 text=calls_table.transcription_text,
                 separators='sentence'
             )
         )
 
-        @pxt.expr_udf
-        def e5_embed(text: str) -> np.ndarray:
-            return sentence_transformer(text, model_id='intfloat/e5-large-v2')
-
-        sentences_view.add_embedding_index('text', string_embed=e5_embed)
+        sentences_view.add_embedding_index(
+            'text', string_embed=sentence_transformer.using(model_id='intfloat/e5-large-v2')
+        )
 
         progress(0.2, desc="Creating UDFs...")
 
-        # Custom User-Defined Function (UDF) for Generating Insights
-        @pxt.udf
-        def generate_insights(transcription: str) -> list[dict]:
-            system_msg = 'You are an AI assistant that analyzes call transcriptions. Analyze the following call transcription and provide insights on: 1. Main topics discussed 2. Action items 3. Sentiment analysis 4. Key questions asked'
-            user_msg = f'Transcription: "{transcription}"'
-            return [
-                {'role': 'system', 'content': system_msg},
-                {'role': 'user', 'content': user_msg}
-            ]
-        
         # Apply the UDF to create a new column
-        calls_table['insights_prompt'] = generate_insights(calls_table.transcription_text)
-        
+        calls_table.add_computed_column(insights_prompt=generate_insights(calls_table.transcription_text))
+
         progress(0.4, desc="Generating insights...")
-        
+
         # Generate insights using OpenAI's chat completion API
-        calls_table['insights_response'] = openai.chat_completions(messages=calls_table.insights_prompt, model='gpt-3.5-turbo', max_tokens=500)
-        
+        calls_table.add_computed_column(
+            insights_response=openai.chat_completions(
+                messages=calls_table.insights_prompt, model='gpt-4o-mini', model_kwargs={'max_tokens': 500}
+            )
+        )
+
         # Extract the content of the response
-        calls_table['insights'] = calls_table.insights_response.choices[0].message.content
-        
+        calls_table.add_computed_column(insights=calls_table.insights_response.choices[0].message.content)
+
         if not video_file:
-            return "Please upload a video file.", ""
+            return '', '', None, 'Please upload a video file.'
 
         # Check video file size
         video_size = os.path.getsize(video_file) / (1024 * 1024)  # Convert to MB
         if video_size > MAX_VIDEO_SIZE_MB:
-            return f"The video file is larger than {MAX_VIDEO_SIZE_MB} MB. Please upload a smaller file.", ""
+            return '', '', None, f'The video file is larger than {MAX_VIDEO_SIZE_MB} MB. Please upload a smaller file.'
 
         progress(0.6, desc="Processing video...")
         
@@ -101,71 +96,64 @@ def process_video(video_file, progress=gr.Progress()):
         return transcription, insights, audio, "Processing complete"
 
     except Exception as e:
-        return f"An error occurred during video processing: {str(e)}", ""
+        return '', '', None, f'An error occurred during video processing: {str(e)}'
 
 # Perform similarity search
 def similarity_search(query, num_results, progress=gr.Progress()):
     sentences_view = pxt.get_table('gong_demo.sentences')
-    
+
     progress(0.5, desc="Performing search...")
 
-    sim = sentences_view.text.similarity(query)
-    results = sentences_view.order_by(sim, asc=False).limit(num_results).select(sentences_view.text).collect().to_pandas()
+    sim = sentences_view.text.similarity(string=query)
+    results = sentences_view.order_by(sim, asc=False).limit(int(num_results)).select(sentences_view.text).collect().to_pandas()
     
     progress(1.0, desc="Search complete")
     return results
 
 def chatbot_response(message, chat_history):
-    @pxt.udf
-    def create_chatbot_prompt(context: str, question: str) -> list[dict]:
-        system_message = "You are an AI assistant that answers questions about a call based on the provided context. If the answer cannot be found in the context, say that you don't know."
-        user_message = f"Context:\n{context}\n\nQuestion: {question}"
-        return [
-            {"role": "system", "content": system_message},
-            {"role": "user", "content": user_message}
-        ]
-
     try:
         sentences_view = pxt.get_table('gong_demo.sentences')
 
         # Perform similarity search to get relevant context
-        sim = sentences_view.text.similarity(message)
+        sim = sentences_view.text.similarity(string=message)
         context = sentences_view.order_by(sim, asc=False).limit(5).select(sentences_view.text, sim=sim).collect()
-        
+
         # Prepare the context for the prompt
-        context_text = "\n".join([row['text'] for row in context])
+        context_text = '\n'.join([row['text'] for row in context])
 
         # Create a temporary table for the chatbot interaction
-        temp_table = pxt.create_table('gong_demo.temp_chatbot', {'question': pxt.StringType()})
+        temp_table = pxt.create_table('gong_demo.temp_chatbot', {'question': pxt.String})
         temp_table.insert([{'question': message}])
 
