​
module TrueFoundry.ml

​
Global Variables

  • TRACKING_HOST_GLOBAL

​
function get_client

Initializes and returns the truefoundry client. Args:
  • disable_analytics (bool, optional): To turn off usage analytics collection, pass True. By default, this is set to False.
Returns:
  • MLFoundry: Instance of MLFoundry class which represents a run.
Examples:
  • Get client
import truefoundry.ml as tfm

client = tfm.get_client()

​
class MlFoundry

MlFoundry.

​
function create_data_directory

Create DataDirectory to Upload the files Args:
  • ml_repo (str): Name of the ML Repo in which you want to create data_directory
  • name (str): Name of the DataDirectory to be created.
  • description (str): Description of the Datset
  • metadata (Dict <str>: Any): Metadata about the data_directory in Dictionary form.
Returns:
  • DataDirectory : An instance of class DataDirectory
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
data_directory = client.create_data_directory(name="<data_directory-name>", ml_repo="<repo-name>")
print(data_directory.fqn)

​
function create_ml_repo

Creates an ML Repository. Args:
  • name (str, optional): The name of the Repository you want to create. if not given, it creates a name by itself.
  • storage_integration_fqn (str): The storage integration FQN to use for the experiment for saving artifacts.
Examples:
  • Create Repository
import truefoundry.ml as tfm

client = tfm.get_client()

ml_repo = client.create_ml_repo(name="my-repo", storage_integration_fqn="truefoundry:example:name:blob-storage:blob")

​
function create_run

Initialize a run. In a machine learning experiment run represents a single experiment conducted under a project. Args:
  • ml_repo (str): The name of the project under which the run will be created. ml_repo should only contain alphanumerics (a-z,A-Z,0-9) or hyphen (-). The user must have ADMIN or WRITE access to this project.
  • run_name (Optional[str], optional): The name of the run. If not passed, a randomly generated name is assigned to the run. Under a project, all runs should have a unique name. If the passed run_name is already used under a project, the run_name will be de-duplicated by adding a suffix. run name should only contain alphanumerics (a-z,A-Z,0-9) or hyphen (-).
  • tags (Optional[Dict[str, Any]], optional): Optional tags to attach with this run. Tags are key-value pairs.
kwargs: Returns:
  • MlFoundryRun: An instance of MlFoundryRun class which represents a run.
Examples:
  • Create a run under current user.
import truefoundry.ml as tfm

client = tfm.get_client()

tags = {"model_type": "svm"}
run = client.create_run(
     ml_repo="my-classification-project", run_name="svm-with-rbf-kernel", tags=tags
)

run.end()
  • Creating a run using context manager.
import truefoundry.ml as tfm

client = tfm.get_client()
with client.create_run(
     ml_repo="my-classification-project", run_name="svm-with-rbf-kernel"
) as run:
     # ...
     # Model training code
     ...
# `run` will be automatically marked as `FINISHED` or `FAILED`.
  • Create a run in a project owned by a different user.
import truefoundry.ml as tfm

client = tfm.get_client()

tags = {"model_type": "svm"}
run = client.create_run(
     ml_repo="my-classification-project",
     run_name="svm-with-rbf-kernel",
     tags=tags,
)
run.end()

​
function get_all_runs

Returns all the run name and id present under a project. The user must have READ access to the project. Args:
  • ml_repo (str): Name of the project.
Returns:
  • pd.DataFrame: dataframe with two columns- run_id and run_name
Examples:
  • get all the runs from a ml_repo
import truefoundry.ml as tfm

client = tfm.get_client()

run = client.get_all_runs(ml_repo='my-repo')

​
function get_artifact_version

Get the model version to download contents or load it in memory Args:
  • ml_repo (str): ML Repo to which artifact is logged
  • artifact_name (str): Artifact Name
  • artifact_type (str): The type of artifact to fetch (acceptable values: “artifact”, “model”, “plot”, “image”)
  • version (str | int): Artifact Version to fetch (default is the latest version)
Returns:
  • ArtifactVersion : An ArtifactVersion instance of the artifact
Examples: 
import tempfile
import truefoundry.ml as tfm

client = tfm.get_client()
model_version = client.get_artifact_version(ml_repo="ml-repo-name", name="artifact-name", version=1)

