JupyterLab®

Code Workspaces enables you to use JupyterLab® ↗ in Foundry. JupyterLab® in Code Workspaces supports:

• Reading and writing tabular datasets, Iceberg tables, and virtual tables.
• Downloading or uploading files from non-structured datasets.
• Importing Python libraries from conda or PyPI.
• Previewing Dash, Streamlit, or TensorBoard™ applications to facilitate iteration.
• Publishing Dash or Streamlit applications.
• Interacting with the Ontology.

Code Workspaces currently supports Dash ↗ and Streamlit ↗ for Python applications. Users can create applications directly in Code Workspaces with Foundry’s version control, branching, and data governance features built-in.

:::callout{theme="neutral"} Code Workspace applications support branching. If you create a new Workspace branch, publishing a new application or synchronizing the changes will publish a new version of the application on that branch. This allows you to preview your application before exposing it to your users. To publish on the master branch, simply merge your branch into master. :::

Dash applications

:::callout{theme="warning"} Published Dash applications are subject to a 30-second timeout, meaning that the Dash server must start up within 30 seconds of executing your Dash application Python file. Otherwise, your Dash application will fail to start up. :::

To create a Dash ↗ application, follow the instructions below:

1. Open the Applications tab in a JupyterLab® workspace.
2. Select Publish application, choose a location in your Files and Projects for your new application, and provide a module name and variable name. A file will be automatically created at the path corresponding to the provided module name.
3. Optional: Configure advanced settings. By default, the application settings will match the settings of the Jupyter® workspace.
4. Select Publish and sync to register your new Dash application and sync the code to the backing code repository. After CI checks and publishing are complete, you can select the link in the Applications panel to view the published application.

:::callout{theme="neutral"} The module name of the Dash application should be entered as a dotted path, relative to the root of the repository (for example, app, or if your module is inside a folder,folder.app). The variable name is the Python variable within the module file that is declared for the Dash app server, which is usually called server. :::

To develop locally against an existing Dash application:

1. Open a terminal window and execute your newly created file, for example, python app.py.
2. Select Preview in the Dash section of the Applications tab to preview the application.

:::callout{theme="success"} You can ensure the conda/PyPI environment in your JupyterLab® workspace will be automatically restored in your published Dash application using managed conda/PyPI environments in Jupyter® Notebooks. :::

Streamlit applications

:::callout{theme="warning"} Published Streamlit applications are subject to a 30-second timeout, meaning that the Streamlit server must start up within 30 seconds of executing your Streamlit application Python file. Otherwise, your published Streamlit application will fail to start up. :::

To create a Streamlit ↗ application, follow the instructions below:

1. Select Publish application, choose a location in your Files and Projects for your new application, and enter the name of your Python file relative to the root of the repository, for example, app.py.
2. Optional: Configure advanced settings. By default, the application settings will match the settings of the JupyterLab® workspace.
3. Select Publish and sync to register your new Streamlit application and sync the code to the backing code repository. After CI checks and publishing are complete, you can select the link in the Applications panel to view the published application.

To develop locally against an existing Streamlit application:

1. Open a terminal window and run the code with Streamlit using streamlit run app-name.py.
2. Open the Applications tab in a JupyterLab® code workspace and select Preview in the Streamlit section to preview the application you are developing locally.

:::callout{theme="success"} You can ensure the conda/PyPI environment in your JupyterLab® workspace will be automatically restored in your published Streamlit application using managed conda/PyPI environments in Jupyter® Notebooks. :::

TensorBoard™

You can preview TensorBoard™ ↗ directly from Code Workspaces. To do this, run tensorboard --logdir=<$PATH_TO_LOGS_DIRECTORY> from a terminal, filling in the path to the directory containing TensorFlow logs. Then, open the Applications tab in a Jupyter® Workspace and select Preview in the TensorBoard™ section.

Jupyter® Notebook transforms

:::callout{theme="warning"} Jupyter Notebook transforms currently do not support making API calls to external services, even if the network policy was added in the code workspace. To run reproducible pipelines making API calls to external systems, you should use external transforms instead. :::

Code Workspaces can publish Jupyter® notebooks as data transformation pipelines and register transform JobSpecs for output datasets produced by the notebooks.

Follow the steps below to register a transform for an output dataset:

1. Save your Jupyter® notebook as a file in the workspace.

2. Select the blue Build icon next to the dataset in the Data panel.

1. Follow the interface to configure your transform. Code Workspaces will infer the Jupyter® notebook used along with the inputs added to produce the output dataset. If the inference is wrong, you can manually edit the inputs and outputs.

1. Select Sync and run to save the transform configuration and trigger CI checks for publishing the transform and JobSpecs on the output dataset(s).

Once CI checks are completed, your transform is ready for build. You can use other Foundry data integration tools to manage your transform and the data pipeline to which it belongs.

:::callout{theme="neutral"} The transform does not automatically use the environment of the Code Workspace. You should install all the libraries needed at runtime at the top of the script file. The code snippet to install a library can be found in the Libraries panel. For example, to use scikit-learn and scikit-image in your transform, add this code snippet to the top of the file:

!mamba install -y -q scikit-learn scikit-image

:::

Managed conda/PyPI environments in Jupyter® Notebooks

Jupyter® conda/PyPI environments can be consistently restored across Jupyter® workspaces, applications, transforms, and model outputs, using the managed environment feature backed by Maestro. Maestro is a Foundry library management tool that lets you install and manage libraries for Jupyter workspaces. It integrates with conda and pip, making library installation easier and helping maintain consistent environments. To convert commands to Maestro, refer to the instructions below for either conda or PyPI.

Managed environments are consistent across Code Workspace workflows, eliminating the need to re-run the same mamba and pip commands every time you restart your workspace, start an application, or run a Jupyter® Notebook transform. With a managed environment, installed libraries in your working environment are tracked through files in your /home/user/repo folder, allowing the environment to be restored to its original state across workflows.

The following two files are used to track your managed environment:

1. /home/user/repo/.envs/maestro/meta.yaml: Known as the environment manifest, it contains the set of requested libraries and versions for your environment. Learn more about the meta.yaml file.
2. /home/user/repo/.envs/maestro/hawk.lock - Known as the environment lock file, it contains the set of resolved libraries and versions for your environment, solved from the constraints specified in the manifest. The hawk.lock file replaces conda.lock, which has been deprecated. Learn more about the hawk.lock file.

