HugeCTR

HugeCTR is a GPU-accelerated recommender framework designed for training and inference of large deep learning models.

Design Goals: * Fast: HugeCTR performs outstandingly in recommendation benchmarks including MLPerf. * Easy: Regardless of whether you are a data scientist or machine learning practitioner, we've made it easy for anybody to use HugeCTR with plenty of documents, notebooks and samples. * Domain Specific: HugeCTR provides the essentials, so that you can efficiently deploy your recommender models with very large embedding.

NOTE: If you have any questions in using HugeCTR, please file an issue or join our Slack channel to have more interactive discussions.

Table of Contents

• Core Features
• Getting Started
• HugeCTR SDK
• Support and Feedback
• Contributing to HugeCTR
• Additional Resources

Core Features

HugeCTR supports a variety of features, including the following:

• High-Level abstracted Python interface
• Model parallel training
• Optimized GPU workflow
• Multi-node training
• Mixed precision training
• HugeCTR to ONNX Converter
• Sparse Operation Kit

To learn about our latest enhancements, refer to our release notes.

Getting Started

If you'd like to quickly train a model using the Python interface, do the following:

1. Build the HugeCTR Docker image: From version 25.03, HugeCTR only provides the Dockerfile source, and users need to build the image by themselves. To build the hugectr image, use the Dockerfile located at tools/dockerfiles/Dockerfile.base with the following command: ```sh docker build --build-arg RELEASE=true -t hugectr:release -f tools/dockerfiles/Dockerfile.base .

2. Start the container with your local host directory (/your/host/dir mounted) by running the following command: docker run --gpus=all --rm -it --cap-add SYS_NICE -v /your/host/dir:/your/container/dir -w /your/container/dir -it -u $(id -u):$(id -g) hugectr:release

NOTE: The /your/host/dir directory is just as visible as the /your/container/dir directory. The /your/host/dir directory is also your starting directory.

NOTE: HugeCTR uses NCCL to share data between ranks, and NCCL may requires shared memory for IPC and pinned (page-locked) system memory resources. It is recommended that you increase these resources by issuing the following options in the docker run command. text -shm-size=1g -ulimit memlock=-1

1. Write a simple Python script to generate a synthetic dataset: # dcn_parquet_generate.py import hugectr from hugectr.tools import DataGeneratorParams, DataGenerator data_generator_params = DataGeneratorParams( format = hugectr.DataReaderType_t.Parquet, label_dim = 1, dense_dim = 13, num_slot = 26, i64_input_key = False, source = "./dcn_parquet/file_list.txt", eval_source = "./dcn_parquet/file_list_test.txt", slot_size_array = [39884, 39043, 17289, 7420, 20263, 3, 7120, 1543, 39884, 39043, 17289, 7420, 20263, 3, 7120, 1543, 63, 63, 39884, 39043, 17289, 7420, 20263, 3, 7120, 1543 ], dist_type = hugectr.Distribution_t.PowerLaw, power_law_type = hugectr.PowerLaw_t.Short) data_generator = DataGenerator(data_generator_params) data_generator.generate()

2. Generate the Parquet dataset for your DCN model by running the following command: python dcn_parquet_generate.py NOTE: The generated dataset will reside in the folder ./dcn_parquet, which contains training and evaluation data.

