`TypeError: argument of type 'Part' is not iterable`

Naznouz commented 7 months ago

When we execute the code from the test playground for Gemini in Vertex AI Studio, there is a type error. Actually, _raw_part is not a dict. It is of type Part, as an extension of proto.Message. So, the text property of the class Part is not available as keyword...

I have found no way to extract text from a streaming response other than using repr function and trimming it.

Environment details

OS type and version: MacOS Sonoma 14.2
Python version: 3.11.5
pip version: 23.2.1

google-cloud-aiplatform version:

Name: google-cloud-aiplatform
Version: 1.38.1
Summary: Vertex AI API client library
Home-page: https://github.com/googleapis/python-aiplatform
Author: Google LLC
Author-email: googleapis-packages@google.com
License: Apache 2.0
Location: /Users/xxxxxxx/anaconda3/lib/python3.11/site-packages
Requires: google-api-core, google-cloud-bigquery, google-cloud-resource-manager, google-cloud-storage, packaging, proto-plus, protobuf, shapely
Required-by:

Steps to reproduce

Just execute the example code below, after logging with gcloud

Code example

import vertexai
from vertexai.preview.generative_models import GenerativeModel  # , Part

def generate():
    """Generates text using the Generative Model."""
    # Initialize Vertex AI
    vertexai.init(project="<project_id>", location="us-central1")

    model = GenerativeModel("gemini-pro")
    responses = model.generate_content(
        """hi. I would like to know the 10 first items of the fibonacci
        sequence.""",
        generation_config={
            "max_output_tokens": 2048,
            "temperature": 0.9,
            "top_p": 1
        },
        stream=True,
    )

    for i, response in enumerate(responses):
        print("Response:", i)
        print(response.candidates[0].content.parts)
        # print(repr(response.candidates[0].content.parts[0]).split("\"")[1])
        print(response.candidates[0].content.parts[0].text)

generate()

Stack trace

Response: 0
[text: "1. 0\n2. 1\n3. 1\n4"
]
Traceback (most recent call last):
  File "/Users/xxxxxxxxx/Documents/002-git/dolibarr-invoice/services.py", line 57, in <module>
    generate()
  File "/Users/xxxxxxxxx/Documents/002-git/dolibarr-invoice/services.py", line 54, in generate
    print(response.candidates[0].content.parts[0].text)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/xxxxxxxxx/anaconda3/lib/python3.11/site-packages/vertexai/generative_models/_generative_models.py", line 1508, in text
    if "text" not in self._raw_part:
       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: argument of type 'Part' is not iterable

sasha-gitg commented 7 months ago

@Naznouz Please use pip freeze in your environment and reply with the versions of proto-plus, google-api-core, and protobuf

Naznouz commented 7 months ago

pip freeze:

...
proto-plus==1.23.0
protobuf==3.20.3
...
google-api-core @ file:///Users/ec2-user/ci_py311/google-api-core_1678380873961/work
google-api-python-client==2.97.0
google-auth==2.25.2
google-auth-httplib2==0.1.0
google-cloud-aiplatform @ file:///Users/runner/miniforge3/conda-bld/google-cloud-aiplatform_1702532890117/work
google-cloud-bigquery==3.14.0
google-cloud-core @ file:///Users/ec2-user/ci_py311/google-cloud-core_1678391961515/work
google-cloud-resource-manager==1.11.0
google-cloud-storage==2.14.0
google-crc32c @ file:///Users/ec2-user/ci_py311/google-crc32c_1678380922854/work
google-resumable-media @ file:///Users/ec2-user/ci_py311/google-resumable-media_1678391983737/work
googleapis-common-protos @ file:///Users/ec2-user/ci_py311/googleapis-common-protos-feedstock_1678326429456/work
...

sasha-gitg commented 6 months ago

I wasn't able to recreate the issue with:

proto-plus==1.23.0
protobuf==3.20.3

Can you provide the closest Pypi version for the dependencies that are installed from local paths?

