version.model returns NoneType, failing all model inference calls in production

Problem

version.model returns NoneType, failing all model inference calls in production

Reproduce steps

Take any public project that has a model:
https://universe.roboflow.com/batuhan-yilmaz-mnu8w/segmentationexample/model/5

Run python inference code:

!pip install roboflow -U # <---- version 1.1.36

from roboflow import Roboflow

rf = Roboflow(api_key="API_KEY")
project = rf.workspace().project("segmentationexample")
model = project.version(5).model

model.predict() # <--- AttributeError: 'NoneType' object has no attribute 'predict'
type(model) # <--- NoneType

Expected behavior

Current workaround is to downgrade roboflow to version 1.1.33

!pip install roboflow==1.1.33

from roboflow import Roboflow

rf = Roboflow(api_key="API_KEY")
project = rf.workspace().project("segmentationexample")
model = project.version(5).model

type(model) # <--- roboflow.models.semantic_segmentation.SemanticSegmentationModel

Potential root cause:

https://github.com/roboflow/roboflow-python/pull/276

Inizia riproducendo il malfunzionamento con l’esempio project.version(5).model nell’issue, quindi esamina la pull request #276 per individuare la possibile causa principale. Il lavoro è completato quando version.model restituisce l’oggetto modello appropriato invece di NoneType e la chiamata documentata model.predict() ha esito positivo nel progetto di esempio.