From 4e47be6cd7cefedad15d9a6f54d348b81a708b62 Mon Sep 17 00:00:00 2001 From: Przemyslaw Bartosik Date: Sun, 30 Oct 2022 01:50:10 +0200 Subject: [PATCH] docs: minor changes in Resnet50 notebook --- notebooks/Resnet50-example.ipynb | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) diff --git a/notebooks/Resnet50-example.ipynb b/notebooks/Resnet50-example.ipynb index f75a2b0e64..a7d3d4eddd 100644 --- a/notebooks/Resnet50-example.ipynb +++ b/notebooks/Resnet50-example.ipynb @@ -208,8 +208,7 @@ "\n", "## 3. Running the model without optimizations\n", "\n", - "\n", - "PyTorch has a model repository called `timm`, which is a source for high quality implementations of computer vision models. We can get our EfficientNet model from there pretrained on ImageNet." + "\n" ] }, { @@ -850,7 +849,7 @@ "# The compiled module will have precision as specified by \"op_precision\".\n", "# Here, it will have FP16 precision.\n", "trt_model_fp16 = torch_tensorrt.compile(model, inputs = [torch_tensorrt.Input((128, 3, 224, 224), dtype=torch.half)],\n", - " enabled_precisions = {torch.half}, # Run with FP32\n", + " enabled_precisions = {torch.half}, # Run with FP16\n", " workspace_size = 1 << 22\n", ")\n" ] @@ -894,7 +893,7 @@ "\n", "## 5. Conclusion\n", "\n", - "In this notebook, we have walked through the complete process of compiling TorchScript models with Torch-TensorRT for EfficientNet-B0 model and test the performance impact of the optimization. With Torch-TensorRT, we observe a speedup of **1.84x** with FP32, and **5.2x** with FP16 on an NVIDIA 3090 GPU. These acceleration numbers will vary from GPU to GPU(as well as implementation to implementation based on the ops used) and we encorage you to try out latest generation of Data center compute cards for maximum acceleration.\n", + "In this notebook, we have walked through the complete process of compiling TorchScript models with Torch-TensorRT for ResNet-50 model and test the performance impact of the optimization. With Torch-TensorRT, we observe a speedup of **1.84x** with FP32, and **5.2x** with FP16 on an NVIDIA 3090 GPU. These acceleration numbers will vary from GPU to GPU(as well as implementation to implementation based on the ops used) and we encorage you to try out latest generation of Data center compute cards for maximum acceleration.\n", "\n", "### What's next\n", "Now it's time to try Torch-TensorRT on your own model. If you run into any issues, you can fill them at https://github.com/pytorch/TensorRT. Your involvement will help future development of Torch-TensorRT.\n"