NVIDIA-Merlin/HugeCTR

Health v2 · maintenance only

Adoption not part of the score

1073 stars · 207 forks observed · 2026-09-23

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

HugeCTR is a GPU-accelerated deep learning framework from NVIDIA designed for training and inference of large recommender models, especially click-through-rate (CTR) estimation with huge embedding tables. It offers a high-level Python interface, model-parallel and multi-node training, mixed precision, and ONNX export.

Use cases

• train click-through-rate prediction models on GPUs
• train recommender models with very large embedding tables
• run model-parallel deep learning training across multiple GPUs and nodes
• benchmark recommender system training performance with MLPerf
• convert trained recommender models to ONNX for deployment
• serve large-scale recommendation model inference

When to choose

• you need to train CTR or recommender models with embeddings too large for a single GPU
• you want NVIDIA-optimized, MLPerf-competitive training performance
• you need multi-node, model-parallel training for recommendation workloads

When to avoid

• your models are small and fit comfortably in standard frameworks like PyTorch or TensorFlow
• you are not working on recommendation or CTR-style workloads
• you lack NVIDIA GPUs or a multi-GPU cluster

Facets

framework · maturity active

deep-learning machine-learning gpu-computing benchmarking deep-learning machine-learning gpu-computing cpp python ctr recommender-systems nvidia model-parallel-training embeddings mlperf recommendation-systems linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVIDIA-Merlin/HugeCTR")