:::callout{theme="warning"} We do not recommend modifying the hawk.lock lock file. Instead, rely on installation commands to update the files on your behalf. Unless there is a git conflict, you do not need to modify the meta.yaml file either; it will automatically update when libraries are interactively installed. :::

Your managed environment is installed in your /home/user/envs/default folder. This environment is restored before the following stages:

• Starting up your Jupyter® workspace
• Starting up applications produced from your workspace
• Running Jupyter® Notebook transforms produced from your workspace
• Running models

Usage

1. Enable managed environments

To enable this feature, toggle the Managed Conda environments advanced feature in your workspace settings, located on the left sidebar. Then, restart the workspace:

Once the feature is enabled, a managed environment named default will be available for you to use:

2. Update your managed environment

Follow the steps below to update your managed environment and manage libraries.

Manage conda libraries with Maestro

• Install using the Libraries panel: Search for the library name (for example, numpy) and install.

• Install using the terminal: To install the latest version of a library, run the following command in the terminal:

maestro env conda install <library_name>

To install a specific version, run the following command:

maestro env conda install <library_name>==<library_version>

• Uninstall using the terminal: To uninstall a library, run the following command in the terminal:

maestro env uninstall --conda-dependencies <library_names>

Manage PyPI libraries with Maestro

• Install using the Libraries panel: Search for the library name (for example, numpy) and install.

• Install using the terminal: To install the latest version of a library, run the following command in the terminal:

maestro env pip install <library_name>

To install a specific version, run the following command:

maestro env pip install <library_name>==<library_version>

• Uninstall using the terminal: To uninstall a library, run the following command in the terminal:

maestro env uninstall --pip-dependencies <library_names>

:::callout{theme="warning"} When installing a library through the Libraries tab, the install command is executed directly in the Jupyter® terminal, rather than executing it within your notebook or application code. Every time an install command is run, the meta.yaml manifest will be updated with your requested library, and the hawk.lock file will reflect the libraries installed in the current environment. :::

3. View currently installed libraries in your managed environment

Installed libraries are visible in the Libraries tab:

4. View environment files in the JupyterLab® interface

Navigate to the View dropdown menu in the workspace and select Show Hidden Files.

Then, navigate to the .envs/maestro folder to view the meta.yaml and hawk.lock files.

5. Synchronize your managed environment with changes to the hawk.lock file

After checking out a new branch or pulling new git changes, the contents of your hawk.lock file may change.

To synchronize your managed environment, execute the following command in your workspace terminal:

maestro env install

:::callout{theme="neutral"} maestro is the command line interface that updates managed environments. The above command re-resolves the lock file from the manifest and installs the environment into the /home/user/envs/default folder. :::

6. Resolve git conflicts to your environment files

To resolve git conflicts in your managed environment, first delete the hawk.lock file. Navigate to the meta.yaml file to resolve any conflicts, then merge your branch.

After merging, run the maestro env install command in your terminal to regenerate the hawk.lock file and reinstall your managed environment.

7. Upgrade your managed environment

To update your managed environment, first remove the hawk.lock file. Then, delete your /home/user/envs/default folder (rm -rf /home/user/envs/default). Finally, run the maestro env install command in your terminal to regenerate your hawk.lock file. Your environment will be re-installed with upgraded library versions.

Example: meta.yaml file

The meta.yaml file represents a specification of requested libraries. This file conforms to the Conda metadata specification ↗, and can contain specifications for both conda and PyPI dependencies. Only libraries specified in the requirements.run and requirements.pip section are used for Jupyter® managed environments.

Below is an example of a meta.yaml file:

package:
  name: '{{ PACKAGE_NAME }}'
  version: '{{ PACKAGE_VERSION }}'
requirements:
  build: []
  run:
    - ipykernel
    - pip
    - foundry-transforms-lib-python
    - pandas
    - polars
    - dash
    - streamlit
    - tensorboard
  pip:
    - Flask-Testing
  run_constrained: []
source:
  path: ../src
build: null
test: null
about: null

Example: hawk.lock file

The hawk.lock file contains the resolved libraries, versions, and channels of a solved managed environment.

Below is an example of a hawk.lock file:

# This file is autogenerated. Though not necessary, it can be committed to the git repository.
# To avoid build failures, we do not recommend changing this file,

