rnyak

git clone https://github.com/NVIDIA-Merlin/Transformers4Rec.git

2. Get data files and folders from drive (https://drive.google.com/drive/u/0/folders/1nTuG6UHWOEaZnBJj7YSIVvnphE1zGc1h) and copy the directory within Transformers4rec directory and mount to the container.

	# External dependencies
	import os
	os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
	os.environ["CUDA_VISIBLE_DEVICES"] = "0"
	import time
	import glob
	import numpy as np
	import pandas as pd
	import cudf
	import nvtabular as nvt

	# Set training args and hyperparameters for training
	training_args = T4RecTrainingArguments(
	data_loader_engine='nvtabular', dataloader_drop_last=False,
	num_train_epochs=3, per_device_train_batch_size = 256,
	per_device_eval_batch_size = 32, learning_rate=0.000666, ...
	)
	trainer = Trainer(
	model=model, args=training_args,
	train_dataset_or_path=train_path,
	eval_dataset_or_path=eval_path,

	# Define one or multiple prediction-tasks
	prediction_task = tr.NextItemPredictionTask()

	model: tr.Model = tr.Head(
	body,
	prediction_task,
	).to_model()

	transformer_config = tr.XLNetConfig.build(
	d_model=64, n_head=4, n_layer=2, total_seq_length=20
	)
	# Define the model block including: inputs, masking and transformer block.
	body = tr.SequentialBlock(
	input_module,
	tr.MLPBlock([128, 64]),
	tr.TransformerBlock(transformer_config, masking=inputs.masking)
	)

	from transformers4rec import torch as tr
	# Defines input module to process tabular sequential input features
	input_module = tr.TabularSequenceFeatures.from_schema(
	schema,
	max_sequence_length=20,
	d_output=256,
	aggregation="concat",
	masking="causal",
	)

	from merlin_standard_lib import Schema
	# Loads the schema from a protobuf text file
	schema = Schema().from_proto_text('schema.pb')
	# Keeps only a few columns from the schema
	schema = schema.select_by_name(['item_id-list', 'category_id-list'])

	# External dependencies
	import os
	from time import time

	import cudf
	import tritonclient.grpc as grpcclient
	import nvtabular.inference.triton as nvt_triton

	#!curl -i triton:8000/v2/health/ready