         # Create computed columns for the prompt and response
-        temp_table['chatbot_prompt'] = create_chatbot_prompt(context_text, temp_table.question)
-        temp_table['chatbot_response'] = openai.chat_completions(
+        temp_table.add_computed_column(chatbot_prompt=create_chatbot_prompt(context_text, temp_table.question))
+        temp_table.add_computed_column(chatbot_response=openai.chat_completions(
             messages=temp_table.chatbot_prompt,
-            model='gpt-4o-mini-2024-07-18',
-            max_tokens=300
-        ) 
-        temp_table['answer'] = temp_table.chatbot_response.choices[0].message.content
-        
+            model='gpt-4o-mini',
+            model_kwargs={'max_tokens': 300}
+        ))
+        temp_table.add_computed_column(answer=temp_table.chatbot_response.choices[0].message.content)
+
         answer = temp_table.select(temp_table.answer).collect()['answer'][0]
-        
+
         # Clean up the temporary table
         pxt.drop_table('gong_demo.temp_chatbot', force=True)
-        
-        chat_history.append((message, answer))
-        return "", chat_history  # Return both expected outputs
+
+        chat_history.append({'role': 'user', 'content': message})
+        chat_history.append({'role': 'assistant', 'content': answer})
+        return '', chat_history  # Return both expected outputs
     except Exception as e:
-        error_message = f"An error occurred: {str(e)}"
-        chat_history.append((message, error_message))
-        return "", chat_history  # Return both expec
+        error_message = f'An error occurred: {str(e)}'
+        chat_history.append({'role': 'user', 'content': message})
+        chat_history.append({'role': 'assistant', 'content': error_message})
+        return '', chat_history  # Return both expected outputs
 
 # Gradio interface
 with gr.Blocks(theme=gr.themes.Base()) as demo:
     gr.Markdown(
         """
         <div style="text-align: left; margin-bottom: 20px;">
-            <img src="https://raw.githubusercontent.com/pixeltable/pixeltable/main/docs/source/data/pixeltable-logo-large.png" alt="Pixeltable" style="max-width: 150px;" />
+            <img src="https://raw.githubusercontent.com/pixeltable/pixeltable/main/docs/resources/pixeltable-logo-large.png" alt="Pixeltable" style="max-width: 150px;" />
             <h1 style="margin-top: 10px;">Call Analysis AI Tool</h1>
         </div>
         """
@@ -177,6 +165,11 @@ with gr.Blocks(theme=gr.themes.Base()) as demo:
         </p>
         """
     )
+    if not OPENAI_API_KEY:
+        gr.Markdown(
+            '**`OPENAI_API_KEY` is not set.** Add it as a Space secret (Settings > Secrets) '
+            'and restart the Space to enable analysis, search, and the AI assistant.'
+        )
     
     with gr.Row():
         with gr.Column():
@@ -231,7 +224,7 @@ with gr.Blocks(theme=gr.themes.Base()) as demo:
                     )
             
                 with gr.TabItem("🤖 AI Assistant"):
-                    chatbot = gr.Chatbot(height=400, label="Chat with AI about the call")
+                    chatbot = gr.Chatbot(height=400, label="Chat with AI about the call", type='messages')
                     with gr.Row():
                         msg = gr.Textbox(label="Ask a question about the call", placeholder="e.g., What were the main points discussed?", scale=4)
                         send_btn = gr.Button("Send", variant="secondary", scale=1)
@@ -263,7 +256,7 @@ with gr.Blocks(theme=gr.themes.Base()) as demo:
 
     msg.submit(chatbot_response, [msg, chatbot], [msg, chatbot])
     send_btn.click(chatbot_response, [msg, chatbot], [msg, chatbot])
-    clear.click(lambda: None, None, chatbot, queue=False)
+    clear.click(lambda: [], None, chatbot, queue=False)
 
 if __name__ == "__main__":
     demo.launch(show_api=False, allowed_paths=[os.path.expanduser("~/.pixeltable/media")])

requirements.txt

@@ -1,4 +1,6 @@
-pixeltable
-openai
-sentence-transformers
-spacy
+pixeltable==0.7.8
+openai==3.16.2
+sentence-transformers==6.1.0
+spacy==3.8.16
+en_core_web_sm @ https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl
+numpy==2.4.6

udfs.py

@@ -0,0 +1,23 @@
+import pixeltable as pxt
+
+
+# Pixeltable rejects @pxt.udf defined in the __main__ script or nested inside
+# functions; HF Spaces runs `python app.py`, so UDFs live in this module.
+@pxt.udf
+def generate_insights(transcription: str) -> list[dict]:
+    system_msg = 'You are an AI assistant that analyzes call transcriptions. Analyze the following call transcription and provide insights on: 1. Main topics discussed 2. Action items 3. Sentiment analysis 4. Key questions asked'
+    user_msg = f'Transcription: "{transcription}"'
+    return [
+        {'role': 'system', 'content': system_msg},
+        {'role': 'user', 'content': user_msg},
+    ]
+
+
+@pxt.udf
+def create_chatbot_prompt(context: str, question: str) -> list[dict]:
+    system_message = "You are an AI assistant that answers questions about a call based on the provided context. If the answer cannot be found in the context, say that you don't know."
+    user_message = f'Context:\n{context}\n\nQuestion: {question}'
+    return [
+        {'role': 'system', 'content': system_message},
+        {'role': 'user', 'content': user_message},
+    ]