[conda]
_libgcc_mutex=0.1=conda_forge @ 16fe1c80
_openmp_mutex=4.5=2_gnu @ 16fe1c80
absl-py=2.1.0=pyhd8ed1ab_0 @ c7211bb6
altair=5.3.0=pyhd8ed1ab_0 @ c7211bb6
asttokens=2.4.1=pyhd8ed1ab_0 @ c7211bb6
attrs=23.2.0=pyh71513ae_0 @ c7211bb6
aws-c-auth=0.7.22=h96bc93b_2 @ e1c11c02
aws-c-cal=0.6.14=h88a6e22_1 @ e1c11c02
aws-c-common=0.9.19=h4ab18f5_0 @ e1c11c02
aws-c-compression=0.2.18=h83b837d_6 @ e1c11c02
aws-c-event-stream=0.4.2=ha47c788_12 @ e1c11c02
aws-c-http=0.8.1=h29d6fba_17 @ e1c11c02
aws-c-io=0.14.8=h21d4f22_5 @ e1c11c02
aws-c-mqtt=0.10.4=h759edc4_4 @ e1c11c02
aws-c-s3=0.5.9=h594631b_3 @ e1c11c02
aws-c-sdkutils=0.1.16=h83b837d_2 @ e1c11c02
aws-checksums=0.1.18=h83b837d_6 @ e1c11c02
aws-crt-cpp=0.26.9=he3a8b3b_0 @ e1c11c02
aws-sdk-cpp=1.11.329=hba8bd5f_3 @ e1c11c02
blinker=1.8.2=pyhd8ed1ab_0 @ c7211bb6
brotli-python=1.1.0=py310hc6cd4ac_1 @ e1c11c02
bzip2=1.0.8=hd590300_5 @ e1c11c02
c-ares=1.28.1=hd590300_0 @ e1c11c02
ca-certificates=2024.7.4=hbcca054_0 @ e1c11c02
cachetools=5.3.3=pyhd8ed1ab_0 @ c7211bb6
certifi=2024.7.4=pyhd8ed1ab_0 @ c7211bb6
cffi=1.16.0=py310h2fee648_0 @ e1c11c02
charset-normalizer=3.3.2=pyhd8ed1ab_0 @ c7211bb6
click=8.1.7=unix_pyh707e725_0 @ c7211bb6
comm=0.2.2=pyhd8ed1ab_0 @ c7211bb6
conjure-python-client=2.8.0=pyhd8ed1ab_0 @ c7211bb6
dash=2.17.1=pyhd8ed1ab_0 @ c7211bb6
debugpy=1.8.2=py310h76e45a6_0 @ e1c11c02
decorator=5.1.1=pyhd8ed1ab_0 @ fe451c34
exceptiongroup=1.2.0=pyhd8ed1ab_2 @ c7211bb6
executing=2.0.1=pyhd8ed1ab_0 @ c7211bb6
flask=3.0.3=pyhd8ed1ab_0 @ c7211bb6
foundry-data-sidecar-api=0.626.0=py_0 @ fe451c34
foundry-transforms-lib-python=0.639.0=py_0 @ fe451c34
freetype=2.12.1=h267a509_2 @ e1c11c02
gflags=2.2.2=he1b5a44_1004 @ 16fe1c80
gitdb=4.0.11=pyhd8ed1ab_0 @ c7211bb6
gitpython=3.1.43=pyhd8ed1ab_0 @ c7211bb6
glog=0.7.1=hbabe93e_0 @ e1c11c02
grpcio=1.62.2=py310h1b8f574_0 @ e1c11c02
h2=4.1.0=pyhd8ed1ab_0 @ fe451c34
hpack=4.0.0=pyh9f0ad1d_0 @ fe451c34
hyperframe=6.0.1=pyhd8ed1ab_0 @ fe451c34
idna=3.7=pyhd8ed1ab_0 @ c7211bb6
importlib-metadata=7.2.1=pyha770c72_0 @ c7211bb6
importlib_metadata=7.2.1=hd8ed1ab_0 @ c7211bb6
importlib_resources=6.4.0=pyhd8ed1ab_0 @ c7211bb6
ipykernel=6.29.5=pyh3099207_0 @ c7211bb6
ipython=8.26.0=pyh707e725_0 @ c7211bb6
ipywidgets=8.1.3=pyhd8ed1ab_0 @ c7211bb6
itsdangerous=2.2.0=pyhd8ed1ab_0 @ c7211bb6
jedi=0.19.1=pyhd8ed1ab_0 @ c7211bb6
jinja2=3.1.4=pyhd8ed1ab_0 @ c7211bb6
jsonschema=4.23.0=pyhd8ed1ab_0 @ c7211bb6
jsonschema-specifications=2023.12.1=pyhd8ed1ab_0 @ c7211bb6
jupyter_client=8.6.2=pyhd8ed1ab_0 @ c7211bb6
jupyter_core=5.7.2=py310hff52083_0 @ e1c11c02
jupyterlab_widgets=3.0.11=pyhd8ed1ab_0 @ c7211bb6
keyutils=1.6.1=h166bdaf_0 @ 16fe1c80
krb5=1.21.3=h659f571_0 @ e1c11c02
lcms2=2.16=hb7c19ff_0 @ e1c11c02
ld_impl_linux-64=2.40=hf3520f5_7 @ e1c11c02
lerc=4.0.0=h27087fc_0 @ 16fe1c80
libabseil=20240116.2=cxx17_h59595ed_0 @ e1c11c02
libarrow=16.1.0=hcb6531f_6_cpu @ e1c11c02
libarrow-acero=16.1.0=hac33072_6_cpu @ e1c11c02
libarrow-dataset=16.1.0=hac33072_6_cpu @ e1c11c02
libarrow-substrait=16.1.0=h7e0c224_6_cpu @ e1c11c02
libblas=3.9.0=22_linux64_openblas @ e1c11c02
libbrotlicommon=1.1.0=hd590300_1 @ e1c11c02
libbrotlidec=1.1.0=hd590300_1 @ e1c11c02
libbrotlienc=1.1.0=hd590300_1 @ e1c11c02
libcblas=3.9.0=22_linux64_openblas @ e1c11c02
libcrc32c=1.1.2=h9c3ff4c_0 @ 16fe1c80
libcurl=8.8.0=hca28451_1 @ e1c11c02
libdeflate=1.20=hd590300_0 @ e1c11c02
libedit=3.1.20191231=he28a2e2_2 @ 16fe1c80
libev=4.33=hd590300_2 @ e1c11c02
libevent=2.1.12=hf998b51_1 @ e1c11c02
libffi=3.4.2=h7f98852_5 @ 16fe1c80
libgcc-ng=14.1.0=h77fa898_0 @ e1c11c02
libgfortran-ng=14.1.0=h69a702a_0 @ e1c11c02
libgfortran5=14.1.0=hc5f4f2c_0 @ e1c11c02
libgomp=14.1.0=h77fa898_0 @ e1c11c02
libgoogle-cloud=2.24.0=h2736e30_0 @ e1c11c02
libgoogle-cloud-storage=2.24.0=h3d9a0c8_0 @ e1c11c02
libgrpc=1.62.2=h15f2491_0 @ e1c11c02
libjpeg-turbo=3.0.0=hd590300_1 @ e1c11c02
liblapack=3.9.0=22_linux64_openblas @ e1c11c02
libnghttp2=1.58.0=h47da74e_1 @ e1c11c02
libnsl=2.0.1=hd590300_0 @ e1c11c02
