cmf-query

name	cmf-query
description	Use when querying CMF metadata programmatically — listing pipelines, stages, executions, artifacts; tracing lineage; comparing hyperparameters across runs; or exporting metadata for analysis. Uses the CmfQuery Python API or CLI listing commands.
version	1.0.0

Write the query the user needs using CmfQuery or the CMF CLI. Identify what they want to find (pipeline overview, lineage, hyperparameter comparison, specific artifacts) and produce the code.

Initialize

from cmflib.cmfquery import CmfQuery
query = CmfQuery(filepath="mlmd")   # or path to a downloaded mlmd file

Common patterns

List pipelines and stages

pipelines = query.get_pipeline_names()           # ['my_pipeline', ...]
stages = query.get_pipeline_stages("my_pipeline") # ['Prepare', 'Train', ...]

Executions with hyperparameters

runs = query.get_all_executions_in_stage("Train")
# runs is a pandas DataFrame — one row per run, columns = custom_properties
print(runs[["name", "learning_rate", "n_estimators"]])

Artifact lineage

# What fed INTO this artifact?
parents = query.get_one_hop_parent_artifacts("models/rf.pkl")

# What was produced FROM this artifact?
children = query.get_one_hop_child_artifacts("data/train.csv")

# Which execution produced this artifact?
execs = query.get_one_hop_parent_executions("models/rf.pkl")

Compare runs across a stage

runs = query.get_all_executions_in_stage("Train")
print(runs[["name", "learning_rate", "n_estimators"]].sort_values("learning_rate"))

Walk full lineage (two hops)

model = "models/rf.pkl"
parents = query.get_one_hop_parent_artifacts(model)
for p in parents["name"].tolist():
    grandparents = query.get_one_hop_parent_artifacts(p)
    print(f"{p} ← {grandparents['name'].tolist()}")

All models in a pipeline

artifacts = query.get_all_artifacts_for_pipeline("my_pipeline")
models = artifacts[artifacts["type"] == "Model"]
print(models[["name", "uri"]])

CLI quick inspection

cmf pipeline list
cmf execution list -p my_pipeline
cmf artifact list  -p my_pipeline

See references/query-api.md for the full CmfQuery method reference.

Troubleshooting

• Empty DataFrame — check pipeline/stage names with get_pipeline_names() / get_pipeline_stages(pipeline)
• mlmd file not found — run cmf metadata pull -p <name> first if querying a teammate's data
• AttributeError — query results are pandas.DataFrame; use .tolist() or standard DataFrame operations

---

Docs: CmfQuery API reference · Web UI executions · Web UI lineage · MCP for natural language queries