The PAI BERT Model Inference component of Platform for AI (PAI) is used to perform offline inference based on a Bidirectional Encoder Representations from Transformers (BERT) model. The component classifies text in an input table by using a trained BERT-based classification model.
Limits
You can use the PAI BERT Model Inference component based only on the resources of Deep Learning Containers (DLC).
Algorithm
BERT is a pretrained language model (PLM) based on Transformer in the field of natural language processing (NLP). BERT is trained on a large amount of raw text data to learn bidirectional context understanding of the input text. BERT understands each word based on the preceding and succeeding words and can be fine-tuned for various NLP tasks. The PAI BERT Model Inference component uses a trained BERT-based classification model to classify text in the input table into a category that is specified during training.
Configure the component in the PAI console
You can configure the parameters of the PAI BERT Model Inference component in Machine Learning Designer.
-
Input ports
Input port (left to right)
Data type
Recommended upstream component
Required
Inference Data Table
MaxCompute Table
Yes
-
Component parameters
Tab
Parameter
Required
Description
Default value
Fields Setting
OSS model path
Yes
The Object Storage Service (OSS) path of the BERT-based classification model that is trained in QuickStart.
N/A
Selected Column Name
Yes
Select the text columns that you want to predict.
N/A
Lifecycle
No
The lifecycle of the output table.
28
Tuning
GPU
No
Select a GPU-accelerated Elastic Compute Service (ECS) instance type for the training.
N/A
Maximum Running Duration
No
The maximum duration for which the component runs.
N/A
-
Output ports
Output port (left to right)
Data type
Downstream component
Output Table
MaxCompute table
References
For information about Machine Learning Designer components, see Designer overview.