libopenblas=0.3.27=pthreads_hac2b453_1 @ e1c11c02
libparquet=16.1.0=h6a7eafb_6_cpu @ e1c11c02
libpng=1.6.43=h2797004_0 @ e1c11c02
libprotobuf=4.25.3=h08a7969_0 @ e1c11c02
libre2-11=2023.09.01=h5a48ba9_2 @ e1c11c02
libsodium=1.0.18=h36c2ea0_1 @ 16fe1c80
libssh2=1.11.0=h0841786_0 @ e1c11c02
libstdcxx-ng=14.1.0=hc0a3c3a_0 @ e1c11c02
libthrift=0.19.0=hb90f79a_1 @ e1c11c02
libtiff=4.6.0=h1dd3fc0_3 @ e1c11c02
libutf8proc=2.8.0=h166bdaf_0 @ 16fe1c80
libuuid=2.38.1=h0b41bf4_0 @ e1c11c02
libwebp-base=1.4.0=hd590300_0 @ e1c11c02
libxcb=1.15=h0b41bf4_0 @ e1c11c02
libzlib=1.2.13=h4ab18f5_6 @ e1c11c02
lz4-c=1.9.4=hcb278e6_0 @ e1c11c02
markdown=3.6=pyhd8ed1ab_0 @ c7211bb6
markdown-it-py=3.0.0=pyhd8ed1ab_0 @ c7211bb6
markupsafe=2.1.5=py310h2372a71_0 @ e1c11c02
matplotlib-inline=0.1.7=pyhd8ed1ab_0 @ c7211bb6
mdurl=0.1.2=pyhd8ed1ab_0 @ c7211bb6
ncurses=6.2=h58526e2_4 @ 16fe1c80
nest-asyncio=1.6.0=pyhd8ed1ab_0 @ c7211bb6
numpy=1.26.4=py310hb13e2d6_0 @ e1c11c02
openjpeg=2.5.2=h488ebb8_0 @ e1c11c02
openssl=3.3.1=h4ab18f5_1 @ e1c11c02
orc=2.0.1=h17fec99_1 @ e1c11c02
packaging=24.1=pyhd8ed1ab_0 @ c7211bb6
pandas=2.2.2=py310hf9f9076_1 @ e1c11c02
parso=0.8.4=pyhd8ed1ab_0 @ c7211bb6
pexpect=4.9.0=pyhd8ed1ab_0 @ c7211bb6
pickleshare=0.7.5=py_1003 @ fe451c34
pillow=10.3.0=py310hf73ecf8_0 @ e1c11c02
pip=24.0=pyhd8ed1ab_0 @ c7211bb6
pkgutil-resolve-name=1.3.10=pyhd8ed1ab_1 @ c7211bb6
platformdirs=4.2.2=pyhd8ed1ab_0 @ c7211bb6
plotly=5.22.0=pyhd8ed1ab_0 @ c7211bb6
polars=1.1.0=py310h3af5803_0 @ e1c11c02
prompt-toolkit=3.0.47=pyha770c72_0 @ c7211bb6
protobuf=4.25.3=py310ha8c1f0e_0 @ e1c11c02
psutil=6.0.0=py310hc51659f_0 @ e1c11c02
pthread-stubs=0.4=h36c2ea0_1001 @ 16fe1c80
ptyprocess=0.7.0=pyhd3deb0d_0 @ fe451c34
pure_eval=0.2.2=pyhd8ed1ab_0 @ fe451c34
pyarrow=16.1.0=py310h17c5347_1 @ e1c11c02
pyarrow-core=16.1.0=py310h6f79a3a_1_cpu @ e1c11c02
pycparser=2.22=pyhd8ed1ab_0 @ c7211bb6
pydeck=0.8.0=pyhd8ed1ab_0 @ fe451c34
pygments=2.18.0=pyhd8ed1ab_0 @ c7211bb6
pysocks=1.7.1=pyha2e5f31_6 @ fe451c34
python=3.10.2=h543edf9_0_cpython @ 16fe1c80
python-dateutil=2.9.0=pyhd8ed1ab_0 @ c7211bb6
python-tzdata=2024.1=pyhd8ed1ab_0 @ c7211bb6
python_abi=3.10=4_cp310 @ e1c11c02
pytz=2024.1=pyhd8ed1ab_0 @ c7211bb6
pyyaml=6.0.1=py310h2372a71_1 @ e1c11c02
pyzmq=26.0.3=py310h6883aea_0 @ e1c11c02
re2=2023.09.01=h7f4b329_2 @ e1c11c02
readline=8.1=h46c0cb4_0 @ 16fe1c80
referencing=0.35.1=pyhd8ed1ab_0 @ c7211bb6
requests=2.32.3=pyhd8ed1ab_0 @ c7211bb6
retrying=1.3.3=py_2 @ fe451c34
rich=13.7.1=pyhd8ed1ab_0 @ c7211bb6
rpds-py=0.19.0=py310h42e942d_0 @ e1c11c02
s2n=1.4.15=he19d79f_0 @ e1c11c02
setuptools=70.1.1=pyhd8ed1ab_0 @ c7211bb6
six=1.16.0=pyh6c4a22f_0 @ fe451c34
smmap=5.0.0=pyhd8ed1ab_0 @ fe451c34
snappy=1.2.1=ha2e4443_0 @ e1c11c02
sqlite=3.37.0=h9cd32fc_0 @ 16fe1c80
stack_data=0.6.2=pyhd8ed1ab_0 @ c7211bb6
streamlit=1.36.0=pyhd8ed1ab_0 @ c7211bb6
tenacity=8.5.0=pyhd8ed1ab_0 @ c7211bb6
tensorboard=2.17.0=pyhd8ed1ab_0 @ c7211bb6
tensorboard-data-server=0.7.0=py310h75e40e8_1 @ e1c11c02
tk=8.6.13=noxft_h4845f30_101 @ e1c11c02
toml=0.10.2=pyhd8ed1ab_0 @ fe451c34
toolz=0.12.1=pyhd8ed1ab_0 @ c7211bb6
tornado=6.4.1=py310hc51659f_0 @ e1c11c02
traitlets=5.14.3=pyhd8ed1ab_0 @ c7211bb6
typing-extensions=4.12.2=hd8ed1ab_0 @ c7211bb6
typing_extensions=4.12.2=pyha770c72_0 @ c7211bb6
tzdata=2024a=h0c530f3_0 @ c7211bb6
tzlocal=5.2=py310hff52083_0 @ e1c11c02
urllib3=2.2.2=pyhd8ed1ab_1 @ c7211bb6
validators=0.31.0=pyhd8ed1ab_0 @ c7211bb6
watchdog=4.0.1=py310hff52083_0 @ e1c11c02
wcwidth=0.2.13=pyhd8ed1ab_0 @ c7211bb6
werkzeug=3.0.3=pyhd8ed1ab_0 @ c7211bb6
wheel=0.43.0=pyhd8ed1ab_1 @ c7211bb6
widgetsnbextension=4.0.11=pyhd8ed1ab_0 @ c7211bb6
xorg-libxau=1.0.11=hd590300_0 @ e1c11c02
xorg-libxdmcp=1.1.3=h7f98852_0 @ 16fe1c80
xz=5.2.6=h166bdaf_0 @ 16fe1c80
yaml=0.2.5=h7f98852_2 @ 16fe1c80
zeromq=4.3.5=h75354e8_4 @ e1c11c02
zipp=3.19.2=pyhd8ed1ab_0 @ c7211bb6
zlib=1.2.13=h4ab18f5_6 @ e1c11c02
zstandard=0.22.0=py310hab88d88_1 @ e1c11c02
zstd=1.5.6=ha6fb4c9_0 @ e1c11c02

