chauhang

Following the steps for https://pytorch.org/docs/stable/torch.compiler_inductor_profiling.html#torchinductor-gpu-profiling


=========================================== On A10g g3.xlarge AWS instance =========================================

$> TORCHINDUCTOR_UNIQUE_KERNEL_NAMES=1 TORCHINDUCTOR_BENCHMARK_KERNEL=1 python -u benchmarks/dynamo/timm_models.py --backend inductor --amp --performance --dashboard --only mixnet_l  --disable-cudagraphs --training
/opt/conda/envs/nllm3/lib/python3.10/site-packages/transformers/utils/generic.py:441: UserWarning: torch.utils._pytree._register_pytree_node is deprecated. Please use torch.utils._pytree.register_pytree_node instead.
  _torch_pytree._register_pytree_node(
loading model: 0it [00:03, ?it/s]
cuda train mixnet_l                           

chauhang

# Open AI Gym Render does not work with default NVidia OpenGL install. Use the steps below to install NVidia Drivers / Cuda without OpenGL

# Install Pre-req
sudo apt install gcc make

# Download installers
mkdir ~/Downloads/nvidia
cd ~/Downloads/nvidia
wget https://developer.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda_10.1.243_418.87.00_linux.run
wget http://us.download.nvidia.com/XFree86/Linux-x86_64/440.36/NVIDIA-Linux-x86_64-440.36.run