Artifact Creation Modes

Create a new artifact version using a single run, collaboratively using distributed writers, or as a patch against a prior version.
New versions of an Artifact can be created in one of three ways:
Simple: A single run provides all the data for a new version. This is the most common case and is best suited for when the run fully recreates the needed data. For example: outputting saved models or model predictions in a table for analysis.
Collaborative: A set of runs collectively provides all the data for a new version. This is best suited for distributed jobs which have multiple runs generating data, often in parallel. For example: evaluating a model in a distributed manner, and outputting the predictions.
Patch: (coming soon) A single run provides a patch of the differences to be applied. This is best suited when a run wants to add data to an artifact without needing to recreate all the already existing data. For example: you have a golden dataset which is created by running a daily web scraper - in this case, you want the run to append new data to the dataset.
Simple Mode
To log a new version of an Artifact with a single run which produces all the files in the artifact, use the Simple Mode:
with wandb.init() as run:
	artifact = wandb.Artifact("artifact_name", "artifact_type")
	# Add Files and Assets to the artifact using 
	# `.add`, `.add_file`, `.add_dir`, and `.add_reference`
	artifact.add_file("image1.png")
	run.log_artifact(artifact)
This pattern is so common, that we provide a method of creating the version without explicitly starting a run using Artifact.save():
artifact = wandb.Artifact("artifact_name", "artifact_type")
# Add Files and Assets to the artifact using 
# `.add`, `.add_file`, `.add_dir`, and `.add_reference`
artifact.add_file("image1.png")
artifact.save()
Collaborative Mode
To allow a collection of runs to collaborate on a version before committing it, use the Collaborative Mode. There are two key ideas to keep in mind when using Collaborative Mode:
Each Run in the collection needs to be aware of the same, unique ID (called distributed_id) in order to collaborate on the same version. As a default, if present, we will use the run's group as set by wandb.init(group=GROUP) as the distributed_id.
There must be a final run which "commits" the version, permanently locking its state.
Consider the following example. Take note that rather than using log_artifact we use upsert_artifact to add the the collaborative artifact and finish_artifact to finalize the commit.:
Run 1:
with wandb.init() as run:
	artifact = wandb.Artifact("artifact_name", "artifact_type")
	# Add Files and Assets to the artifact using 
	# `.add`, `.add_file`, `.add_dir`, and `.add_reference`
	artifact.add_file("image1.png")
	run.upsert_artifact(artifact, distributed_id="my_dist_artifact")
Run 2:
with wandb.init() as run:
	artifact = wandb.Artifact("artifact_name", "artifact_type")
	# Add Files and Assets to the artifact using 
	# `.add`, `.add_file`, `.add_dir`, and `.add_reference`
	artifact.add_file("image2.png")
	run.upsert_artifact(artifact, distributed_id="my_dist_artifact")
Run 3: Must run after Run 1 and Run 2 complete. The Run which calls finish_artifact is welcome to include files in the artifact, but does not need to.
with wandb.init() as run:
	artifact = wandb.Artifact("artifact_name", "artifact_type")
	# Add Files and Assets to the artifact using 
	# `.add`, `.add_file`, `.add_dir`, and `.add_reference`
	artifact.add_file("image3.png")
	run.finish_artifact(artifact, distributed_id="my_dist_artifact")
Patch Mode (Coming Soon)
To create a new version of an Artifact by modifying a previous version, use Patch Mode. A code snippet for Patch Mode will be provided once available for use.
FAQ
How do I log a Table in Collaborative Mode?
For large datasets multiple parallel runs might need to contribute to a single Table. You can use the following pattern to construct such parallel artifacts. The key idea is that each worker puts their own Table in a directory in the artifact. Then, the final worker adds a PartitionTable to the artifact which points to the folder of "parts".
import wandb
import time
​
# We will use ray to launch our runs in parallel
# for demonstration purposes. You can orchestrate
# your parallel runs however you want.
import ray
​
ray.init()
​
artifact_type = "dataset"
artifact_name = "parallel-artifact"
table_name = "distributed_table"
parts_path = "parts"
num_parallel = 5
​
# Each batch of parallel writers should have its own
# unique group name.
group_name = "writer-group-{}".format(round(time.time()))
​
@ray.remote
def train(i):
  """
  Our writer job. Each writer will add one image to the artifact.
  """
  with wandb.init(group=group_name) as run:
    artifact = wandb.Artifact(name=artifact_name, type=artifact_type)
    
    # Add data to a wandb table. In this case we use example data
    table = wandb.Table(columns=["a", "b", "c"], data=[[i, i*2, 2**i]])
    
    # Add the table to folder in the artifact
    artifact.add(table, "{}/table_{}".format(parts_path, i))
    
    # Upserting the artifact creates or appends data to the artifact
    run.upsert_artifact(artifact)
  
# Launch your runs in parallel
result_ids = [train.remote(i) for i in range(num_parallel)]
​
# Join on all the writers to make sure their files have
# been added before finishing the artifact. 
ray.get(result_ids)
​
# Once all the writers are done writing, finish the artifact
# to mark it ready.
with wandb.init(group=group_name) as run:
  artifact = wandb.Artifact(artifact_name, type=artifact_type)
  
  # Create a "PartitionTable" pointing to the folder of tables
  # and add it to the artifact.
  artifact.add(wandb.data_types.PartitionedTable(parts_path), table_name)
  
  # Finish artifact finalizes the artifact, disallowing future "upserts"
  # to this version.
  run.finish_artifact(artifact)
Model Versioning
Artifact References
Last updated 2 months ago