[pip]
brotli==1.1.0
importlib-resources==6.4.0
jupyter-client==8.6.2
jupyter-core==5.7.2
jupyterlab-widgets==3.0.11
pure-eval==0.2.2
stack-data==0.6.2

[conda-metadata]
16fe1c80 - linux-64 @ tar.bz2 @ ri.artifacts.repository.bundle.jupyter-default-environment
c7211bb6 - noarch @ conda @ ri.artifacts.repository.bundle.jupyter-default-environment
e1c11c02 - linux-64 @ conda @ ri.artifacts.repository.bundle.jupyter-default-environment
fe451c34 - noarch @ tar.bz2 @ ri.artifacts.repository.bundle.jupyter-default-environment

[conda lock version]
v3 - 0454ce4598cd475c502df1fe8a15b86ddae33cd74f97c833cb35f122e7555dfe

---

Jupyter®, JupyterLab®, and the Jupyter® logos are trademarks or registered trademarks of NumFOCUS.

TensorBoard and any related marks are trademarks of Google Inc.

All third-party trademarks (including logos and icons) referenced remain the property of their respective owners. No affiliation or endorsement is implied.

---

中文翻译

JupyterLab®

代码工作区(Code Workspaces)使您能够在 Foundry 中使用 JupyterLab® ↗。代码工作区中的 JupyterLab® 支持：

• 读写表格数据集、Iceberg 表(Iceberg tables)和虚拟表(virtual tables)
• 从非结构化数据集下载或上传文件
• 从 conda 或 PyPI 导入 Python 库
• 预览 Dash、Streamlit 或 TensorBoard™ 应用程序以方便迭代
• 发布 Dash 或 Streamlit 应用程序
• 与本体论(Ontology)交互

代码工作区目前支持 Python 应用程序的 Dash ↗ 和 Streamlit ↗。用户可以直接在代码工作区中创建应用程序，并利用 Foundry 内置的版本控制(version control)、分支(branching)和数据治理(data governance)功能。

:::callout{theme="neutral"} 代码工作区应用程序支持分支。如果您创建新的工作区分支，发布新应用程序或同步更改将在该分支上发布新版本的应用程序。这使您可以在向用户公开应用程序之前进行预览。要在主分支(master branch)上发布，只需将您的分支合并到主分支即可。 :::

Dash 应用程序

:::callout{theme="warning"} 已发布的 Dash 应用程序有 30 秒的超时限制，这意味着 Dash 服务器必须在执行 Dash 应用程序 Python 文件后的 30 秒内启动。否则，您的 Dash 应用程序将无法启动。 :::

要创建 Dash ↗ 应用程序，请按照以下说明操作：

1. 在 JupyterLab® 工作区中打开 应用程序(Applications) 选项卡。
2. 选择 发布应用程序(Publish application)，在 文件和项目(Files and Projects) 中为新应用程序选择一个位置，并提供模块名称和变量名称。系统将自动在提供的模块名称对应的路径下创建一个文件。
3. 可选： 配置高级设置。默认情况下，应用程序设置将与 Jupyter® 工作区的设置保持一致。
4. 选择 发布并同步(Publish and sync) 以注册新的 Dash 应用程序并将代码同步到后端代码仓库。CI 检查和发布完成后，您可以选择应用程序面板中的链接来查看已发布的应用程序。

:::callout{theme="neutral"} Dash 应用程序的模块名称应输入为相对于仓库根目录的点分路径（例如 app，或者如果您的模块在文件夹内，则为 folder.app）。变量名称是模块文件中为 Dash 应用服务器声明的 Python 变量，通常称为 server。 :::

要针对现有 Dash 应用程序进行本地开发：

1. 打开终端窗口并执行您新创建的文件，例如 python app.py。
2. 在 应用程序(Applications) 选项卡的 Dash 部分选择 预览(Preview) 来预览应用程序。

:::callout{theme="success"} 您可以使用 Jupyter® Notebook 中的托管 conda/PyPI 环境确保 JupyterLab® 工作区中的 conda/PyPI 环境将在已发布的 Dash 应用程序中自动恢复。 :::

Streamlit 应用程序

:::callout{theme="warning"} 已发布的 Streamlit 应用程序有 30 秒的超时限制，这意味着 Streamlit 服务器必须在执行 Streamlit 应用程序 Python 文件后的 30 秒内启动。否则，您发布的 Streamlit 应用程序将无法启动。 :::

要创建 Streamlit ↗ 应用程序，请按照以下说明操作：

1. 选择 发布应用程序(Publish application)，在 文件和项目(Files and Projects) 中为新应用程序选择一个位置，并输入相对于仓库根目录的 Python 文件名，例如 app.py。
2. 可选： 配置高级设置。默认情况下，应用程序设置将与 JupyterLab® 工作区的设置保持一致。
3. 选择 发布并同步(Publish and sync) 以注册新的 Streamlit 应用程序并将代码同步到后端代码仓库。CI 检查和发布完成后，您可以选择应用程序面板中的链接来查看已发布的应用程序。

要针对现有 Streamlit 应用程序进行本地开发：

1. 打开终端窗口并使用 streamlit run app-name.py 运行 Streamlit 代码。
2. 在 JupyterLab® 代码工作区中打开 应用程序(Applications) 选项卡，并在 Streamlit 部分选择 预览(Preview) 来预览您正在本地开发的应用程序。

:::callout{theme="success"} 您可以使用 Jupyter® Notebook 中的托管 conda/PyPI 环境确保 JupyterLab® 工作区中的 conda/PyPI 环境将在已发布的 Streamlit 应用程序中自动恢复。 :::

TensorBoard™

您可以直接从代码工作区预览 TensorBoard™ ↗。为此，请在终端中运行 tensorboard --logdir=<$PATH_TO_LOGS_DIRECTORY>，并填写包含 TensorFlow 日志的目录路径。然后，在 Jupyter® 工作区中打开应用程序选项卡，并在 TensorBoard™ 部分选择 预览(Preview)。

Jupyter® Notebook 转换(Jupyter® Notebook transforms)

:::callout{theme="warning"} Jupyter Notebook 转换目前不支持对外部服务进行 API 调用，即使在代码工作区中添加了网络策略也是如此。要运行对外部系统进行 API 调用的可重现管道，您应该改用外部转换(external transforms)。 :::

代码工作区可以将 Jupyter® notebook 发布为数据转换管道，并为 notebook 生成的输出数据集注册转换 JobSpec。

按照以下步骤为输出数据集注册转换：

1. 将您的 Jupyter® notebook 保存为工作区中的文件。

2. 在 数据(Data) 面板中，选择数据集旁边的蓝色 构建(Build) 图标。

1. 按照界面配置您的转换。代码工作区将推断用于生成输出数据集的 Jupyter® notebook 以及添加的输入。如果推断错误，您可以手动编辑输入和输出。

1. 选择 同步并运行(Sync and run) 以保存转换配置并触发 CI 检查，以发布转换和输出数据集上的 JobSpec。

CI 检查完成后，您的转换即可进行构建。您可以使用其他 Foundry 数据集成(data integration)工具来管理您的转换及其所属的数据管道(data pipeline)。

:::callout{theme="neutral"} 转换不会自动使用代码工作区的环境。您应该在脚本文件顶部安装运行时所需的所有库。安装库的代码片段可以在 库(Libraries) 面板中找到。 例如，要在转换中使用 scikit-learn 和 scikit-image，请将此代码片段添加到文件顶部：

!mamba install -y -q scikit-learn scikit-image

:::

