Skip to main content

Transformer Lab SDK

The Transformer Lab Python SDK provides a way for ML scripts to integrate with Transformer Lab.

Install

pip install transformerlab-internal

Usage

from lab import lab

# Initialize with experiment ID
lab.init("my-experiment")
lab.log("Job initiated")

config_artifact_path = lab.save_artifact(<config_file>, "training_config.json")
lab.log(f"Saved training config: {config_artifact_path}")
lab.update_progress(1)

...
lab.update_progress(99)

model_path = lab.save_model(<training_output_dir>, name="trained_model")
lab.log("Saved model file to {model_path}")

lab.finish("Training completed successfully")

Shared storage

Each user has a personal area of the team workspace for large files reused across experiments and jobs — a base model, a tokenized corpus, a reference checkpoint. Upload once from the terminal with lab storage upload, then read it from any job:

from lab import lab

corpus = lab.storage_download("tokenized-corpus")                 # directory → local dir
weights = lab.storage_download("models/base-model.safetensors")   # file → local path

for entry in lab.storage_list("models"):
    print(entry["relpath"], entry["size"])

lab.storage_upload("./derived-index.faiss", "indexes/wiki.faiss")

Downloads always land on the node's own disk — the returned path is a file you can open. By default they go to a cache under ~/.transformerlab/cache/user_storage/ (pass dest= to choose a location) and read straight from object storage, so a multi-GB file moves at full bandwidth without routing through the API server. A local copy whose size already matches is reused, so re-running a job does not re-fetch it; pass force=True to override.

Every path is relative to your own storage root. Async variants exist for all three (async_storage_download, async_storage_upload, async_storage_list).

Another job's artifacts

A job can read what an earlier job saved with lab.save_artifact, for example to evaluate an adapter a training job produced:

from lab import lab

for entry in lab.list_job_artifacts("1234"):               # same experiment
    print(entry["name"], entry["size"])

adapter = lab.download_job_artifact("1234", "adapter")    # directory → local dir
report = lab.download_job_artifact("1234", "eval.json", experiment_id="sweeps")

Downloads behave like storage_download: they land on local disk (cached under ~/.transformerlab/cache/job_artifacts/ unless you pass dest=), and a copy of the right size is reused unless you pass force=True.

Which jobs you can reach follows the team's experiment privacy setting. Jobs in the current experiment are always reachable. When the team keeps experiments private, another experiment's jobs are reachable only if you created that experiment, and experiments with no recorded creator are refused. Either case raises lab.job_artifacts.JobArtifactAccessError. Async variants are async_list_job_artifacts and async_download_job_artifact.

Reporting metrics

Pass a dict of named metrics to lab.finish(score=...). The dict shape is required — metrics are stored under job_data.score and surfaced by lab job list (Score column) and lab job info, and read by sweep / autoresearch flows for optimization.

lab.finish(message="Done!")                              # success, no score
lab.finish(message="Done!", score={"accuracy": 0.78})    # success with one metric
lab.finish(score={"accuracy": 0.78, "f1": 0.83})         # multiple metrics

Do not pass a scalar (e.g. lab.finish(score=0.78)) — wrap it in a dict instead: lab.finish(score={"score": 0.78}).

Sample scripts can be found at https://github.com/transformerlab/transformerlab-platform/tree/main/lab-sdk/scripts/examples

Development

The code for this can be found in the lab-sdk directory of https://github.com/transformerlab/transformerlab-platform

To develop locally in editable mode and run automated tests:

cd lab-sdk
uv venv
uv pip install -e .
uv run pytest  # Run tests

Metadata

Release files for transformerlab-internal 0.1.66

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for transformerlab-internal 0.1.66
File Size Uploaded Provenance Actions
transformerlab_internal-0.1.66.tar.gz 131.7 kB Without provenance Details for transformerlab_internal-0.1.66.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for transformerlab-internal 0.1.66
File Interpreter ABI Platform Actions
transformerlab_internal-0.1.66-py3-none-any.whl Python 3 none any Details for transformerlab_internal-0.1.66-py3-none-any.whl

Total release size: 232.8 kB

Release files / transformerlab_internal-0.1.66.tar.gz

Download URL transformerlab_internal-0.1.66.tar.gz
Size 131.7 kB
Tags Source
SHA-256 checksum
How to use checksums
8049965aa05207ee987e4908e13cb263315462d90e55b09cd16211278f77796b
BLAKE2b-256 checksum
How to use checksums
f11cdf386751f4072d325bf8b8a0c0f781e80d8ce806abed43beaf07d3af843a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.13.16

Release files / transformerlab_internal-0.1.66-py3-none-any.whl

Download URL transformerlab_internal-0.1.66-py3-none-any.whl
Size 101.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
9e62f3bef72a767628a24652cfe9522b66a551412e59e8e7d1d683c4fb32c4f8
BLAKE2b-256 checksum
How to use checksums
e4f3a6f00f326ba54ad84a674344645ee2f4fda1cfa6bdaf3b23613f45c7a63f
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.13.16

Security for transformerlab-internal 0.1.66

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page