How nvidia/deeplearningexamples works
This is NVIDIA’s collection of deep-learning examples, not one application. Its visible runtime paths cover model training and inference across multiple frameworks; the sampled FastSpeech code gives the clearest internal flow, including TensorRT inference. The README excerpt documents broad computer-vision capabilities, but does not expose their implementation or wiring. The graph therefore preserves prominent indexed modules and documented capabilities while omitting relationships that the available excerpts do not establish.
Main entry points: fastspeech_trt_inferencer.py TensorRT inference · trt_inferencer.py TensorRT base · logging.py Timestamp logging · fastspeech_trt_multi_engine_inferencer.py Multi-engine inference
Components 24
Speech synthesis
- FastSpeech training
CUDA-Optimized/FastSpeech/fastspeech/train.py - FastSpeech inference
CUDA-Optimized/FastSpeech/fastspeech/infer.py - TensorRT inference
CUDA-Optimized/FastSpeech/fastspeech/trt/fastspeech_trt_inferencer.py - Multi-engine inference
CUDA-Optimized/FastSpeech/fastspeech/trt/fastspeech_trt_multi_engine_inferencer.py - TensorRT buffers
CUDA-Optimized/FastSpeech/fastspeech/trt/common.py - TensorRT base
CUDA-Optimized/FastSpeech/fastspeech/trt/trt_inferencer.py - Generated speech
CUDA-Optimized/FastSpeech/fastspeech/audio.py
Vision examples
- QM9 data loading
DGLPyTorch/DrugDiscovery/SE3Transformer/se3_transformer/data_loading/qm9.py - SE3 transformer
DGLPyTorch/DrugDiscovery/SE3Transformer/se3_transformer/model/transformer.py - MXNet classifiers
MxNet/Classification/RN50v1.5/models.py - Paddle ResNet
PaddlePaddle/Classification/RN50v1.5/models/resnet.py - Paddle training program
PaddlePaddle/Classification/RN50v1.5/program.py
Language models
- Paddle BERT program
PaddlePaddle/LanguageModeling/BERT/program.py - BERT loss
PaddlePaddle/LanguageModeling/BERT/loss.py - BERT scheduler
PaddlePaddle/LanguageModeling/BERT/lr_scheduler.py - BERT optimizer
PaddlePaddle/LanguageModeling/BERT/optimizer.py
Triton deployment
- GPUNet deployment stages
PyTorch/Classification/GPUNet/triton/runner/pipeline.py - TFT deployment stages
PyTorch/Forecasting/TFT/triton/runner/pipeline.py - BERT deployment stages
PyTorch/LanguageModeling/BERT/triton/runner/pipeline.py
Shared runtime
- Timestamp logging
CUDA-Optimized/FastSpeech/fastspeech/utils/logging.py - Elapsed-time helper
CUDA-Optimized/FastSpeech/fastspeech/utils/time.py - NVTX instrumentation
CUDA-Optimized/FastSpeech/fastspeech/utils/nvtx.py - PyTorch utilities
CUDA-Optimized/FastSpeech/fastspeech/utils/pytorch.py
Other components
- Speech user
Connections 7
- Speech user → FastSpeech inference: requests synthesis
- TensorRT inference → TensorRT buffers: creates buffers
CUDA-Optimized/FastSpeech/fastspeech/trt/fastspeech_trt_inferencer.py - TensorRT inference → TensorRT base: uses base
CUDA-Optimized/FastSpeech/fastspeech/trt/fastspeech_trt_inferencer.py - Multi-engine inference → TensorRT inference: extends inferencer
CUDA-Optimized/FastSpeech/fastspeech/trt/fastspeech_trt_multi_engine_inferencer.py - TensorRT base → Timestamp logging: logs messages
CUDA-Optimized/FastSpeech/fastspeech/trt/trt_inferencer.py - Elapsed-time helper → Timestamp logging: prints timing
CUDA-Optimized/FastSpeech/fastspeech/utils/time.py - GPUNet deployment stages → TFT deployment stages: similar stages