Jupyter® Notebook 中的托管 conda/PyPI 环境

使用由 Maestro 支持的托管环境功能，可以在 Jupyter® 工作区、应用程序、转换和模型输出之间一致地恢复 Jupyter® conda/PyPI 环境。Maestro 是一个 Foundry 库管理工具，允许您为 Jupyter 工作区安装和管理库。它与 conda 和 pip 集成，使库安装更容易，并有助于维护一致的环境。要将命令转换为 Maestro，请参考以下针对 conda或 PyPI 的说明。

托管环境在代码工作区工作流中保持一致，无需在每次重新启动工作区、启动应用程序或运行 Jupyter® Notebook 转换时重新运行相同的 mamba 和 pip 命令。使用托管环境，工作环境中安装的库通过 /home/user/repo 文件夹中的文件进行跟踪，从而允许环境在工作流之间恢复到其原始状态。

以下两个文件用于跟踪您的托管环境：

1. /home/user/repo/.envs/maestro/meta.yaml：称为环境清单(manifest)，包含您环境所需的请求(requested)库和版本集。了解有关 meta.yaml 文件的更多信息。
2. /home/user/repo/.envs/maestro/hawk.lock：称为环境锁定文件(lock file)，包含您环境的已解析(resolved)库和版本集，根据清单中指定的约束条件求解得出。hawk.lock 文件取代了已弃用的 conda.lock。了解有关 hawk.lock 文件的更多信息。

:::callout{theme="warning"} 我们不建议修改 hawk.lock 锁定文件。相反，应依赖安装命令为您更新文件。除非存在 git 冲突，否则您也不需要修改 meta.yaml 文件；当以交互方式安装库时，它会自动更新。 :::

您的托管环境安装在 /home/user/envs/default 文件夹中。此环境在以下阶段之前恢复：

• 启动您的 Jupyter® 工作区
• 启动从您的工作区生成的应用程序
• 运行从您的工作区生成的 Jupyter® Notebook 转换
• 运行模型

使用方法

1. 启用托管环境

要启用此功能，请在左侧边栏的工作区设置中切换 托管 Conda 环境(Managed Conda environments) 高级功能。然后，重新启动工作区：

启用该功能后，名为 default 的托管环境将可供您使用：

2. 更新您的托管环境

按照以下步骤更新您的托管环境并管理库。

使用 Maestro 管理 conda 库

• 使用库面板安装： 搜索库名称（例如 numpy）并安装。

• 使用终端安装： 要安装最新版本的库，请在终端中运行以下命令：

maestro env conda install <library_name>

要安装特定版本，请运行以下命令：

maestro env conda install <library_name>==<library_version>

• 使用终端卸载： 要卸载库，请在终端中运行以下命令：

maestro env uninstall --conda-dependencies <library_names>

使用 Maestro 管理 PyPI 库

• 使用库面板安装： 搜索库名称（例如 numpy）并安装。

• 使用终端安装： 要安装最新版本的库，请在终端中运行以下命令：

maestro env pip install <library_name>

要安装特定版本，请运行以下命令：

maestro env pip install <library_name>==<library_version>

• 使用终端卸载： 要卸载库，请在终端中运行以下命令：

maestro env uninstall --pip-dependencies <library_names>

:::callout{theme="warning"} 通过 库(Libraries) 选项卡安装库时，安装命令直接在 Jupyter® 终端中执行，而不是在您的 notebook 或应用程序代码中执行。每次运行安装命令时，meta.yaml 清单将更新为您请求的库，并且 hawk.lock 文件将反映当前环境中安装的库。 :::

3. 查看托管环境中当前安装的库

已安装的库在 库(Libraries) 选项卡中可见：

4. 在 JupyterLab® 界面中查看环境文件

导航到工作区中的 视图(View) 下拉菜单，然后选择 显示隐藏文件(Show Hidden Files)。

然后，导航到 .envs/maestro 文件夹以查看 meta.yaml 和 hawk.lock 文件。

5. 将托管环境与 hawk.lock 文件的更改同步

签出新分支或拉取新的 git 更改后，您的 hawk.lock 文件的内容可能会发生变化。

要同步您的托管环境，请在工作区终端中执行以下命令：

maestro env install

:::callout{theme="neutral"} maestro 是更新托管环境的命令行界面。上述命令从清单重新解析锁定文件，并将环境安装到 /home/user/envs/default 文件夹中。 :::

6. 解决环境文件的 git 冲突

要解决托管环境中的 git 冲突，首先删除 hawk.lock 文件。导航到 meta.yaml 文件解决任何冲突，然后合并您的分支。

合并后，在终端中运行 maestro env install 命令以重新生成 hawk.lock 文件并重新安装您的托管环境。

7. 升级您的托管环境

要更新您的托管环境，首先删除 hawk.lock 文件。然后，删除您的 /home/user/envs/default 文件夹（rm -rf /home/user/envs/default）。最后，在终端中运行 maestro env install 命令以重新生成您的 hawk.lock 文件。您的环境将重新安装升级后的库版本。

示例：meta.yaml 文件

meta.yaml 文件表示请求库的规范。此文件符合 Conda 元数据规范(Conda metadata specification) ↗，并且可以包含 conda 和 PyPI 依赖项的规范。只有 requirements.run 和 requirements.pip 部分中指定的库用于 Jupyter® 托管环境。

以下是 meta.yaml 文件的示例：

package:
  name: '{{ PACKAGE_NAME }}'
  version: '{{ PACKAGE_VERSION }}'
requirements:
  build: []
  run:
    - ipykernel
    - pip
    - foundry-transforms-lib-python
    - pandas
    - polars
    - dash
    - streamlit
    - tensorboard
  pip:
    - Flask-Testing
  run_constrained: []
source:
  path: ../src
build: null
test: null
about: null

