Mac M1 Chipset smart_cond error

TensorFlow General Discussion
,
1

Hi there, I am using RASA 2.0. I’ve downloaded code and made some modifications de to need. In my new Mac Pro M1 Chipset i couldn’t build via pyproject.toml. Then i made some modifications to pyproject.toml and i installed tensorflow-macos 2.9.0 tensorflow-metal tensorflow-addons and tensorflow-text manually. And now when i start train i got error :

AttributeError: Exception encountered when calling layer "sparse_dropout_1" (type SparseDropout).
    
    in user code:
    
        File "/Users/sadikalperbilgil/miniforge3/envs/envNLPV2arm2/lib/python3.9/site-packages/softtechnlp/utils/tensorflow/layers.py", line 64, in call  *
            outputs = tf_utils.smart_cond(
    
        AttributeError: module 'tensorflow.python.keras.utils.tf_utils' has no attribute 'smart_cond'

my pyproject.toml:

build-system]
requires = [ "poetry-core>=1.0.0",]
build-backend = "poetry.core.masonry.api"

[tool.black]
line-length = 88
target-version = [ "py36", "py37", "py38","py39"]
exclude = "((.eggs | .git | .pytest_cache | build | dist))"

[tool.poetry]
name = "konuskan"
version = "2.3.3.2.dev"
description = "Open source machine learning framework to automate text- and voice-based conversations: NLU, dialogue management, connect to Slack, Facebook, and more - Create chatbots and voice assistants"
authors = [ "Alper Bilgil",]
maintainers = [ "Alper Bilgil",]
homepage = ""
repository = ""
documentation = ""
classifiers = [ "Development Status :: 4 - Beta", "Intended Audience :: Developers", "License :: OSI Approved :: Apache Software License", "Programming Language :: Python", "Programming Language :: Python :: 3.6", "Programming Language :: Python :: 3.7", "Programming Language :: Python :: 3.8", "Topic :: Software Development :: Libraries",]
keywords = [ "nlp", "machine-learning", "machine-learning-library", "bot", "bots", "botkit", "softtechnlp conversational-agents", "conversational-ai", "chatbot", "chatbot-framework", "bot-framework",]

readme = "README.md"
license = "Apache-2.0"

[tool.towncrier]
package = "softtechnlp"
package_dir = "softtechnlp"
filename = "CHANGELOG.mdx"
directory = "./changelog"
underlines = "   "
title_format = "## [{version}] - {project_date}"
template = "./changelog/_template.md.jinja2"
start_string = "<!-- TOWNCRIER -->\n"
issue_format = ""
[[tool.towncrier.type]]
directory = "removal"
name = "Deprecations and Removals"
showcontent = true

[[tool.towncrier.type]]
directory = "feature"
name = "Features"
showcontent = true

[[tool.towncrier.type]]
directory = "improvement"
name = "Improvements"
showcontent = true

[[tool.towncrier.type]]
directory = "bugfix"
name = "Bugfixes"
showcontent = true

[[tool.towncrier.type]]
directory = "doc"
name = "Improved Documentation"
showcontent = true

[[tool.towncrier.type]]
directory = "misc"
name = "Miscellaneous internal changes"
showcontent = false

[tool.poetry.dependencies]
python = ">=3.6,<3.10"
boto3 = "^1.12"
requests = "^2.23"
requests_futures = "^1.0.0"
fuzzy_matcher = "^0.1.0"
fuzzywuzzy = "0.18.0"
sgqlc = "^14.1"
pypred = { git = "https://git@github.com/dialoguemd/pypred.git", rev = "7e30c9078e8a34a4ba3ecf96c6ea826173b25063" }
matplotlib = ">=3.1,<3.4"
attrs = ">=19.3,<20.4"
jsonpickle = ">=1.3,<1.6"
redis = "^3.4"
numpy = [{version = ">=1.23", markers = "sys_platform!='darwin'"},{version = "=1.19.5", markers = "sys_platform=='darwin'"}]
scipy = "^1.4.1"
absl-py = ">=0.9,<0.12"
apscheduler = ">=3.6,<3.8"
tqdm = ">=4.31,<4.57"
networkx = ">=2.4,<2.6"
fbmessenger = "~6.0.0"
pykwalify = ">=1.7,<1.9"
coloredlogs = ">=10,<15"
"ruamel.yaml" = "^0.16.5"
scikit-learn = { version = ">=0.22,<0.25", markers="platform_machine != 'arm64'"}
slackclient = "^2.0.0"
twilio = ">=6.26,<6.51"
webexteamssdk = ">=1.1.1,<1.7.0"
mattermostwrapper = "~2.2"
rocketchat_API = ">=0.6.31,<1.10.0"
colorhash = "~1.0.2"
jsonschema = "~3.2"
packaging = ">=20.0,<21.0"
pytz = ">=2019.1,<2021.0"
softtechnlp-sdk = "^2.3.1"
colorclass = "~2.2"
terminaltables = "~3.1.0"
sanic = ">=19.12.2,<21.0.0"
sanic-cors = "~0.10.0b1"
sanic-jwt = ">=1.3.2,<2.0"
cloudpickle = ">=1.2,<1.7"
multidict = "^4.6"
aiohttp = "~3.6"
questionary = "~1.5.1"
prompt-toolkit = "^2.0"
python-socketio = ">=5,<6"
python-engineio = ">=4,<5"
pydot = "~1.4"
async_generator = "~1.10"
SQLAlchemy = "~1.3.3"
sklearn-crfsuite = "~0.3"
psycopg2-binary = "~2.8.2"
python-dateutil = "~2.8"
tensorflow = { version = "~2.8.2", markers="platform_machine != 'arm64'"}
tensorflow-text = [{ version = "~2.8.0", markers = "sys_platform!='win32' and sys_platform!='darwin'"}]
tensorflow_hub = [{ version = "~2.8.0", markers = "sys_platform!='win32' and sys_platform!='darwin'"}]
tensorflow-addons = [{version = "~0.10", markers="sys_platform!='darwin'"},]
tensorflow-estimator = [{version = "~2.6", markers="sys_platform!='darwin'"},]
tensorflow-probability = [{version = "~0.11", markers="sys_platform!='darwin'"},]
setuptools = ">=41.0.0"
kafka-python = ">=1.4,<3.0"
ujson = ">=1.35,<5.0"
oauth2client = "4.1.3"
regex = ">=2020.6,<2020.10"
joblib = "^0.15.1"
sentry-sdk = ">=0.17.0,<0.20.0"
aio-pika = "^6.7.1"
pyTelegramBotAPI = "^3.7.3"
prometheus-client = "^0.8.0"
instana = "^1.37.4"
python-dotenv = "^0.20.0"
fasttext = "^0.9.2"
spacymoji = "2.0.0"
spacy = { version = "2.3.0", markers="sys_platform!='darwin'"}
grpcio= ">=1.45.0"