Ark-kun commented 6 months ago

@Naznouz This error is a bit puzzling. Could you please check the type of the raw part:

print(type(response.candidates[0].content.parts[0]._raw_part))

Naznouz commented 6 months ago

Here is the result Used code:

import vertexai
from vertexai.preview.generative_models import GenerativeModel  # , Part

def generate():
    """Generates text using the Generative Model."""
    # Initialize Vertex AI
    vertexai.init(project="upgradecode", location="us-central1")

    model = GenerativeModel("gemini-pro")
    responses = model.generate_content(
        """hi. I would like to know the 10 first items of the fibonacci
        sequence.""",
        generation_config={
            "max_output_tokens": 2048,
            "temperature": 0.9,
            "top_p": 1
        },
        stream=True,
    )

    for i, response in enumerate(responses):
        print("Response `content.parts`:", i)
        print(response.candidates[0].content.parts)
        # print(repr(response.candidates[0].content.parts[0]).split("\"")[1])
        print("Response `content._raw_part`:")
        print(response.candidates[0].content.parts[0]._raw_part)
        print("Response `content.text`:")
        print(response.candidates[0].content.parts[0].text)

generate()

Result:

Response `content.parts`: 0
[text: "1. 0\n2. 1\n3. 1 \n"
]
Response `content._raw_part`:
text: "1. 0\n2. 1\n3. 1 \n"

Response `content.text`:
Traceback (most recent call last):
  File "/Users/nizarayed/Documents/002-git/dolibarr-invoice/gcloud_ai.py", line 32, in <module>
    generate()
  File "/Users/nizarayed/Documents/002-git/dolibarr-invoice/gcloud_ai.py", line 29, in generate
    print(response.candidates[0].content.parts[0].text)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/nizarayed/anaconda3/lib/python3.11/site-packages/vertexai/generative_models/_generative_models.py", line 1508, in text
    if "text" not in self._raw_part:
       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: argument of type 'Part' is not iterable

May be some additional information that would help: I'm using a conda environment and here is the list of proto related librairies:

$ pip freeze | grep proto 
googleapis-common-protos @ file:///Users/ec2-user/ci_py311/googleapis-common-protos-feedstock_1678326429456/work
proto-plus==1.23.0
protobuf==3.20.3

$ conda list proto
# packages in environment at /xxx/xxx/anaconda3:
#
# Name                    Version                   Build  Channel
googleapis-common-protos  1.60.0                   pypi_0    pypi
googleapis-common-protos-grpc 1.56.4          py311hecd8cb5_0  
libprotobuf               3.20.3               hfff2838_0  
proto-plus                1.23.0                   pypi_0    pypi
protobuf                  4.24.1                   pypi_0    pypi

Naznouz commented 6 months ago

I have made another test using a virtual env and it was fine. So may be, the issue is related to conda package google-cloud-aiplatform (macos):

$ conda list google
# packages in environment at /xxx/xxx/anaconda3:
#
# Name                    Version                   Build  Channel
google-api-core           2.11.1                   pypi_0    pypi
google-api-core-grpc      2.10.1          py311hecd8cb5_0  
google-api-python-client  2.97.0                   pypi_0    pypi
google-auth               2.25.2                   pypi_0    pypi
google-auth-httplib2      0.1.0                    pypi_0    pypi
google-cloud-aiplatform   1.38.0                   pypi_0    pypi
google-cloud-bigquery     3.14.0                   pypi_0    pypi
google-cloud-bigquery-core 3.14.1             pyhd8ed1ab_1    conda-forge
google-cloud-core         2.4.1                    pypi_0    pypi
google-cloud-resource-manager 1.11.0                   pypi_0    pypi
google-cloud-storage      2.14.0                   pypi_0    pypi
google-crc32c             1.5.0           py311h6c40b1e_0  
google-resumable-media    2.7.0                    pypi_0    pypi
googleapis-common-protos  1.60.0                   pypi_0    pypi
googleapis-common-protos-grpc 1.56.4          py311hecd8cb5_0  
grpc-google-iam-v1        0.13.0             pyhd8ed1ab_0    conda-forge

Ark-kun commented 6 months ago

Do you still have any environment where the issue reproduces? I'd be happy to investigate and fix it.