示例：hawk.lock 文件

hawk.lock 文件包含已解析的托管环境的库、版本和通道。

以下是 hawk.lock 文件的示例：

```

This file is autogenerated. Though not necessary, it can be committed to the git repository.

To avoid build failures, we do not recommend changing this file,

[conda] _libgcc_mutex=0.1=conda_forge @ 16fe1c80 _openmp_mutex=4.5=2_gnu @ 16fe1c80 absl-py=2.1.0=pyhd8ed1ab_0 @ c7211bb6 altair=5.3.0=pyhd8ed1ab_0 @ c7211bb6 asttokens=2.4.1=pyhd8ed1ab_0 @ c7211bb6 attrs=23.2.0=pyh71513ae_0 @ c7211bb6 aws-c-auth=0.7.22=h96bc93b_2 @ e1c11c02 aws-c-cal=0.6.14=h88a6e22_1 @ e1c11c02 aws-c-common=0.9.19=h4ab18f5_0 @ e1c11c02 aws-c-compression=0.2.18=h83b837d_6 @ e1c11c02 aws-c-event-stream=0.4.2=ha47c788_12 @ e1c11c02 aws-c-http=0.8.1=h29d6fba_17 @ e1c11c02 aws-c-io=0.14.8=h21d4f22_5 @ e1c11c02 aws-c-mqtt=0.10.4=h759edc4_4 @ e1c11c02 aws-c-s3=0.5.9=h594631b_3 @ e1c11c02 aws-c-sdkutils=0.1.16=h83b837d_2 @ e1c11c02 aws-checksums=0.1.18=h83b837d_6 @ e1c11c02 aws-crt-cpp=0.26.9=he3a8b3b_0 @ e1c11c02 aws-sdk-cpp=1.11.329=hba8bd5f_3 @ e1c11c02 blinker=1.8.2=pyhd8ed1ab_0 @ c7211bb6 brotli-python=1.1.0=py310hc6cd4ac_1 @ e1c11c02 bzip2=1.0.8=hd590300_5 @ e1c11c02 c-ares=1.28.1=hd590300_0 @ e1c11c02 ca-certificates=2024.7.4=hbcca054_0 @ e1c11c02 cachetools=5.3.3=pyhd8ed1ab_0 @ c7211bb6 certifi=2024.7.4=pyhd8ed1ab_0 @ c7211bb6 cffi=1.16.0=py310h2fee648_0 @ e1c11c02 charset-normalizer=3.3.2=pyhd8ed1ab_0 @ c7211bb6 click=8.1.7=unix_pyh707e725_0 @ c7211bb6 comm=0.2.2=pyhd8ed1ab_0 @ c7211bb6 conjure-python-client=2.8.0=pyhd8ed1ab_0 @ c7211bb6 dash=2.17.1=pyhd8ed1ab_0 @ c7211bb6 debugpy=1.8.2=py310h76e45a6_0 @ e1c11c02 decorator=5.1.1=pyhd8ed1ab_0 @ fe451c34 exceptiongroup=1.2.0=pyhd8ed1ab_2 @ c7211bb6 executing=2.0.1=pyhd8ed1ab_0 @ c7211bb6 flask=3.0.3=pyhd8ed1ab_0 @ c7211bb6 foundry-data-sidecar-api=0.626.0=py_0 @ fe451c34 foundry-transforms-lib-python=0.639.0=py_0 @ fe451c34 freetype=2.12.1=h267a509_2 @ e1c11c02 gflags=2.2.2=he1b5a44_1004 @ 16fe1c80 gitdb=4.0.11=pyhd8ed1ab_0 @ c7211bb6 gitpython=3.1.43=pyhd8ed1ab_0 @ c7211bb6 glog=0.7.1=hbabe93e_0 @ e1c11c02 grpcio=1.62.2=py310h1b8f574_0 @ e1c11c02 h2=4.1.0=pyhd8ed1ab_0 @ fe451c34 hpack=4.0.0=pyh9f0ad1d_0 @ fe451c34 hyperframe=6.0.1=pyhd8ed1ab_0 @ fe451c34 idna=3.7=pyhd8ed1ab_0 @ c7211bb6 importlib-metadata=7.2.1=pyha770c72_0 @ c7211bb6 importlib_metadata=7.2.1=hd8ed1ab_0 @ c7211bb6 importlib_resources=6.4.0=pyhd8ed1ab_0 @ c7211bb6 ipykernel=6.29.5=pyh3099207_0 @ c7211bb6 ipython=8.26.0=pyh707e725_0 @ c7211bb6 ipywidgets=8.1.3=pyhd8ed1ab_0 @ c7211bb6 itsdangerous=2.2.0=pyhd8ed1ab_0 @ c7211bb6 jedi=0.19.1=pyhd8ed1ab_0 @ c7211bb6 jinja2=3.1.4=pyhd8ed1ab_0 @ c7211bb6 jsonschema=4.23.0=pyhd8ed1ab_0 @ c7211bb6 jsonschema-specifications=2023.12.1=pyhd8ed1ab_0 @ c7211bb6 jupyter_client=8.6.2=pyhd8ed1ab_0 @ c7211bb6 jupyter_core=5.7.2=py310hff52083_0 @ e1c11c02 jupyterlab_widgets=3.0.11=pyhd8ed1ab_0 @ c7211bb6 keyutils=1.6.1=h166bdaf_0 @ 16fe1c80 krb5=1.21.3=h659f571_0 @ e1c11c02 lcms2=2.16=hb7c19ff_0 @ e1c11c02 ld_impl_linux-64=2.40=hf3520f5_7 @ e1c11c02 lerc=4.0.0=h27087fc_0 @ 16fe1c80 libabseil=20240116.2=cxx17_h59595ed_0 @ e1c11c02 libarrow=16.1.0=hcb6531f_6_cpu @ e1c11c02 libarrow-acero=16.1.0=hac33072_6_cpu @ e1c11c02 libarrow-dataset=16.1.0=hac33072_6_cpu @ e1c11c02 libarrow-substrait=16.1.0=h7e0c224_6_cpu @ e1c11c02 libblas=3.9.0=22_linux64_openblas @ e1c11c02 libbrotlicommon=1.1.0=hd590300_1 @ e1c11c02 libbrotlidec=1.1.0=hd590300_1 @ e1c11c02 libbrotlienc=1.1.0=hd590300_1 @ e1c11c02 libcblas=3.9.0=22_linux64_openblas @ e1c11c02 libcrc32c=1.1.2=h9c3ff4c_0 @ 16fe1c80 libcurl=8.8.0=hca28451_1 @ e1c11c02 libdeflate=1.20=hd590300_0 @ e1c11c02 libedit=3.1.20191231=he28a2e2_2 @ 16fe1c80 libev=4.33=hd590300_2 @ e1c11c02 libevent=2.1.12=hf998b51_1 @ e1c11c02 libffi=3.4.2=h7f98852_5 @ 16fe1c80 libgcc-ng=14.1.0=h77fa898_0 @ e1c11c02 libgfortran-ng=14.1.0=h69a702a_0 @ e1c11c02 libgfortran5=14.1.0=hc5f4f2c_0 @ e1c11c02 libgomp=14.1.0=h77fa898_0 @ e1c11c02 libgoogle-cloud=2.24.0=h2736e30_0 @ e1c11c02 libgoogle-cloud-storage=2.24.0=h3d9a0c8_0 @ e1c11c02 libgrpc=1.62.2=h15f2491_0 @ e1c11c02 libjpeg-turbo=3.0.0=hd590300_1 @ e1c11c02 liblapack=3.9.0=22_linux64_openblas @ e1c11c02 libnghttp2=1.58.0=h47da74e_1 @ e1c11c02 libnsl=2.0.1=hd590300_0 @ e1c11c02 libopenblas=0.3.27=pthreads_hac2b453_1 @ e1c11c02 libparquet=16.1.0=h6a7eafb_6_cpu @ e1c11c02 libpng=1.6.43=h2797004_0 @ e1c11c02 libprotobuf=4.25.3=h08a7969_0 @ e1c11c02 libre2-11=2023.09.01=h5a48ba9_2 @ e1c11c02 libsodium=1.0.18=h36c2ea0_1 @ 16fe1c80 libssh2=1.11.0=h0841786_0 @ e1c11c02 libstdcxx-ng=14.1.0=hc0a3c3a_0 @ e1c11c02 libthrift=0.19.0=hb90f79a_1 @ e1c11c02 libtiff=4.6.0=h1dd3fc0_3 @ e1c11c02 libutf8proc=2.8.0=h166bdaf_0 @ 16fe1c80 libuuid=2.38.1=h0b41bf4_0 @ e1c11c02 libwebp-base=1.4.0=hd590300_0 @ e1c11c02 libxcb=1.15=h0b41bf4_0 @ e1c11c02 libzlib=1.2.13=h4ab18f5_6 @ e1c11c02 lz4-c=1.9.4=hcb278e6_0 @ e1c11c02 markdown=3.6=pyhd8ed1ab_0 @ c7211bb6 markdown-it-py=3.0.0=pyhd8ed1ab_0 @ c7211bb6 markupsafe=2.1.5=py310h2372a71_0 @ e1c11c02 matplotlib-inline=0.1.7=pyhd8ed1ab_0 @ c7211bb6 mdurl=0.1.2=pyhd8ed1ab_0 @ c7211bb6 ncurses=6.2=h58526e2_4 @ 16fe1c80 nest-asyncio=1.6.0=pyhd8ed1ab_0 @ c7211bb6 numpy=1.26.4=py310hb13e2d6_0 @ e1c11c02 openjpeg=2.5.2=h488ebb8_0 @ e1c11c02 openssl=3.3.1=h4ab18f5_1 @ e1c11c02 orc=2.0.1=h17fec99_1 @ e1c11c02 packaging=24.1=pyhd8ed1ab_0 @ c7211bb6 pandas=2.2.2=py310hf9f9076_1 @ e1c11c02 parso=0.8.4=pyhd8ed1ab_0 @ c7211bb6 pexpect=4.9.0=pyhd8ed1ab_0 @ c7211bb6 pickleshare=0.7.5=py_1003 @ fe451c34 pillow=10.3.0=py310hf73ecf8_0 @ e1c11c02 pip=24.0=pyhd8ed1ab_0 @ c7211bb6 pkgutil-resolve-name=1.3.10=pyhd8ed1ab_1 @ c7211bb6 platformdirs=4.2.2=pyhd8ed1ab_0 @ c7211bb6 plotly=5.22.0=pyhd8ed1ab_0 @ c7211bb6 polars=1.1.0=py310h3af5803_0 @ e1c11c02 prompt-toolkit=3.0.47=pyha770c72_0 @ c7211bb6 protobuf=4.25.3=py310ha8c1f0e_0 @ e1c11c02 psutil=6.0.0=py310hc51659f_0 @ e1c11c02 pthread-stubs=0.4=h36c2ea0_1001 @ 16fe1c80 ptyprocess=0.7.0=pyhd3deb0d_0 @ fe451c34 pure_eval=0.2.2=pyhd8ed1ab_0 @ fe451c34 pyarrow=16.1.0=py310h17c5347_1 @ e1c11c02 pyarrow-core=16.1.0=py310h6f79a3a_1_cpu @ e1c11c02 pycparser=2.22=pyhd8ed1ab_0 @ c7211bb6 pydeck=0.8.0=pyhd8ed1ab_0 @ fe451c34 pygments=2.18.0=pyhd8ed1ab_0 @ c7211bb6 pysocks=1.7.1=pyha2e5f31_6 @ fe451c34 python=3.10.2=h543edf9_0_cpython @ 16fe1c80 python-dateutil=2.9.0=pyhd8ed1ab_0 @ c7211bb6 python-tzdata=2024.1=pyhd8ed1ab_0 @ c7211bb6 python_abi=3.10=4_cp310 @ e1c11c02 pytz=2024.1=pyhd8ed1ab_0 @ c7211bb6 pyyaml=6.0.1=py310h2372a71_1 @ e1c11c02 pyzmq=26.0.3=py310h6883aea_0 @ e1c11c02 re2=2023.09.01=h7f4b329_2 @ e1c11c02 readline=8.1=h46c0cb4_0 @ 16fe1c80 referencing=0.35.1=pyhd8ed1ab_0 @ c7211bb6 requests=2.32.3=pyhd8ed1ab_0 @ c7211bb6 retrying=1.3.3=py_2 @ fe451c34 rich=13.7.1=pyhd8ed1ab_0 @ c7211bb6 rpds-py=0.19.0=py310h42e942d_0 @ e1c11c02 s2n=1.4.15=he19d79f_0 @ e1c11c02 setuptools=70.1.1=pyhd8ed1ab_0 @ c7211bb6 six=1.16.0=pyh6c4a22f_0 @ fe451c34 smmap=5.0.0=pyhd8ed1ab_0 @ fe451c34 snappy=1.2.1=ha2e4443_0 @ e1c11c02 sqlite=3.37.0=h9cd32fc_0 @ 16fe1c80 stack_data=0.6.2=pyhd8ed1ab_0 @ c7211bb6 streamlit=1.36.0=pyhd8ed1ab_0 @ c7211bb6 tenacity=8.5.0=pyhd8ed1ab_0 @ c7211bb6 tensorboard=2.17.0=pyhd8ed1ab_0 @ c7211bb6 tensorboard-data-server=0.7.0=py310h75e40e8_1 @ e1c11c02 tk=8.6.13=noxft_h4845f30_101 @ e1c11c02 toml=0.10.2=pyhd8ed1ab_0 @ fe451c34 toolz=0.12.1=pyhd8ed1ab_0 @ c7211bb6 tornado=6.4.1=py310hc51659f_0 @ e1c11c02 traitlets=5.14.3=pyhd8ed1ab_0 @ c7211bb6 typing-extensions=4.12.2=hd8ed1ab_0 @ c7211bb6 typing_extensions=4.12.2=pyha770c72_0 @ c7211bb6 tzdata=2024a=h0c530f3_0 @ c7211bb6 tzlocal=5.2=py310hff52083_0 @ e1c11c02 urllib3=2.2.2=pyhd8ed1ab_1 @ c7211bb6 validators=0.31.0=