# load the model into memory
clf = model_version.load()

# download the model to disk
temp = tempfile.TemporaryDirectory()
download_info = model_version.download(path=temp.name)
print(download_info)

​
function get_artifact_version_by_fqn

Get the artifact version to download contents Args:
  • fqn (str): Fully qualified name of the artifact version.
Returns:
  • ArtifactVersion : An ArtifactVersion instance of the artifact
Examples: 
import tempfile
import truefoundry.ml as tfm

client = tfm.get_client()
artifact_version = client.get_artifact_version_by_fqn(
     fqn="artifact:truefoundry/my-classification-project/sklearn-artifact:1"
)

# download the artifact to disk
temp = tempfile.TemporaryDirectory()
download_info = artifact_version.download(path=temp.name)
print(download_info)

​
function get_data_directory

Get an existing data_directory by name. Args:
  • ml_repo (str): name of an the project of which the data-directory is part of.
  • name (str): the name of the data-directory
Returns:
  • DataDirectory: An instance of class DataDirectory
Examples: 
import truefoundry.ml as tfm
client = tfm.get_client()
data_directory = client.get_data_directory(ml_repo='my-repo', name="<data-directory-name>")
with open("artifact.txt", "w") as f:
     f.write("hello-world")
data_directory.add_files(
     artifact_paths=[tfm.DataDirectoryPath('artifact.txt', 'a/b/')]
)
# print the path of files and folder in the data_directory
for file in data_directory.list_files():
     print(file.path)

​
function get_data_directory_by_fqn

Get the DataDirectory by DataDirectory FQN Args:
  • fqn (str): Fully qualified name of the artifact version.
Returns:
  • DataDirectory : An instance of class DataDirectory
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
data_directory = client.get_data_directory_by_fqn(fqn="<data-dir-fqn>")
with open("artifact.txt", "w") as f:
     f.write("hello-world")

data_directory.add_files(
     artifact_paths=[tfm.DataDirectoryPath('artifact.txt', 'a/b/')]
)
# print the path of files and folder in the data_directory
for file in data_directory.list_files():
     print(file.path)

​
function get_data_directory_by_id

Get the DataDirectory From the DataDirectory ID Args:
  • id (uuid.UUID): Id of the data_directory.
Returns:
  • DataDirectory : An instance of class DataDirectory
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
data_directory = client.get_data_directory_by_id(id="<data_directory-id>")
with open("artifact.txt", "w") as f:
     f.write("hello-world")

data_directory.add_files(
     artifact_paths=[tfm.DataDirectoryPath('artifact.txt', 'a/b/')]
)
# print the path of files and folder in the data_directory
for file in data_directory.list_files():
     print(file.path)

​
function get_model_version

Get the model version to download contents or load it in memory Args:
  • ml_repo (str): ML Repo to which model is logged
  • name (str): Model Name
  • version (str | int): Model Version to fetch (default is the latest version)
Returns:
  • ModelVersion: The ModelVersion instance of the model.
Examples:
  • Sklearn
# See `truefoundry.ml.mlfoundry_api.MlFoundry.log_model` examples to understand model logging
import tempfile
import joblib
import truefoundry.ml as tfm

client = tfm.get_client()
model_version = client.get_model_version(
    ml_repo="my-classification-project",
    name="my-sklearn-model",
    version=1
)

# Download the model to disk
temp = tempfile.TemporaryDirectory()
download_info = model_version.download(path=temp.name)
print(download_info.model_dir, download_info.model_filename)

# Deserialize and Load
model = joblib.load(
    os.path.join(download_info.model_dir, download_info.model_filename)
)
  • Huggingface Transformers
# See `truefoundry.ml.mlfoundry_api.MlFoundry.log_model` examples to understand model logging
import torch
from transformers import pipeline

import truefoundry.ml as tfm

client = tfm.get_client()
model_version = client.get_model_version(
    ml_repo="my-llm-project",
    name="my-transformers-model",
    version=1
)

