Pipeline diagrams

NexuML can write a Mermaid representation of a compiled pipeline when diagram output is enabled in LoggingSpec.

Configure diagram output

from nexuml.core.types import DiagramSpec, LoggingSpec

logging = LoggingSpec(
    diagram=DiagramSpec(
        enabled=True,
        depth=2,
        direction="TB",
        show_params=True,
        show_shapes=True,
        show_metrics=True,
        output_dir=".experiments/diagrams",
    )
)

Generate during build

nexuml resolve my-scenario
nexuml build configs/my-scenario.yaml

build compiles the pipeline and, when diagram output is enabled, writes <output_dir>/<scenario-name>.md. The command itself reports the compiled pipeline summary; it does not rely on printing the full Mermaid source to standard output.

Generate during training

nexuml train performs the same configured diagram export before the training session begins. You do not need a separate build invocation just to create the diagram.

Diagram generation is diagnostic: a diagram-export failure is reported as a warning rather than turning a valid pipeline compile into a failure.

Rendering

Material for MkDocs renders Mermaid code fences directly. The exported .md can also be rendered by any Mermaid-compatible tool.

See also

• Architecture
• TensorDict data flow
• nexuml.core.diagram