Rename master to main everwhere (#162)

Summary:
Tested by running a CIFAR sweep

Pull Request resolved: #162

Reviewed By: pdollar

Differential Revision: D30412710

Pulled By: mannatsingh

fbshipit-source-id: 2eedd709a5fa7463f487c5e5f861f2fb519b0fd8

docs/CONTRIBUTING.md

@@ -5,7 +5,7 @@ possible.
 ## Pull Requests
 We actively welcome your pull requests.
 
-1. Fork the repo and create your branch from `master`.
+1. Fork the repo and create your branch from `main`.
 2. If you've added code that should be tested, add tests.
 3. If you've changed APIs, update the documentation.
 4. Ensure the test suite passes.

pycls/core/checkpoint.py

@@ -57,8 +57,8 @@ def has_checkpoint():
 
 def save_checkpoint(model, model_ema, optimizer, epoch, test_err, ema_err):
     """Saves a checkpoint and also the best weights so far in a best checkpoint."""
-    # Save checkpoints only from the master process
-    if not dist.is_master_proc():
+    # Save checkpoints only from the main process
+    if not dist.is_main_proc():
         return
     # Ensure that the checkpoint dir exists
     pathmgr.mkdirs(get_checkpoint_dir())

pycls/core/distributed.py

@@ -38,15 +38,15 @@ def __call__(self):
         self.fun()
 
 
-def is_master_proc(local=False):
+def is_main_proc(local=False):
     """
-    Determines if the current process is the master process.
+    Determines if the current process is the main process.
 
-    Master process is responsible for logging, writing and loading checkpoints. In
-    the multi GPU setting, we assign the master role to the rank 0 process. When
-    training using a single GPU, there is a single process which is considered master.
+    Main process is responsible for logging, writing and loading checkpoints. In
+    the multi GPU setting, we assign the main role to the rank 0 process. When
+    training using a single GPU, there is a single process which is considered main.
 
-    If local==True, then check if the current process is the master on the current node.
+    If local==True, then check if the current process is the main on the current node.
     """
     m = cfg.MAX_GPUS_PER_NODE if local else cfg.NUM_GPUS
     return cfg.NUM_GPUS == 1 or torch.distributed.get_rank() % m == 0
@@ -95,10 +95,10 @@ def setup_distributed(cfg_state):
     torch.cuda.set_device(local_rank)
 
 
-def single_proc_run(local_rank, fun, master_port, cfg_state, world_size):
+def single_proc_run(local_rank, fun, main_port, cfg_state, world_size):
     """Executes fun() on a single GPU in a multi-GPU setup."""
     os.environ["MASTER_ADDR"] = "localhost"
-    os.environ["MASTER_PORT"] = str(master_port)
+    os.environ["MASTER_PORT"] = str(main_port)
     os.environ["RANK"] = str(local_rank)
     os.environ["LOCAL_RANK"] = str(local_rank)
     os.environ["WORLD_SIZE"] = str(world_size)
@@ -129,15 +129,15 @@ def multi_proc_run(num_proc, fun):
             slurm_constraint=launch.GPU_TYPE,
             slurm_additional_parameters={"mail-user": launch.EMAIL, "mail-type": "END"},
         )
-        master_port = random.randint(cfg.PORT_RANGE[0], cfg.PORT_RANGE[1])
-        job = executor.submit(SubmititRunner(master_port, fun, cfg))
+        main_port = random.randint(cfg.PORT_RANGE[0], cfg.PORT_RANGE[1])
+        job = executor.submit(SubmititRunner(main_port, fun, cfg))
         print("Submitted job_id {} with out_dir: {}".format(job.job_id, cfg.OUT_DIR))
         if not use_slurm:
             job.wait()
     elif num_proc > 1:
-        master_port = random.randint(cfg.PORT_RANGE[0], cfg.PORT_RANGE[1])
+        main_port = random.randint(cfg.PORT_RANGE[0], cfg.PORT_RANGE[1])
         mp_runner = torch.multiprocessing.start_processes
-        args = (fun, master_port, cfg, num_proc)
+        args = (fun, main_port, cfg, num_proc)
         # Note: using "fork" below, "spawn" causes time and error regressions. Using
         # spawn changes the default multiprocessing context to spawn, which doesn't
         # interact well with the dataloaders (likely due to the use of OpenCV).

pycls/core/logging.py

@@ -43,8 +43,8 @@ def ignore(*_objects, _sep=" ", _end="\n", _file=sys.stdout, _flush=False):
 
 def setup_logging():
     """Sets up the logging."""
-    # Enable logging only for the master process
-    if dist.is_master_proc():
+    # Enable logging only for the main process
+    if dist.is_main_proc():
         # Clear the root logger to prevent any existing logging config
         # (e.g. set by another module) from messing with our setup
         logging.root.handlers = []

pycls/core/trainer.py

@@ -33,7 +33,7 @@
 
 def setup_env():
     """Sets up environment for training or testing."""
-    if dist.is_master_proc():
+    if dist.is_main_proc():
         # Ensure that the output dir exists
         pathmgr.mkdirs(cfg.OUT_DIR)
         # Save the config
@@ -76,7 +76,7 @@ def setup_model():
 
 def get_weights_file(weights_file):
     """Download weights file if stored as a URL."""
-    download = dist.is_master_proc(local=True)
+    download = dist.is_main_proc(local=True)
     weights_file = cache_url(weights_file, cfg.DOWNLOAD_CACHE, download=download)
     if cfg.NUM_GPUS > 1:
         torch.distributed.barrier()

pycls/models/model_zoo.py

@@ -19,7 +19,7 @@
 _URL_WEIGHTS = "https://dl.fbaipublicfiles.com/pycls"
 
 # URL prefix for model config files
-_URL_CONFIGS = "https://raw.githubusercontent.com/facebookresearch/pycls/master/configs"
+_URL_CONFIGS = "https://raw.githubusercontent.com/facebookresearch/pycls/main/configs"
 
 # Model weights download cache directory
 _DOWNLOAD_CACHE = "/tmp/pycls-download-cache"

tools/sweep_launch_job.py

@@ -67,7 +67,7 @@ def main():
     # Print info about run
     job_id = os.environ["SLURM_ARRAY_JOB_ID"]
     task_id = os.environ["SLURM_ARRAY_TASK_ID"]
-    prt("Job array master job ID: {}".format(job_id))
+    prt("Job array main job ID: {}".format(job_id))
     prt("Job array task ID (index): {}".format(task_id))
     prt("Running job on: {}".format(str(os.uname())))
     # Load what we need