# Download the model to disk
temp = tempfile.TemporaryDirectory()
download_info = model_version.download(path=temp.name)
print(download_info.model_dir)

# Deserialize and Load
pln = pipeline("text-generation", model=download_info.model_dir, torch_dtype=torch.float16)

​
function get_model_version_by_fqn

Get the model version to download contents or load it in memory Args:
  • fqn (str): Fully qualified name of the model version.
Returns:
  • ModelVersion: The ModelVersion instance of the model.
Examples:
  • Sklearn
# See `truefoundry.ml.mlfoundry_api.MlFoundry.log_model` examples to understand model logging
import tempfile
import joblib
import truefoundry.ml as tfm

client = tfm.get_client()
model_version = client.get_model_version_by_fqn(
     fqn="model:truefoundry/my-classification-project/my-sklearn-model:1"
)

# Download the model to disk
temp = tempfile.TemporaryDirectory()
download_info = model_version.download(path=temp.name)
print(download_info.model_dir, download_info.model_filename)

# Deserialize and Load
model = joblib.load(
     os.path.join(download_info.model_dir, download_info.model_filename)
)
  • Huggingface Transformers
# See `mlfoundry.mlfoundry_api.MlFoundry.log_model` examples to understand model logging
import torch
from transformers import pipeline

import truefoundry.ml as tfm

client = tfm.get_client()
model_version = client.get_model_version_by_fqn(
     fqn="model:truefoundry/my-llm-project/my-transformers-model:1"
)
# Download the model to disk
temp = tempfile.TemporaryDirectory()
download_info = model_version.download(path=temp.name)
print(download_info.model_dir)

# Deserialize and Load
pln = pipeline("text-generation", model=download_info.model_dir, torch_dtype=torch.float16)

​
function get_run_by_fqn

Get an existing run by fqn. fqn stands for Fully Qualified Name. A run fqn has the following pattern: tenant_name/ml_repo/run_name If a run svm under the project cat-classifier in truefoundry tenant, the fqn will be truefoundry/cat-classifier/svm. Args:
  • run_fqn (str): fqn of an existing run.
Returns:
  • MlFoundryRun: An instance of MlFoundryRun class which represents a run.
Examples:
  • get run by run fqn
import truefoundry.ml as tfm

client = tfm.get_client()

run = client.get_run_by_fqn(run_fqn='truefoundry/my-repo/svm')

​
function get_run_by_id

Get an existing run by the run_id. Args:
  • run_id (str): run_id or fqn of an existing run.
Returns:
  • MlFoundryRun: An instance of MlFoundryRun class which represents a run.
Examples:
  • Get run by the run id
import truefoundry.ml as tfm

client = tfm.get_client()

run = client.get_run_by_id(run_id='a8f6dafd70aa4baf9437a33c52d7ee90')

​
function get_run_by_name

Get an existing run by run_name. Args:
  • ml_repo (str): name of the ml_repo of which the run is part of.
  • run_name (str): the name of the run required
Returns:
  • MlFoundryRun: An instance of MlFoundryRun class which represents a run.
Examples:
  • get run by name
import truefoundry.ml as tfm

client = tfm.get_client()

run = client.get_run_by_name(run_name='svm', ml_repo='my-repo')

​
function get_tracking_uri

Get the current tracking URI. Returns:
The tracking URI. Examples: 
import tempfile
import truefoundry.ml as tfm

client = tfm.get_client()
tracking_uri = client.get_tracking_uri()
print("Current tracking uri: {}".format(tracking_uri))

​
function list_artifact_versions

Get all the version of na artifact to download contents or load them in memory Args:
  • ml_repo (str): Repository in which the model is stored.
  • name (str): Name of the artifact whose version is required
  • artifact_type (ArtifactType): Type of artifact you want for example model, image, etc.
Returns:
  • Iterator[ArtifactVersion]: An iterator that yields non deleted artifact-versions of an artifact under a given ml_repo sorted reverse by the version number
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
artifact_versions = client.list_artifact_versions(ml_repo="my-repo", name="artifact-name")

for artifact_version in artifact_versions:
     print(artifact_version)