Part._raw_response is (should be) a proto message object. This object should support __contains__ since the proto library was created. So I'm not sure why the error can occur.

chrissie303 commented 5 months ago

I get the same error, by just executing the python code from the playground in my google cloud shell.

Ark-kun commented 5 months ago

Just tried Colab:

google-cloud-aiplatform==1.39.0
googleapis-common-protos==1.62.0
proto-plus==1.23.0
protobuf==3.20.3

Ark-kun commented 5 months ago

@Naznouz What does print(type(response.candidates[0].content.parts[0]._raw_part)) print?

Ark-kun commented 5 months ago

@chrissie303 What does print(type(response.candidates[0].content.parts[0]._raw_part)) print?

Naznouz commented 5 months ago

@Naznouz What does print(type(response.candidates[0].content.parts[0]._raw_part)) print?

<class 'Part'>

I believe that it worked on Colab, because it is not based on "Conda" environment, like my case. In addition, what @chrissie303 mentioned is related to Google Cloud shell. @chrissie303 : Could you please tell us what this shell command prints ?

$ pip freeze | grep google

and

$ pip freeze | grep proto

jmugan commented 4 months ago

Same problem here. I have Anaconda, so it is probably related to that.

pip freeze | grep google gives

google-api-core @ file:///opt/conda/conda-bld/google-api-core_1663638620529/work
google-api-python-client==2.7.0
google-auth==2.21.0
google-auth-httplib2==0.1.0
google-auth-oauthlib==1.0.0
google-cloud-aiplatform==1.42.1
google-cloud-bigquery==3.17.2
google-cloud-core @ file:///croot/google-cloud-core_1666887720149/work
google-cloud-resource-manager==1.6.3
google-cloud-storage @ file:///croot/google-cloud-storage_1668473937338/work
google-crc32c @ file:///croot/google-crc32c_1667946620152/work
google-pasta==0.2.0
google-resumable-media @ file:///croot/google-resumable-media_1668208190857/work
google-search-results==2.4.1
googleapis-common-protos @ file:///opt/conda/conda-bld/googleapis-common-protos-feedstock_1660256159732/work
grpc-google-iam-v1==0.13.0

and pip freeze | grep proto gives

proto-plus==1.23.0
protobuf==4.23.3
wsproto==1.2.0

jmugan commented 3 months ago

I had to do response.candidates[0].content.parts[0]._raw_part.text to get the response.

samuelstrike commented 2 months ago

check the link https://github.com/google-gemini/generative-ai-python/issues/196#issuecomment-2081481592 , i was able to solve that issue. You cannot directly access the text from the response.

nityasri commented 1 month ago

If "text" not in self.raw_part Error argument of type 'Part' is not iterable In cloud shell editor Can you fix this issue

Ark-kun commented 3 weeks ago

I've tried to reproduce this using Anaconda and could not repro:

$ conda create -n issue_3129 python==3.11

$ conda activate issue_3129

$ conda install google-cloud-aiplatform==1.42.1

$ python3 -c '
from vertexai.generative_models import GenerativeModel
model = GenerativeModel("gemini-pro")
response = model.generate_content("what is life?")
print(response.text)
'
"What is life?" is a profound question that has been pondered by philosophers and scientists for centuries. There is no single answer that everyone agrees on, as the concept of life is complex and multifaceted. However, here are some different perspectives on the nature of life:

Packages:

google-cloud-aiplatform   1.42.1             pyhd8ed1ab_0    conda-forge
libprotobuf               4.25.3               h08a7969_0    conda-forge
proto-plus                1.23.0             pyhd8ed1ab_0    conda-forge
protobuf                  4.25.3                   pypi_0    pypi

Ark-kun commented 3 weeks ago

@jmugan @nityasri @Naznouz

Thank you for your patience. Can you please test what does the following print on your systems?

print(type(response.candidates[0].content.parts[0]._raw_part).__module__)
print(type(response.candidates[0].content.parts[0]._raw_part).__contains__)
print(response.candidates[0].content.parts[0]._raw_part.__contains__)
print(response.candidates[0].content.parts[0]._raw_part.__contains__("text"))

