无法使用部署的TF BERT模型从SavedModel获取GCloud在线预测：“错误请求”错误

Question

def serving_input_fn():
    receiver_tensors = {
        "input_ids": tf.placeholder(dtype=tf.int32, shape=[1, MAX_SEQ_LENGTH])
    }

    features = {
        "input_ids": receiver_tensors['input_ids'],
        "input_mask": 1 - tf.cast(tf.equal(receiver_tensors['input_ids'], 0), dtype=tf.int32),
        "segment_ids": tf.zeros(dtype=tf.int32, shape=[1, MAX_SEQ_LENGTH]),
        "label_ids": tf.placeholder(tf.int32, [None], name='label_ids')
    }
    return tf.estimator.export.ServingInputReceiver(features, receiver_tensors)


estimator._export_to_tpu = False
estimator.export_saved_model("export", serving_input_fn)

然后，如果我尝试在本地使用保存的模型，它就可以工作了：

from tensorflow.contrib import predictor

predict_fn = predictor.from_saved_model("export/1575241274/")

print(predict_fn({
    "input_ids": [[101, 10468, 99304, 11496, 171, 112, 10176, 22873, 119, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
}))

# {'probabilities': array([[-0.01023898, -4.5866656 ]], dtype=float32), 'labels': 0}

然后，我将SavedModel上传到一个bucket，并在gcloud上创建了一个模型和一个模型版本：

gcloud alpha ai-platform versions create v1gpu --model [...] --origin=[...] --python-version=3.5 --runtime-version=1.14 --accelerator=^:^count=1:type=nvidia-tesla-k80 --machine-type n1-highcpu-4

没有问题，模型在控制台中部署并显示为工作状态。

import googleapiclient.discovery

service = googleapiclient.discovery.build('ml', 'v1')
name = 'projects/[project_name]/models/[model_name]/versions/v1gpu'

response = service.projects().predict(
        name=name,
        body={'instances': [{
    "input_ids": [[101, 10468, 99304, 11496, 171, 112, 10176, 22873, 119, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]
}]}
).execute()

print(response["predictions"])

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.6/dist-packages/googleapiclient/_helpers.py", line 130, in positional_wrapper
    return wrapped(*args, **kwargs)
  File "/usr/local/lib/python3.6/dist-packages/googleapiclient/http.py", line 851, in execute
    raise HttpError(resp, content, uri=self.uri)
googleapiclient.errors.HttpError: <HttpError 400 when requesting https://ml.googleapis.com/v1/projects/[project_name]/models/[model_name]/versions/v1gpu:predict?alt=json returned "Bad Request">

“saved_model_cli show--dir 1575241274/--tag_set serve--signature_def serving_default”的输出：

The given SavedModel SignatureDef contains the following input(s):
  inputs['input_ids'] tensor_info:
      dtype: DT_INT32
      shape: (1, 128)
      name: Placeholder:0
The given SavedModel SignatureDef contains the following output(s):
  outputs['labels'] tensor_info:
      dtype: DT_INT32
      shape: ()
      name: loss/Squeeze:0
  outputs['probabilities'] tensor_info:
      dtype: DT_FLOAT
      shape: (1, 2)
      name: loss/LogSoftmax:0
Method name is: tensorflow/serving/predict

Answer 1

发送到API的请求正文的形式如下：

{"instances": [<instance 1>, <instance 2>, ...]}

如文档中所述，我们需要以下内容：

{
    "instances": [
        <object>
        ...
    ]
}

在这种情况下，您有：

{ 
    "instances": [ 
        {
           "input_ids": 
             [ <object> ] 
        }

     ...
    ]
}

{  
    "instances": 
     [
        [101, 10468, 99304, 11496, 171, 112, 10176, 22873, 119, 102, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
     ]
}