Update and Verify tweetsgenerator Example (#1058)

* Add new images

* Remove param??

* Fixes

* Add 4o instead

* Readme

.github/workflows/publish_executor_containers.yaml

@@ -44,3 +44,7 @@ jobs:
           docker push tensorlake/pdf-structured-extraction-inkwell-example
           docker push tensorlake/openai-image
           docker push tensorlake/base-image
+          docker push tensorlake/base-image:3.10
+          docker push tensorlake/base-image:3.11
+          docker push tensorlake/openai-image:3.10
+          docker push tensorlake/openai-image:3.11

examples/tweetsgenerator/README.md

@@ -15,6 +15,9 @@ This project demonstrates how to build a tweet generation and ranking pipeline u
 - Docker and Docker Compose (for containerized setup)
 - OpenAI API key
 
+NOTE: The example is setup to work with python 3.10. See `image` param in the workflow and the image name used in
+the docker compose file. Please update them as needed while following the instructions below for Docker deployment.
+
 ## Installation and Usage
 
 ### Option 1: Local Installation - In Process

examples/tweetsgenerator/docker-compose.yml

@@ -15,7 +15,7 @@ services:
       - data:/tmp/indexify-blob-storage
 
   openai:
-    image: tensorlake/openai-image:latest
+    image: tensorlake/openai-image:3.10
     environment:
       - OPENAI_API_KEY=${OPENAI_API_KEY}
     command: ["indexify-cli", "executor", "--server-addr", "indexify:8900"]
@@ -25,7 +25,7 @@ services:
       - data:/tmp/indexify-blob-storage
 
   base-executor:
-    image: tensorlake/base-image:latest
+    image: tensorlake/base-image:3.10
     command:
       [
         "indexify-cli",

examples/tweetsgenerator/workflow.py

@@ -1,6 +1,7 @@
 import logging
 import os
-from typing import Dict, List
+from typing import List
+
 from pydantic import BaseModel, Field
 from indexify import RemoteGraph
 from indexify.functions_sdk.graph import Graph
@@ -17,82 +18,106 @@ class Tweets(BaseModel):
 class RankedTweets(BaseModel):
     scores: List[float] = Field(description="List of scores for the tweets. Higher score means better tweet.")
 
+
 # Define custom image
+base_image_3_10 = (
+    Image()
+    .name("tensorlake/base-image")
+    .base_image("python:3.10-slim-bookworm")
+    .tag("3.10")
+)
 
-base_image = (
+base_image_3_11 = (
     Image()
     .name("tensorlake/base-image")
+    .base_image("python:3.11-slim-bookworm")
+    .tag("3.11")
+)
+
+openai_image_3_10 = (
+    Image()
+    .name("tensorlake/openai-image")
+    .base_image("python:3.10-slim-bookworm")
+    .tag("3.10")
+    .run("pip install openai")
 )
-openai_image = (
+
+openai_image_3_11 = (
     Image()
     .name("tensorlake/openai-image")
+    .base_image("python:3.11-slim-bookworm")
+    .tag("3.11")
     .run("pip install openai")
 )
 
-@indexify_function(image=openai_image)
+
+@indexify_function(image=openai_image_3_10)
 def generate_tweet_topics(subject: str) -> List[str]:
     """Generate topics for tweets about a given subject."""
     import openai
     from pydantic import BaseModel, Field
     from typing import List
+
     client = openai.OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
 
     class Topics(BaseModel):
         topics: List[str] = Field(default_factory=list)
 
-    response = client.chat.completions.create(
-        model="gpt-4",
+    response = client.beta.chat.completions.parse(
+        model="gpt-4o-mini-2024-07-18",
         messages=[
             {"role": "system", "content": "You are a helpful assistant that generates topics for a tweet about a given subject."},
             {"role": "user", "content": f"Generate 5 topics for a tweet about {subject}"},
         ],
-        response_model=Topics
+        response_format=Topics
     )
-    topics = response.choices[0].message.content
+    topics = response.choices[0].message.parsed
     return topics.topics
 
-@indexify_function(image=openai_image)
+@indexify_function(image=openai_image_3_10)
 def generate_tweet(topic: str) -> str:
     """Generate a tweet about a given topic."""
     import openai
     from pydantic import BaseModel, Field
+
     client = openai.OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
 
     class Tweet(BaseModel):
         tweet: str = Field(description="a tweet about the given topic")
 
-    response = client.chat.completions.create(
-        model="gpt-4",
+    response = client.beta.chat.completions.parse(
+        model="gpt-4o-mini-2024-07-18",
         messages=[
             {"role": "system", "content": "You are a helpful assistant that generates a tweet about a given topic."},
             {"role": "user", "content": f"Generate a tweet about {topic}"},
         ],
-        response_model=Tweet
+        response_format=Tweet
     )
-    tweet = response.choices[0].message.content
+    tweet = response.choices[0].message.parsed
     return tweet.tweet
 
-@indexify_function(image=base_image,accumulate=Tweets)
+@indexify_function(image=base_image_3_10, accumulate=Tweets)
 def accumulate_tweets(acc: Tweets, tweet: str) -> Tweets:
     """Accumulate generated tweets."""
     acc.tweets.append(tweet)
     return acc
 
-@indexify_function(image=openai_image)
+@indexify_function(image=openai_image_3_10)
 def score_and_rank_tweets(tweets: Tweets) -> RankedTweets:
     """Score and rank the accumulated tweets."""
     import openai
     client = openai.OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
     tweet_contents = "\n".join(tweets.tweets)
-    response = client.chat.completions.create(
-        model="gpt-4",
+
+    response = client.beta.chat.completions.parse(
+        model="gpt-4o-mini-2024-07-18",
         messages=[
             {"role": "system", "content": "You are a helpful assistant that scores and ranks tweets based on their relevance to a given topic."},
             {"role": "user", "content": f"Score and rank the following tweets, separated by new lines: {tweet_contents}"},
         ],
-        response_model=RankedTweets
+        response_format=RankedTweets
     )
-    ranked_tweets = response.choices[0].message.content
+    ranked_tweets = response.choices[0].message.parsed
     return ranked_tweets
 
 def create_tweets_graph():
@@ -116,7 +141,11 @@ def run_workflow(mode: str, server_url: str = 'http://localhost:8900'):
         raise ValueError("Invalid mode. Choose 'in-process-run' or 'remote-run'.")
 
     import httpx
-    subject = httpx.get("https://discord.com/blog/how-discord-reduced-websocket-traffic-by-40-percent").text
+    # NOTE contact Tensorlake to get a key!
+    subject = httpx.get(
+        url="https://discord.com/blog/how-discord-reduced-websocket-traffic-by-40-percent",
+    ).text
+
     logging.info(f"Generating tweets for subject: {subject[:100]}...")
 
     invocation_id = graph.run(block_until_done=True, subject=subject)