Runs
Access training runs through client.runs in the Datature Vi SDK.
A run is a single training execution. Each run belongs to a flow that defines its configuration (the dataset, model architecture, hyperparameters, and schedule). You can create runs from flows, monitor their progress, kill running runs, and retrieve model checkpoints from completed runs via the Models API.
- Vi SDK installed with authentication configured
- A secret key for API authentication
- A training project with at least one workflow configured
Methods
list()
List all training runs in your organization.
runs = client.runs.list()
for run in runs.items:
print(f"Run: {run.run_id}")
print(f"Created: {run.metadata.time_created}")from vi.api.types import PaginationParams
runs = client.runs.list(pagination=PaginationParams(page_size=50))
for run in runs.items:
print(f"{run.run_id}")for page in client.runs.list():
for run in page.items:
status = "Unknown"
if run.status.conditions:
status = run.status.conditions[-1].condition.value
print(f"{run.run_id}: {status}")completed_runs = []
for page in client.runs.list():
for run in page.items:
if run.status.conditions:
latest = run.status.conditions[-1]
if latest.condition.value == "Succeeded":
completed_runs.append(run)
print(f"Found {len(completed_runs)} completed runs")all_runs = list(client.runs.list().all_items())
print(f"Total runs: {len(all_runs)}")Returns: PaginatedResponse[Run]
get()
Get a specific training run by ID.
run = client.runs.get("run_abc123")
print(f"Run ID: {run.run_id}")
print(f"Organization: {run.organization_id}")run = client.runs.get("run_abc123")
if run.status.conditions:
latest = run.status.conditions[-1]
print(f"Status: {latest.condition.value}")
print(f"Message: {latest.message}")run = client.runs.get("run_abc123")
print(f"Flow Name: {run.spec.flow.name}")
print(f"Schema: {run.spec.flow.schema}")
print(f"Blocks: {len(run.spec.flow.blocks)}")
if run.spec.training_project:
print(f"Training Dataset: {run.spec.training_project}")run = client.runs.get("run_abc123")
run.info() # Prints formatted run summaryimport time
def wait_for_run(run_id: str, max_wait: int = 3600, check_interval: int = 30) -> str:
"""Wait for a run to complete and return the final status."""
start_time = time.time()
while time.time() - start_time < max_wait:
run = client.runs.get(run_id)
if run.status.conditions:
latest = run.status.conditions[-1]
status = latest.condition.value
print(f"Status: {status}")
if status in ["Succeeded", "Failed", "Error", "Cancelled"]:
return status
time.sleep(check_interval)
raise TimeoutError("Run did not complete in time")
status = wait_for_run("run_abc123")
if status == "Succeeded":
model = client.get_model("run_abc123")Returns: Run
create()
Create and start a new training run from a flow configuration.
run = client.runs.create(
flow="flow_abc123",
training_project="project_xyz789"
)
print(f"Started run: {run.run_id}")run = client.runs.create(
flow={
"name": "Quick Training",
"schema": "v1",
"blocks": [],
"settings": {},
"tolerations": {}
}
)run = client.runs.create(
flow="flow_abc123",
metadata={"experiment": "baseline", "version": "1.0"}
)Returns: Run
start()
Alias for create(). Creates and starts a new training run.
run = client.runs.start(
flow="flow_abc123",
training_project="project_xyz789"
)
print(f"Started run: {run.run_id}")Returns: Run
update()
Update an existing training run. Uses PATCH semantics: only provided fields are modified.
run = client.runs.update(
run_id="run_abc123",
metadata={"experiment": "baseline_v2", "notes": "improved accuracy"}
)run = client.runs.update(
run_id="run_abc123",
spec={"training_project": "project_new123"}
)Returns: Run
kill()
Terminate a running training run.
run = client.runs.kill(run_id="run_abc123")
print(f"Run killed at: {run.spec.killed_at}")run = client.runs.kill(run_id="run_abc123")
updated_run = client.runs.get(run_id="run_abc123")
if updated_run.status.conditions:
latest = updated_run.status.conditions[-1]
print(f"Status: {latest.condition.value}")Returns: Run
delete()
Delete a training run permanently. This also deletes all associated model checkpoints.
deleted = client.runs.delete(run_id="run_abc123")
print(f"Deleted: {deleted.id}")from vi import ViNotFoundError
try:
client.runs.delete(run_id="run_abc123")
print("Run deleted successfully")
except ViNotFoundError:
print("Run not found or already deleted")Deleting a run permanently removes all training data including logs, metrics, and model checkpoints. Download important models using client.models.download() before deleting a run.
Returns: DeletedResource
Condition values
The condition field in ResourceCondition reports the current run state:
Response types
Run
from vi.api.resources.runs.responses import RunMethods: info() → prints a formatted run summary.
RunSpec
from vi.api.resources.runs.responses import RunSpecRunStatus
from vi.api.resources.runs.responses import RunStatusRelated resources
Updated 30 days ago