​
function list_artifact_versions_by_fqn

List versions for a given artifact Args:
  • artifact_fqn: FQN of the Artifact to list versions for.
  • An artifact_fqn looks like {artifact_type}`: {org}/{user}/{project}/{artifact_name}`
  • or {artifact_type}`: {user}/{project}/{artifact_name}`
where artifact_type can be on of (“model”, “image”, “plot”) Returns:
  • Iterator[ArtifactVersion]: An iterator that yields non deleted artifact versions under the given artifact_fqn sorted reverse by the version number
Yields:
  • ArtifactVersion: An instance of mlfoundry.ArtifactVersion
Examples: 
import truefoundry.ml as tfm

tfm.login(tracking_uri=https://your.truefoundry.site.com")
client = tfm.get_client()
artifact_fqn = "artifact:org/my-project/my-artifact"
for av in client.list_artifact_versions(artifact_fqn=artifact_fqn):
     print(av.name, av.version, av.description)

​
function list_data_directories

Get the list of DataDirectory in a ml_repo Args:
  • ml_repo (str): Name of the ML Repository
  • max_results (int): Maximum number of Data Directory to list
  • offset (int): Skip these number of instance of DataDirectory and then give the result from these number onwards
Returns:
  • DataDirectory : An instance of class DataDirectory
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
data_directories = client.list_data_directories(ml_repo="<ml-repo-nam>")

for data_directory in data_directories:
     print(data_directory.name)

​
function list_ml_repos

Returns a list of names of ML Repos accessible by the current user. Returns:
  • List[str]: A list of names of ML Repos

​
function list_model_versions

Get all the version of a model to download contents or load them in memory Args:
  • ml_repo (str): Repository in which the model is stored.
  • name (str): Name of the model whose version is required
Returns:
  • Iterator[ModelVersion]: An iterator that yields non deleted model versions of a model under a given ml_repo sorted reverse by the version number
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
model_versions = client.list_model_version(ml_repo="my-repo", name="svm")

for model_version in model_versions:
     print(model_version)

​
function list_model_versions_by_fqn

List versions for a given model Args:
  • model_fqn: FQN of the Model to list versions for.
  • A model_fqn looks like model`: {org}/{user}/{project}/{artifact_name}`
  • or model`: {user}/{project}/{artifact_name}`
Returns:
  • Iterator[ModelVersion]: An iterator that yields non deleted model versions under the given model_fqn sorted reverse by the version number
Yields:
  • ModelVersion: An instance of mlfoundry.ModelVersion
Examples: 
import truefoundry.ml as tfm

mlfoundry.login(tracking_uri="https://your.truefoundry.site.com")
client = tfm.get_client()
model_fqn = "model:org/my-project/my-model"
for mv in client.list_model_versions(model_fqn=model_fqn):
     print(mv.name, mv.version, mv.description)

