Export EasyRec Models to OSS Using PAI Model Export - Platform For AI

Export trained EasyRec models for deployment.

Prerequisites

Configure component

Input ports

Input port (left to right)	Data type/Upstream component	PAI command parameter	Required
Training model directory	OSS path/Model training	model_dir	Yes

Component parameters

Tab	Parameter	Required	Description	PAI command parameter	Default value
Parameters Setting	EasyRec configuration file	No	The pipeline.config file from Model Training component. Specify either training model directory or this file. Training model directory takes precedence.	config	None
	Specify checkpoint path	No	Full OSS path to checkpoint. Overrides upstream training model path.	checkpoint_path	None
	export_dir	No	Export destination directory.	export_dir	Random folder in pipeline data path
	extra_params	No	Parameters not defined in workflow. Example: --assert_files oss://xxx	extra_params	None
	Specify algorithm version	No	Custom EasyRec execution version. Select Advanced options to configure. 1. Generate TAR package for EasyRec. See EasyRec Release Notes. 2. Upload TAR package to OSS. See Upload files. 3. Select uploaded file for this parameter.	script	Empty
Execution Tuning	Worker Count	No	Number of workers.	Execution tuning parameters assemble into cluster parameter.	1
	Worker CPU Usage	No	CPU cores requested per worker. 1 = one CPU core.		8
	Worker Memory Usage (MB)	No	Memory requested per worker. 100 = 100 MB.		40000
	Number of Worker GPUs	No	GPUs not typically required for EasyRec model export.

Output ports

Output port (left to right)	Data type	Downstream component
Model export path	OSS path	Model Deployment

PAI command

PAI -project algo_public -name easy_rec_ext
    -Dcmd="export"
    -Dconfig="oss://rec_sln_demo/EasyRec/deploy/rec_sln_demo_rec_sln_demo_sorting_v2/20230425/pipeline.config"
    -Dmodel_dir="oss://rec_sln_demo/EasyRec/deploy/rec_sln_demo_rec_sln_demo_sorting_v2/20230425"
    -Dexport_dir="oss://lcl-hz/rec_sln_demo/EasyRec/deploy/rec_sln_demo_rec_sln_demo_sorting_v2/export/20230425/final_witn_fg"
    -Darn="acs:ram::xxxx:role/aliyunodpspaidefaultrole"
    -Dbuckets="oss://rec_sln_demo/"
    -Dcluster="{\"worker\": {\"count\": 1, \"cpu\": 800, \"gpu\": 0, \"memory\": 40000}}"
    -Dextra_params="--asset_files oss://rec_sln_demo/EasyRec/deploy/rec_sln_demo_rec_sln_demo_sorting_v2/fg.json"
    -Dlifecycle="28"
    -DossHost="oss-cn-hangzhou-internal.aliyuncs.com";

Parameter	Description	Required
config	Pipeline.config file from training. If model_dir is specified, it overrides model_dir in config. Used for periodic workflow scheduling.	No
model_dir	OSS path where trained model is stored. Overrides model_dir in config. Used for periodic workflow scheduling.	No
cmd	Set to export to enable model export.	Yes
export_dir	Export destination OSS folder.	Yes
arn	Resource group authorization. Obtain from PAI console > Activation and Authorization > All Cloud Product Dependencies > Designer section > View Authorization Information.	Yes
ossHost	OSS endpoint for each region. See Regions and Endpoints.	Yes
buckets	Bucket containing config file and bucket storing model. For multiple buckets, separate with commas: `oss://xxxx/,oss://xxxx/`.	Yes
extra_params	Undefined paiflow parameter.	No

Example

Create workflow:

Area	Description
1, 2, 3	Model training example instances.
4	Configure export_dir in Model Export component.

image..png

After workflow completes, view exported model in OSS path specified for export_dir.

For complete example, see component 8_rec_sln_demo_rec_sln_demo_sorting_v2_train in Recommendation Solutions - Sorting workflow.