Execute a Pipeline

Basic Execution

When you have a pre-built pipeline, here is how you can execute it:

# Initialize the pipeline
pipeline = step_1 | step_2 | step_3  # or replace it with your prebuilt pipeline

# Prepare initial state
initial_state = {
    "user_query": "What is machine learning?",
    "history": "",
    "context": ""
}

# Execute the pipeline
final_state = await pipeline.invoke(initial_state)

# Access the results
response = final_state["response"]
references = final_state["references"]

Execution with Configuration

You can also run the pipeline with configuration. Configuration can include debugging, caching, timeouts, retries, and other runtime parameters (top_k for retriever, etc). The configuration is passed to the invoke() method and affects how the pipeline executes.

# Execute with custom configuration
config = {
    "debug_state": True,  # Include state logs in output
    "cache_enabled": True,
    "timeout": 30,  # seconds
    "max_retries": 3
    # other configuration
}

final_state = await pipeline.invoke(
    initial_state=initial_state,
    config=config
)