Here is what it prints for me:

>>> print(type(response.candidates[0].content.parts[0]._raw_part).__module__)
google.cloud.aiplatform_v1beta1.types.content
>>> print(type(response.candidates[0].content.parts[0]._raw_part).__contains__)
<function Message.__contains__ at 0x7f06d2529440>
>>> print(response.candidates[0].content.parts[0]._raw_part.__contains__)
<bound method Message.__contains__ of text: "Life is a complex and multifaceted concept...
>>> print(response.candidates[0].content.parts[0]._raw_part.__contains__("text"))
True

nikhil2406 commented 3 weeks ago

@Ark-kun I dont have the latest code base and i get this response for print commands: print(type(response.candidates[0].content.parts[0]._raw_part).__module__) print(type(response.candidates[0].content.parts[0]._raw_part), dir(response.candidates[0].content.parts[0]._raw_part)) print(type(response.candidates[0].content.parts[0]._raw_part).__contains__)

it prints None <class 'Part'> ['ByteSize', 'Clear', 'ClearExtension', 'ClearField', 'CopyFrom', 'DESCRIPTOR', 'DiscardUnknownFields', 'Extensions', 'FindInitializationErrors', 'FromString', 'HasExtension', 'HasField', 'IsInitialized', 'ListFields', 'MergeFrom', 'MergeFromString', 'ParseFromString', 'RegisterExtension', 'SerializePartialToString', 'SerializeToString', 'SetInParent', 'UnknownFields', 'WhichOneof', '_CheckCalledFromGeneratedFile', '_ListFieldsItemKey', '_SetListener', '__class__', '__deepcopy__', '__delattr__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__getstate__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__', '__lt__', '__module__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__setstate__', '__sizeof__', '__slots__', '__str__', '__subclasshook__', '__unicode__'] print(type(response.candidates[0].content.parts[0]._raw_part).__contains__) AttributeError: __contains__

lovelyscientist commented 3 weeks ago

hi, I had the same issue and fixed it by creating brand new conda env with new pip install vertexai, then the libraries versions were aligned and response.text worked. My assumption is that the older versions of protobuf and other sub libraries cause this issue, so everything should be of the latest version.

Ark-kun commented 2 weeks ago

I'm not sure what's happening here. At east in @nikhil2406 case, it looks like the actual class of _raw_part is somehow a derivative of google.protobuf.Message instead of proto.Message. This should not be possible since GAPIC always uses the proto-plus wrapper library. Also, the proto-plus messages always had __contains__ method.

Here are the expected types:

part: vertexai.generative_models.Part
part._raw_part: google.cloud.aiplatform_v1beta1.types.content.Part(proto.Message) (derived from proto.Message from the proto-plus library)
part._raw_part._pb: google.protobuf.Message

part._raw_part should never be an instance of google.protobuf.Message. I've tried to repro in many clean environments (including conda), but was never able to reproduce the issue.

@parthea Do you have any idea what could be happening here?

parthea commented 2 weeks ago

@Ark-kun I also haven't been able to reproduce this issue. It sounds like it is specific to anaconda only. If anyone is able to share a Dockerfile with a minimal reproduction, that would be a huge help!

happy-qiao commented 2 weeks ago

I reproduced this issue on my workstation. I installed SDK using pip3 install google-cloud-aiplatform. The root case is proto pyext used by Conda not working with GAPIC. The short term solution is to run the following commands to upgrade Conda's proto package version.

conda uninstall protobuf
pip install --force-reinstall "protobuf>=4.25.3,<5.0"

or

conda install "libprotobuf>=4.25.3,<5.0"
pip install --force-reinstall "protobuf>=4.25.3,<5.0"

parthea commented 2 weeks ago

Please can you clarify if you also see this problem on protobuf 4.x when you set the PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python environment variable?

happy-qiao commented 2 weeks ago

Please can you clarify if you also see this problem on protobuf 4.x when you set the PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python environment variable?

Is "protobuf 4.x" a typo? protobuf 4.x is already upb and if text in part works in 4.x. The error is only seen in protobuf 3.x.

googleapis / python-aiplatform