[skip ci] dist.torch.launch -> torchrun (#2450)

* [skip ci] dist.torch.launch -> torchrun * Update config.yml
pytorch · Feb 3, 2022 · 052d66a · 052d66a
1 parent 6d83dd7
commit 052d66a
Show file tree

Hide file tree

Showing 2 changed files with 10 additions and 10 deletions.
diff --git a/.circleci/config.yml b/.circleci/config.yml
@@ -294,12 +294,12 @@ jobs:
             docker exec -it pthd /bin/bash -c "${test_cmd} --num_epochs=7 ${resume_opt}"
 
       - run:
-          name: "Run with NCCL backend using torch dist launch"
+          name: "Run with NCCL backend using torchrun"
           command: |
             export example_path="examples/contrib/cifar10"
             # initial run
             export stop_cmd="--stop_iteration=500"
-            export test_cmd="CI=1 python -u -m torch.distributed.launch --nproc_per_node=2 --use_env ${example_path}/main.py run --backend=nccl --checkpoint_every=200"
+            export test_cmd="CI=1 torchrun --nproc_per_node=2 ${example_path}/main.py run --backend=nccl --checkpoint_every=200"
             docker exec -it pthd /bin/bash -c "${test_cmd} ${stop_cmd}"
             # resume
             export resume_opt="--resume-from=/tmp/output-cifar10/resnet18_backend-nccl-2_stop-on-500/training_checkpoint_400.pt"

diff --git a/examples/contrib/cifar10/README.md b/examples/contrib/cifar10/README.md
@@ -52,8 +52,8 @@ If user would like to provide already downloaded dataset, the path can be setup
 Let's start training on a single node with 2 gpus:
 
 ```bash
-# using torch.distributed.launch
-python -u -m torch.distributed.launch --nproc_per_node=2 --use_env main.py run --backend="nccl"
+# using torchrun
+torchrun --nproc_per_node=2 main.py run --backend="nccl"
 ```
 
 or
@@ -92,18 +92,18 @@ Let's start training on two nodes with 2 gpus each. We assuming that master node
 1. Execute on master node
 
 ```bash
-python -u -m torch.distributed.launch \
+torchrun \
     --nnodes=2 \
     --nproc_per_node=2 \
     --node_rank=0 \
-    --master_addr=master --master_port=2222 --use_env \
+    --master_addr=master --master_port=2222 \
     main.py run --backend="nccl"
 ```
 
 2. Execute on worker node
 
 ```bash
-python -u -m torch.distributed.launch \
+torchrun \
     --nnodes=2 \
     --nproc_per_node=2 \
     --node_rank=1 \
@@ -134,14 +134,14 @@ python main.py run --resume-from=/tmp/output-cifar10/resnet18_backend-None-1_sto
 Initial training on a single node with 2 gpus with a stop on 1000 iteration (~11 epochs):
 
 ```bash
-# using torch.distributed.launch
-python -u -m torch.distributed.launch --nproc_per_node=2 --use_env main.py run --backend="nccl" --stop_iteration=1000
+# using torchrun
+torchrun --nproc_per_node=2 main.py run --backend="nccl" --stop_iteration=1000
 ```
 
 Resume from the latest checkpoint
 
 ```bash
-python -u -m torch.distributed.launch --nproc_per_node=2 --use_env main.py run --backend="nccl" \
+torchrun --nproc_per_node=2 main.py run --backend="nccl" \
     --resume-from=/tmp/output-cifar10/resnet18_backend-nccl-2_stop-on-1000/training_checkpoint_1000.pt
 ```