TL;DR

I recently migrated the content-automation repository to use the meta-llama/llama-4-scout-17b-16e-instruct model, replacing previous models like llama3-70b-8192. This change involved updating multiple files, including main.py, settings.yml, and craft_publisher.py. The goal was to improve performance and reduce errors.

The Problem

The initial problem was the decommissioning of the llama3-70b-8192 model, which was being used in the content-automation project. This model was replaced by the meta-llama/llama-4-scout-17b-16e-instruct model. However, the transition required updates to various parts of the codebase to ensure compatibility and correct functionality.

What I Tried First

My first approach was to identify all occurrences of the llama3-70b-8192 model in the codebase and replace them with the new model. This involved searching through files like main.py, settings.yml, and craft_publisher.py. However, simply replacing the model references was not enough, as the API endpoints and payload formats also needed to be adjusted.

The Implementation

Updating Model References

The first step was to update the model references in main.py and settings.yml. This involved changing the MODEL_SHORTFORM variable to point to the new model.

# src/main.py
MODEL_SHORTFORM = "meta-llama/llama-4-scout-17b-16e-instruct"

# config/settings.yml
MODEL_SHORTFORM: meta-llama/llama-4-scout-17b-16e-instruct

Adjusting Craft API Payload

The Craft API expects an array in the documents payload. This required modifying the _create_document function in craft_publisher.py to correctly format the payload.

# src/craft_publisher.py
def _create_document(
    title: str, content: str, tags: List[str], categories: List[str]
) -> Optional[str]:
    """Creates document in Craft. Returns doc ID or None."""
    payload = {
        "title": title,
        "content": content,
        "tags": tags,
        "categories": categories,
    }
    # Ensure tags and categories are lists
    payload["tags"] = [{"value": tag} for tag in tags]
    payload["categories"] = [{"value": category} for category in categories]
    response = requests.post(CRAFT_API_URL, json=payload)
    if response.status_code == 201:
        return response.json()["id"]
    else:
        return None

Updating API Endpoints

The API endpoint for the new model needed to be updated in craft_publisher.py. This involved changing the connect_base parameter to use the correct endpoint.

# src/craft_publisher.py
def publish_to_craft(doc_id: str, content: str):
    """Publishes content to Craft."""
    url = f"{CRAFT_API_URL}/entries"
    payload = {"doc_id": doc_id, "content": content}
    response = requests.post(url, json=payload)
    if response.status_code == 201:
        print("Published to Craft successfully.")
    else:
        print("Failed to publish to Craft.")

Key Takeaway

The key takeaway from this experience is the importance of thoroughly testing and validating changes when replacing models in a machine learning-based project. Ensuring that all parts of the codebase are updated correctly is crucial to avoid errors and performance issues.

What's Next

The next step is to monitor the performance of the new model and make any necessary adjustments to the codebase. Additionally, I plan to explore other models and compare their performance to further improve the content-automation project.

vibecoding #buildinpublic #AI #Productivity

---

Part of my Build in Public series — sharing the real process of building SaaS projects from Playa del Carmen, México.

Repo: zaerohell/content-automation · 2026-06-24

#playadev #buildinpublic