Skip to main content
In the Track LLM inputs & outputs tutorial, the basics of tracking the inputs and outputs of your LLMs was covered. In this tutorial you will learn how to:
  • Track data as it flows through your application
  • Track metadata at call time

​
Tracking nested function calls

LLM-powered applications can contain multiple LLMs calls and additional data processing and validation logic that is important to monitor. Even deep nested call structures common in many apps, Weave will keep track of the parent-child relationships in nested functions as long as weave.op() is added to every function you’d like to track. Building on the quickstart example, the following code adds additional logic to count the returned items from the LLM and wrap them all in a higher level function. Additionally, the example uses weave.op() to trace every function, its call order, and its parent-child relationship:
  • Python
  • TypeScript
import weave
import json
from openai import OpenAI

client = OpenAI()

# highlight-next-line
@weave.op()
def extract_dinos(sentence: str) -> dict:
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {
                "role": "system",
                "content": """Extract any dinosaur `name`, their `common_name`, \
names and whether its `diet` is a herbivore or carnivore, in JSON format."""
            },
            {
                "role": "user",
                "content": sentence
            }
            ],
            response_format={ "type": "json_object" }
        )
    return response.choices[0].message.content

# highlight-next-line
@weave.op()
def count_dinos(dino_data: dict) -> int:
    # count the number of items in the returned list
    k = list(dino_data.keys())[0]
    return len(dino_data[k])

# highlight-next-line
@weave.op()
def dino_tracker(sentence: str) -> dict:
    # extract dinosaurs using a LLM
    dino_data = extract_dinos(sentence)

    # count the number of dinosaurs returned
    dino_data = json.loads(dino_data)
    n_dinos = count_dinos(dino_data)
    return {"n_dinosaurs": n_dinos, "dinosaurs": dino_data}

# highlight-next-line
weave.init('jurassic-park')

sentence = """I watched as a Tyrannosaurus rex (T. rex) chased after a Triceratops (Trike), \
both carnivore and herbivore locked in an ancient dance. Meanwhile, a gentle giant \
Brachiosaurus (Brachi) calmly munched on treetops, blissfully unaware of the chaos below."""

result = dino_tracker(sentence)
print(result)
Nested functionsWhen you run the above code, you see the the inputs and outputs from the two nested functions (extract_dinos and count_dinos), as well as the automatically-logged OpenAI trace.Nested Weave Trace

​
Tracking metadata

You can track metadata by using the weave.attributes context manager and passing it a dictionary of the metadata to track at call time. Continuing our example from above:
  • Python
  • TypeScript
import weave

weave.init('jurassic-park')

sentence = """I watched as a Tyrannosaurus rex (T. rex) chased after a Triceratops (Trike), \
both carnivore and herbivore locked in an ancient dance. Meanwhile, a gentle giant \
Brachiosaurus (Brachi) calmly munched on treetops, blissfully unaware of the chaos below."""

# track metadata alongside our previously defined function
# highlight-next-line
with weave.attributes({'user_id': 'lukas', 'env': 'production'}):
    result = dino_tracker(sentence)
We recommend tracking metadata at run time, such as your user IDs and your code’s environment status (development, staging, or production).To track system settings, such as a system prompt, we recommend using Weave Models

​
What’s next?

  • Follow the App Versioning tutorial to capture, version, and organize ad-hoc prompt, model, and application changes.
⌘I