feat(tpu): remove nprocs from xla.spawn (#3324)

This parameter will cause issues on recent version of torch_xla.

src/accelerate/commands/launch.py

@@ -854,6 +854,7 @@ def deepspeed_launcher(args):
 
 def tpu_launcher(args):
     import torch_xla.distributed.xla_multiprocessing as xmp
+    from torch_xla import device_count
 
     if args.no_python:
         raise ValueError("--no_python cannot be used with TPU launcher")
@@ -874,13 +875,17 @@ def tpu_launcher(args):
             f"Your training script should have a function named {args.main_training_function}, or you should pass a "
             "different value to `--main_training_function`."
         )
+    if args.num_processes and args.num_processes != device_count():
+        raise ValueError(
+            f"Number of processes ({args.num_processes}) must match the number of TPU devices ({device_count()})"
+        )
 
     # Patch sys.argv
     sys.argv = [mod.__file__] + args.training_script_args
 
     main_function = getattr(mod, args.main_training_function)
     with patch_environment(**current_env):
-        xmp.spawn(PrepareForLaunch(main_function), args=(), nprocs=args.num_processes)
+        xmp.spawn(PrepareForLaunch(main_function), args=())
 
 
 def tpu_pod_launcher(args):

src/accelerate/launchers.py

@@ -135,19 +135,18 @@ def train(*args):
     if (in_colab or in_kaggle) and (os.environ.get("TPU_NAME", None) is not None):
         # TPU launch
         import torch_xla.distributed.xla_multiprocessing as xmp
+        from torch_xla import device_count
 
         if len(AcceleratorState._shared_state) > 0:
             raise ValueError(
                 "To train on TPU in Colab or Kaggle Kernel, the `Accelerator` should only be initialized inside "
                 "your training function. Restart your notebook and make sure no cells initializes an "
                 "`Accelerator`."
             )
-        if num_processes is None:
-            num_processes = 8
 
         launcher = PrepareForLaunch(function, distributed_type="XLA")
-        print(f"Launching a training on {num_processes} TPU cores.")
-        xmp.spawn(launcher, args=args, nprocs=num_processes, start_method="fork")
+        print(f"Launching a training on {device_count()} TPU cores.")
+        xmp.spawn(launcher, args=args, start_method="fork")
     elif in_colab and get_gpu_info()[1] < 2:
         # No need for a distributed launch otherwise as it's either CPU or one GPU.
         if torch.cuda.is_available():

tests/xla_spawn.py

@@ -30,6 +30,7 @@
 from pathlib import Path
 
 import torch_xla.distributed.xla_multiprocessing as xmp
+from torch_xla import device_count
 
 
 def parse_args():
@@ -46,7 +47,13 @@ def parse_args():
     )
 
     # Optional arguments for the launch helper
-    parser.add_argument("--num_cores", type=int, default=1, help="Number of TPU cores to use (1 or 8).")
+    num_devices = device_count()
+    parser.add_argument(
+        "--num_cores",
+        type=int,
+        default=num_devices,
+        help="Number of TPU cores to use (1 or number of available devices).",
+    )
 
     # positional
     parser.add_argument(
@@ -76,7 +83,12 @@ def main():
     mod = importlib.import_module(mod_name)
 
     # Patch sys.argv
-    sys.argv = [args.training_script] + args.training_script_args + ["--tpu_num_cores", str(args.num_cores)]
+    sys.argv = [args.training_script] + args.training_script_args
+    num_cores = args.num_cores
+    if num_cores == device_count() and num_cores != 1:
+        # There is an error in xmp.spawn that causes it to fail when num_cores is specified and not 1, so we set it to
+        # None when it matches the number of devices.
+        num_cores = None
     xmp.spawn(mod._mp_fn, args=(), nprocs=args.num_cores)