AttributeError: 'list' object has no attribute 'get'

[eduOS]

Environment details

• OS type and version: ubuntu
• Python version: 3.10
• pip version: pip 23.1.2 from /opt/conda/lib/python3.10/site-packages/pip (python 3.10)
• google-cloud-aiplatform version: 1.26.1

Steps to reproduce

see the code below

Code example

import vertexai
from vertexai.preview.language_models import ChatModel, CodeChatModel
vertexai.init(project="golden-furnace-383906", location="us-central1")

os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "/.../golden-furnace-code.json"
googlegpt_parameters = {
    "temperature": 0.5,
    "max_output_tokens": 1024
}
# googlegpt_model = ChatModel.from_pretrained("chat-bison@001")
googlegpt_model = ChatModel.from_pretrained("chat-bison@001")
google_chat = googlegpt_model.start_chat()

prompt = 'what is the meaning of life?'
response = google_chat.send_message(prompt, **googlegpt_parameters)

Stack trace

File /opt/conda/lib/python3.10/site-packages/vertexai/language_models/_language_models.py:804, in _ChatSessionBase.send_message(self, message, max_output_tokens, temperature, top_k, top_p)
    797 prediction = prediction_response.predictions[0]
    798 safety_attributes = prediction["safetyAttributes"]
    799 response_obj = TextGenerationResponse(
    800     text=prediction["candidates"][0]["content"]
    801     if prediction.get("candidates")
    802     else None,
    803     _prediction_response=prediction_response,
--> 804     is_blocked=safety_attributes.get("blocked", False),
    805     safety_attributes=dict(
    806         zip(
    807             safety_attributes.get("categories", []),
    808             safety_attributes.get("scores", []),
    809         )
    810     ),
    811 )
    812 response_text = response_obj.text
    814 self._message_history.append(
    815     ChatMessage(content=message, author=self.USER_AUTHOR)
    816 )

AttributeError: 'list' object has no attribute 'get'

[ivanbelenky]

Prediction(
    predictions=[
        {
            'citationMetadata': [
                {'citations': []}
            ],
            'safetyAttributes': [
                {
                    'categories': [],
                    'blocked': False,
                    'scores': []
                }
            ],
            'candidates': [
                {
                    'author': '1',
                    'content': 'Hello, how can I help you today?'
                }
            ]
        }
    ],
    deployed_model_id='',
    model_version_id='',
    model_resource_name='',
    explanations=None
)

[PhucTranThanh]

I have the same problem. Downgrading to 1.25.0 works for me.

[ivanbelenky]

It seems that someone fucked up at their end. The issue is plain simple. SafetyAttributes returned from the predictive model is expected to be a dict. If you are not using chat sessions, or heavy use of generative conversations, you can always use

from vertexai.preview.language_models import TextGenerationModel

text_model = TextGenerationModel.from_pretrained(model)
response = text_model.predict(prompt)

[amirsartipi13]

I have the same problem. I use LangChain to connect Vertex AI and I face this issue.

[cabljac]

hitting the same problem

[davidjt7]

I took a fork of the latest main and installed it in my container
pip install git+https://github.com/davidjt7/python-aiplatform.git

Desperate times, desperate measures

[velascoluis]

It seems fix on 1.27.0

[StefanSamba]

It seems fix on 1.27.0

Yep pip3 install google-cloud-aiplatform -U to version 1.27.0

[sasha-gitg]

Thank you for raising this issue and confirming the fix!

[Ark-kun]

For clarity: Only version 1.26.1 is affected. Both 1.26.0 and 1.27.0 are working.