​
function log_artifact

Logs an artifact for the current ml_repo. An artifact is a list of local files and directories. This function packs the mentioned files and directories in artifact_paths and uploads them to remote storage linked to the ml_repo Args:
  • ml_repo (str): Name of the ML Repo to which an artifact is to be logged.
  • name (str): Name of the Artifact. If an artifact with this name already exists under the current ml_repo, the logged artifact will be added as a new version under that name. If no artifact exist with the given name, the given artifact will be logged as version 1.
  • artifact_paths (List[truefoundry.ml.ArtifactPath], optional): A list of pairs of (source path, destination path) to add files and folders to the artifact version contents. The first member of the pair should be a file or directory path and the second member should be the path inside the artifact contents to upload to.
         E.g. >>> client.log_artifact(
              ...     ml_repo="sample-repo",
              ...     name="xyz",
              ...     artifact_paths=[
                         tfm.ArtifactPath("foo.txt", "foo/bar/foo.txt"),
                         tfm.ArtifactPath("tokenizer/", "foo/tokenizer/"),
                         tfm.ArtifactPath('bar.text'),
                         ('bar.txt', ),
                         ('foo.txt', 'a/foo.txt')
                      ]
              ... )
         would result in
         .
         └── foo/
             ├── bar/
             │   └── foo.txt
             └── tokenizer/
                 └── # contents of tokenizer/ directory will be uploaded here
  • description (Optional[str], optional): arbitrary text upto 1024 characters to store as description. This field can be updated at any time after logging. Defaults to None
  • metadata (Optional[Dict[str, Any]], optional): arbitrary json serializable dictionary to store metadata. For example, you can use this to store metrics, params, notes. This field can be updated at any time after logging. Defaults to None
Returns:
  • truefoundry.ml.ArtifactVersion: an instance of ArtifactVersion that can be used to download the files, or update attributes like description, metadata.
Examples: 
import os
import truefoundry.ml as tfm

with open("artifact.txt", "w") as f:
     f.write("hello-world")

client = tfm.get_client()
ml_repo = "sample-repo"

client.create_ml_repo(ml_repo=ml_repo)
client.log_artifact(
     ml_repo=ml_repo,
     name="hello-world-file",
     artifact_paths=[tfm.ArtifactPath('artifact.txt', 'a/b/')]
)

​
function log_model

Serialize and log a versioned model under the current ml_repo. Each logged model generates a new version associated with the given name and linked to the current run. Multiple versions of the model can be logged as separate versions under the same name. Args:
  • ml_repo (str): Name of the ML Repo to which an artifact is to be logged.
  • name (str): Name of the model. If a model with this name already exists under the current ML Repo, the logged model will be added as a new version under that name. If no models exist with the given name, the given model will be logged as version 1.
  • model_file_or_folder (str): Path to either a single file or a folder containing model files. This folder is usually created using serialization methods of libraries or frameworks e.g. joblib.dump, model.save_pretrained(...), torch.save(...), model.save(...)
  • framework (Union[enums.ModelFramework, str]): Model Framework. Ex:- pytorch, sklearn, tensorflow etc. The full list of supported frameworks can be found in mlfoundry.enums.ModelFramework. Can also be None when model is None.
  • description (Optional[str], optional): arbitrary text upto 1024 characters to store as description. This field can be updated at any time after logging. Defaults to None
  • metadata (Optional[Dict[str, Any]], optional): arbitrary json serializable dictionary to store metadata. For example, you can use this to store metrics, params, notes. This field can be updated at any time after logging. Defaults to None
Returns:
  • truefoundry.ml.ModelVersion: an instance of ModelVersion that can be used to download the files, load the model, or update attributes like description, metadata, schema.
Examples:
  • Sklearn
import truefoundry.ml as tfm
from truefoundry.ml import ModelFramework

import joblib
import numpy as np
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVC

X = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]])
y = np.array([1, 1, 2, 2])
clf = make_pipeline(StandardScaler(), SVC(gamma='auto'))
clf.fit(X, y)
joblib.dump(clf, "sklearn-pipeline.joblib")

client = tfm.get_client()
client.create_ml_repo(  # This is only required once
     ml_repo="my-classification-project",
     # This controls which bucket is used.
     # You can get this from Integrations > Blob Storage. `None` picks the default
     storage_integration_fqn=None
)
model_version = client.log_model(
     ml_repo="my-classification-project",
     name="my-sklearn-model",
     model_file_or_folder="sklearn-pipeline.joblib",
     framework=ModelFramework.SKLEARN,
     metadata={"accuracy": 0.99, "f1": 0.80},
     step=1,  # step number, useful when using iterative algorithms like SGD
)
print(model_version.fqn)
  • Huggingface Transformers
import truefoundry.ml as tfm
from truefoundry.ml import ModelFramework

