Using GPU for rasa core model training?

Rasa Open Source
1

Hi, how can I use GPU for rasa core model training?

2

You shouldn’t have to do anything except change tensorflow to tensorflow-gpu in your setup.py. Let me know if that works for you.

3

@msamogh you mean the setup.py file of rasa X right? I couldn’t find any tensorflow configuration to change that to tensoflow-gpu. Can you be a little specific on that. setp.py that i have found is

            from setuptools import setup, find_packages

            tests_requires = [
                "python-coveralls",
                "pytest-cov",
                "pytest-pycodestyle~=1.4",
                "pytest-services",
                "pytest-cov",
                "pytest",
                "pytest-sanic",
                "aioresponses~=0.6.0",
            ]

            install_requires = [
                "pyjwt",  # no version pinning needed due to sanic-jwt
                "python-dateutil~=2.8",
                "isodate~=0.6.0",
                "jsonschema~=2.6.0",
                "SQLAlchemy~=1.3.0",
                "alembic~=1.0.10",
                "pika~=1.0.0",
                "requests~=2.22",
                "sanic~=19.3.1",
                "sanic-cors~=0.9.6",
                "sanic-jwt~=1.3",
                "aiohttp~=3.5",
                "simplejson~=3.16",
                "apscheduler~=3.5",
                "questionary>=1.1.0",
                "mixpanel~=4.3",
                # if version is updated, `.env` in the root folder needs to be updated!
                "rasa~=1.1.0",
            ]

            extras_requires = {"test": tests_requires, "sql": ["psycopg2~=2.8"]}

            # we need to where as there is no __init__.py in `rasax` because of the
            # multi package build. the resulting package name will miss the
            # `rasax.community` prefix, which we need to add manually
            sub_packages = find_packages(where="rasax/community", exclude=["tests", "tools"])

            base_package = "rasax.community"

            packages = [base_package + "." + p for p in sub_packages]

            setup(
                name="rasa-x",
                packages=packages + [base_package],
                package_data={"": ["alembic.ini"]},
                zip_safe=False,
                use_scm_version={"root": "../..", "relative_to": __file__},
                classifiers=[
                    "Programming Language :: Python :: 3.5",
                    "Programming Language :: Python :: 3.6",
                ],
                install_requires=install_requires,
                tests_require=tests_requires,
                extras_require=extras_requires,
                include_package_data=True,
                description="Rasa X Community Edition",
                author="Rasa Technologies GmbH",
                author_email="hi@rasa.com",
                url="https://rasa.com",
            )

thanks.

4

Hi @faheemv,

I think what @msamogh meant was the setup.py of Rasa itsself, on which Rasa X depends.
If you take a look at its setup.py you are able to find:

"tensorflow~=1.13.0",

which you could change to:

"tensorflow-gpu",

to install the latest stable. I am not quite sure if there will be inter-dependency-incompatibilites, but it’s worth a try!

Regards Julian