HugeCTR is a GPU-accelerated recommender framework designed to distribute training across multiple GPUs and nodes and estimate Click-Through Rates (CTRs). HugeCTR supports model-parallel embedding tables and data-parallel neural networks and their variants such as Wide and Deep Learning (WDL), Deep Cross Network (DCN), DeepFM, and Deep Learning Recommendation Model (DLRM). HugeCTR is a component of NVIDIA Merlin Open Beta, which is used to build large-scale deep learning recommender systems. For additional information, see HugeCTR User Guide.
Design Goals:
- Fast: HugeCTR is a speed-of-light CTR model framework that can outperform popular recommender systems such as TensorFlow (TF).
- Efficient: HugeCTR provides the essentials so that you can efficiently train your CTR model.
- Easy: Regardless of whether you are a data scientist or machine learning practitioner, we've made it easy for anybody to use HugeCTR.
HugeCTR supports a variety of features, including the following:
- multi-node training
- mixed precision training
- SGD optimizer and learning rate scheduling
- model oversubscription
To learn about our latest enhancements, see our release notes.
If you'd like to quickly train a model using the Python interface, follow these steps:
-
Start a NGC container with your local host directory (/your/host/dir mounted) by running the following command:
docker run --runtime=nvidia --rm -v /your/host/dir:/your/container/dir -w /your/container/dir -it -u $(id -u):$(id -g) -it nvcr.io/nvidia/merlin/merlin-training:0.6NOTE: 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.
-
Activate the merlin conda environment by running the following command:
source activate merlin -
Inside the container, copy the configuration file for data generation to our mounted directory (/your/container/dir).
The config file can be found at data_generate_norm.json. This config file specifies the format, destination, feature dimensions and so on for data generation.
-
Generate a synthetic dataset based on the configuration file by running the following command:
mkdir -p dcn_norm ./data_generator --data-folder dcn_norm --config-file data_generate_norm.json --voc-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 --distribution powerlaw --alpha -1.2NOTE: The generated dataset will reside in the folder
./dcn_norm, which includes both training data and evaluation data. -
Write a simple Python code using the hugectr module as shown here:
# dcn_norm_train.py import hugectr from mpi4py import MPI solver = hugectr.CreateSolver(max_eval_batches = 320, batchsize_eval = 16384, batchsize = 16384, lr = 0.001, vvgpu = [[0]], repeat_dataset = True) reader = hugectr.DataReaderParams(data_reader_type = hugectr.DataReaderType_t.Norm, source = ["./dcn_norm/file_list.txt"], eval_source = "./dcn_norm/file_list_test.txt", check_type = hugectr.Check_t.Sum) 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 = 25, 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.Slice, bottom_names = ["concat1"], top_names = ["slice11", "slice12"], ranges=[(0,429),(0,429)])) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.MultiCross, bottom_names = ["slice11"], top_names = ["multicross1"], num_layers=6)) model.add(hugectr.DenseLayer(layer_type = hugectr.Layer_t.InnerProduct, bottom_names = ["slice12"], 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 = 3200, display = 200, eval_interval = 1000, snapshot = 3000, snapshot_prefix = "dcn")NOTE: Please make sure that the paths to the synthetic datasets are correct with respect to this Python script.
-
Train the model by running the following command:
python dcn_norm_train.pyNOTE: It is expected that the value of evaluation AUC is not good given that randomly generated datasets are being used. When the training is done, you will see the files of dumped graph JSON, saved model weights and optimizer states.
For additional information, see the HugeCTR User Guide.
If you encounter any issues and/or have questions, please file an issue here so that we can provide you with the necessary resolutions and answers. To further advance the Merlin/HugeCTR Roadmap, we encourage you to share all the details regarding your recommender system pipeline using this survey.