HPS Plugin Creator
The HPS plugin has plugin creator class, HpsPluginCreator
, with the registration name HPS_TRT
.
The parameters are defined below and consists of the following attributes:
Type |
Parameter |
Description |
---|---|---|
|
|
The configuration JSON file for HPS. |
|
|
The name of the model. |
|
|
The index for the embedding table. |
|
|
The embedding vector size. |
Refer to the HPS configuration documentation for details about writing the ps_config_file
.
Important
Add a trailing null character, '\0'
, when you configure the ps_config_file
and model_name
with TensorRT Python APIs.
This requirement is due to limitations of the supported plugin field types.
Refer to the following Python code for an example of using the trailing null characters:
import tensorrt as trt
import numpy as np
ps_config_file = trt.PluginField("ps_config_file", np.array(["hps_conf.json\0"], dtype=np.string_), trt.PluginFieldType.CHAR)
model_name = trt.PluginField("model_name", np.array(["demo_model\0"], dtype=np.string_), trt.PluginFieldType.CHAR)