import torch
from transformers import AutoTokenizer, AutoConfig, pipeline, AutoModelForCausalLM
pln = pipeline(
     "text-generation",
     model_file_or_folder="EleutherAI/pythia-70m",
     tokenizer="EleutherAI/pythia-70m",
     torch_dtype=torch.float16
)
pln.model.save_pretrained("my-transformers-model")
pln.tokenizer.save_pretrained("my-transformers-model")

client = tfm.get_client()
client.create_ml_repo(  # This is only required once
     ml_repo="my-llm-project",
     # This controls which bucket is used.
     # You can get this from Integrations > Blob Storage. `None` picks the default
     storage_integration_fqn=None
)
model_version = client.log_model(
     ml_repo="my-llm-project",
     name="my-transformers-model",
     model_file_or_folder="my-transformers-model/",
     framework=ModelFramework.TRANSFORMERS
)
print(model_version.fqn)

​
function search_runs

The user must have READ access to the project. Returns an iterator that returns a MLFoundryRun on each next call. All the runs under a project which matches the filter string and the run_view_type are returned. Args:
  • ml_repo (str): Name of the project. filter_string (str, optional): Filter query string, defaults to searching all runs. Identifier required in the LHS of a search expression.
  • Signifies an entity to compare against. An identifier has two parts separated by a period: the type of the entity and the name of the entity. The type of the entity is metrics, params, attributes, or tags. The entity name can contain alphanumeric characters and special characters.
  • You can search using two run attributes : status and artifact_uri. Both attributes have string values. When a metric, parameter, or tag name contains a special character like hyphen, space, period, and so on, enclose the entity name in double quotes or backticks, params.”model-type” or params.model-type
  • run_view_type (str, optional): one of the following values “ACTIVE_ONLY”, “DELETED_ONLY”, or “ALL” runs. order_by (List[str], optional): List of columns to order by (e.g., “metrics.rmse”). Currently supported values are metric.key, parameter.key, tag.key, attribute.key. The order_by column can contain an optional DESC or ASC value. The default is ASC. The default ordering is to sort by start_time DESC.
  • job_run_name (str): Name of the job which are associated with the runs to get that runs
  • max_results (int): max_results on the total numbers of run yielded through filter
Returns:
  • Iterator[MlFoundryRun]: MLFoundryRuns matching the search query.
Examples: 
import truefoundry.ml as tfm

client = tfm.get_client()
with client.create_run(ml_repo="my-project", run_name="run-1") as run1:
     run1.log_metrics(metric_dict={"accuracy": 0.74, "loss": 0.6})
     run1.log_params({"model": "LogisticRegression", "lambda": "0.001"})

with client.create_run(ml_repo="my-project", run_name="run-2") as run2:
     run2.log_metrics(metric_dict={"accuracy": 0.8, "loss": 0.4})
     run2.log_params({"model": "SVM"})

# Search for the subset of runs with logged accuracy metric greater than 0.75
filter_string = "metrics.accuracy > 0.75"
runs = client.search_runs(ml_repo="my-project", filter_string=filter_string)

# Search for the subset of runs with logged accuracy metric greater than 0.7
filter_string = "metrics.accuracy > 0.7"
runs = client.search_runs(ml_repo="my-project", filter_string=filter_string)

# Search for the subset of runs with logged accuracy metric greater than 0.7 and model="LogisticRegression"
filter_string = "metrics.accuracy > 0.7 and params.model = 'LogisticRegression'"
runs = client.search_runs(ml_repo="my-project", filter_string=filter_string)

# Search for the subset of runs with logged accuracy metric greater than 0.7 and
# order by accuracy in Descending order
filter_string = "metrics.accuracy > 0.7"
order_by = ["metric.accuracy DESC"]
runs = client.search_runs(
     ml_repo="my-project", filter_string=filter_string, order_by=order_by
)

filter_string = "metrics.accuracy > 0.7"
runs = client.search_runs(
     ml_repo="transformers", order_by=order_by ,job_run_name='job_run_name', filter_string=filter_string
)

order_by = ["metric.accuracy DESC"]
runs = client.search_runs(
     ml_repo="my-project", filter_string=filter_string, order_by=order_by, max_results=10
)