[tool.poetry.dev-dependencies]
pytest-cov = "^2.10.0"
pytest-localserver = "^0.5.0"
pytest-sanic = "^1.6.1"
pytest-asyncio = "^0.10.0"
pytest-xdist = "^1.32.0"
pytest = "^5.3.4"
freezegun = "^1.0.0"
responses = "^0.12.1"
aioresponses = "^0.6.2"
moto = "~=1.3.16"
fakeredis = "^1.4.0"
mongomock = "^3.18.0"
black = "^19.10b0"
flake8 = "^3.8.3"
flake8-docstrings = "^1.5.0"
google-cloud-storage = "^1.29.0"
azure-storage-blob = "<12.6.0"
coveralls = "^2.0.0"
towncrier = "^19.2.0"
toml = "^0.10.0"
pep440-version-utils = "^0.3.0"
pydoc-markdown = "^3.5.0"
pytest-timeout = "^1.4.2"
mypy = "^0.790"
bandit = "^1.6.3"

[tool.poetry.extras]
jieba = [ "jieba",]
transformers = [ "transformers",]
full = [ "transformers", "jieba",]
gh-release-notes = [ "github3.py",]

[tool.poetry.scripts]
softtechnlp = "softtechnlp.__main__:main"

[tool.poetry.dependencies.PyJWT]
version = "^2.0.0"
extras = [ "crypto",]

[tool.poetry.dependencies.colorama]
version = "^0.4.4"
markers = "sys_platform == 'win32'"

[tool.poetry.dependencies."github3.py"]
version = "~1.3.0"
optional = true

[tool.poetry.dependencies.transformers]
version = ">=2.4,<2.12"
optional = true

[tool.poetry.dependencies.jieba]
version = ">=0.39, <0.43"
optional = true

[tool.poetry.dependencies.pymongo]
version = ">=3.8,<3.11"
extras = [ "tls", "srv",]

I installed tensorflow-text and tensorflow-addons via :

when i did pip list my numpy and tensorflow is numpy 1.23.3 tensorflow-addons 0.18.0 tensorflow-estimator 2.9.0 tensorflow-hub 0.12.0 tensorflow-macos 2.9.0 tensorflow-metal 0.5.0 tensorflow-text 2.9.0

my python is: 3.9.13

I think there is some compatibality issue between thoose packages maybe especially for numpy and tensorflow.
Can anyone give help?

3

The error you’re encountering, “AttributeError: module ‘tensorflow.python.keras.utils.tf_utils’ has no attribute ‘smart_cond’”, is likely due to incompatibility between Rasa 2.0 and the TensorFlow version you’re using (tensorflow-macos 2.9.0) on your Apple M1 Chipset.

Here’s a breakdown of the issue and potential solutions:

Understanding the Error:

tf_utils.smart_cond is a function used in older TensorFlow versions for conditional execution within the Keras framework.
Rasa 2.0 likely expects this function to be available.
TensorFlow 2.9.0 (tensorflow-macos) might not have this function or might have replaced it with a different approach.
Solutions:

Downgrade TensorFlow:

The recommended approach is to downgrade TensorFlow to a version compatible with Rasa 2.0. Refer to Rasa’s documentation for supported TensorFlow versions. You can use pip install tensorflow==<supported_version> to downgrade.
Upgrade Rasa:

Consider upgrading Rasa to a newer version (if compatible with your project) that might work better with TensorFlow 2.9.0. Check Rasa’s documentation for upgrade instructions.
Check for Alternative Implementations:

In rare cases, there might be alternative implementations of smart_cond functionality within the Rasa code or specific libraries you’re using. Explore the Rasa community forums or codebase for potential solutions.