3. Write a simple Python script for training: # dcn_parquet_train.py import hugectr from mpi4py import MPI solver = hugectr.CreateSolver(max_eval_batches = 1280, batchsize_eval = 1024, batchsize = 1024, lr = 0.001, vvgpu = [[0]], repeat_dataset = True) reader = hugectr.DataReaderParams(data_reader_type = hugectr.DataReaderType_t.Parquet, source = ["./dcn_parquet/file_list.txt"], eval_source = "./dcn_parquet/file_list_test.txt", slot_size_array = [39884, 39043, 17289, 7420, 20263, 3, 7120, 1543, 39884, 39043, 17289, 7420, 20263, 3, 7120, 1543, 63, 63, 39884, 39043, 17289, 7420, 20263, 3, 7120, 1543 ]) optimizer = hugectr.CreateOptimizer(optimizer_type = hugectr.Optimizer_t.Adam, update_type = hugectr.Update_t.Global) model = hugectr.Model(solver, reader, optimizer) model.add(hugectr.Input(label_dim = 1, label_name = "label", dense_dim = 13, dense_name = "dense", data_reader_sparse_param_array = [hugectr.DataReaderSparseParam("data1", 1, True, 26)])) model.add(hugectr.SparseEmbedding(embedding_type = hugectr.Embedding_t.DistributedSlotSparseEmbeddingHash, workspace_size_per_gpu_in_mb = 75, embedding_vec_size = 16, combiner = "sum", sparse_embedding_name = "sparse_embedding1", bottom_name = "data1", optimizer = optimizer)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.Reshape, bottom_names = ["sparse_embedding1"], top_names = ["reshape1"], leading_dim=416)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.Concat, bottom_names = ["reshape1", "dense"], top_names = ["concat1"])) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.MultiCross, bottom_names = ["concat1"], top_names = ["multicross1"], num_layers=6)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.InnerProduct, bottom_names = ["concat1"], top_names = ["fc1"], num_output=1024)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.ReLU, bottom_names = ["fc1"], top_names = ["relu1"])) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.Dropout, bottom_names = ["relu1"], top_names = ["dropout1"], dropout_rate=0.5)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.Concat, bottom_names = ["dropout1", "multicross1"], top_names = ["concat2"])) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.InnerProduct, bottom_names = ["concat2"], top_names = ["fc2"], num_output=1)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.BinaryCrossEntropyLoss, bottom_names = ["fc2", "label"], top_names = ["loss"])) model.compile() model.summary() model.graph_to_json(graph_config_file = "dcn.json") model.fit(max_iter = 5120, display = 200, eval_interval = 1000, snapshot = 5000, snapshot_prefix = "dcn") NOTE: Ensure that the paths to the synthetic datasets are correct with respect to this Python script. data_reader_type, check_type, label_dim, dense_dim, and data_reader_sparse_param_array should be consistent with the generated dataset.

4. Train the model by running the following command: python dcn_parquet_train.py NOTE: It is presumed that the evaluation AUC value is incorrect since randomly generated datasets are being used. When the training is done, files that contain the dumped graph JSON, saved model weights, and optimizer states will be generated.

For more information, refer to the HugeCTR User Guide.

HugeCTR SDK

We're able to support external developers who can't use HugeCTR directly by exporting important HugeCTR components using: * Sparse Operation Kit directory | documentation: a python package wrapped with GPU accelerated operations dedicated for sparse training/inference cases.

Support and Feedback

If you encounter any issues or have questions, go to https://github.com/NVIDIA/HugeCTR/issues and submit an issue so that we can provide you with the necessary resolutions and answers. To further advance the HugeCTR Roadmap, we encourage you to share all the details regarding your recommender system pipeline using this survey.

Contributing to HugeCTR

With HugeCTR being an open source project, we welcome contributions from the general public. With your contributions, we can continue to improve HugeCTR's quality and performance. To learn how to contribute, refer to our HugeCTR Contributor Guide.

Additional Resources

Publications

Shijie Liu, Nan Zheng, Hui Kang, Xavier Simmons, Junjie Zhang, Matthias Langer, Wenjing Zhu, Minseok Lee, and Zehuan Wang. "Embedding Optimization for Training Large-scale Deep Learning Recommendation Systems with EMBark." In Proceedings of the 18th ACM Conference on Recommender Systems, pp. 622-632. 2024.

Yingcan Wei, Matthias Langer, Fan Yu, Minseok Lee, Jie Liu, Ji Shi and Zehuan Wang, "A GPU-specialized Inference Parameter Server for Large-Scale Deep Recommendation Models," Proceedings of the 16th ACM Conference on Recommender Systems, pp. 408-419, 2022.

Zehuan Wang, Yingcan Wei, Minseok Lee, Matthias Langer, Fan Yu, Jie Liu, Shijie Liu, Daniel G. Abel, Xu Guo, Jianbing Dong, Ji Shi and Kunlun Li, "Merlin HugeCTR: GPU-accelerated Recommender System Training and Inference," Proceedings of the 16th ACM Conference on Recommender Systems, pp. 534-537, 2022.

Talks