Fix FSDP for Models with Frozen Weights

test/test_torch_distributed_fsdp_frozen_weight.py

@@ -0,0 +1,32 @@
+import sys
+import torch
+import torch.nn as nn
+import torch_xla.core.xla_model as xm
+import torch_xla.distributed.xla_multiprocessing as xmp
+from torch_xla.distributed.fsdp import XlaFullyShardedDataParallel as FSDP
+
+
+def _mp_fn(index):
+  dev = xm.xla_device()
+  if xm.xla_device_hw(dev) not in ('TPU', 'GPU'):
+    print(
+        'Default device {} is not a TPU or GPU device'.format(dev),
+        file=sys.stderr)
+    return
+
+  model = nn.Linear(1024, 1024)
+  model.weight.requires_grad = False  # the weight param is frozen
+
+  model = FSDP(model)  # wrapping the linear module with FSDP
+
+  input = torch.rand((2, 1024), device=xm.xla_device())
+
+  output = model(input)
+  loss = torch.sum(output)
+  loss.backward()
+  assert not any(p._has_full_param for p in model.full_params), \
+    'Expecting all the full params to be freed at this moment.'
+
+
+if __name__ == "__main__":
+  xmp.spawn(_mp_fn, args=())

torch_xla/distributed/fsdp/xla_fully_sharded_data_parallel.py

@@ -1299,13 +1299,19 @@ def _wait_for_post_backward(self) -> None:
     # A backward pass is done, clean up below.
     def _finalize_parameters(fsdp_module: XlaFullyShardedDataParallel) -> None:
       """Helper used below on all fsdp modules."""
+      frozen_params = []
       for p in fsdp_module.full_params:
         if not p.requires_grad:
-          continue
+          frozen_params.append(p)
         if hasattr(p, "_shard_bwd_hook"):
           assert len(p._shard_bwd_hook) == 2, len(p._shard_bwd_hook)
           p._shard_bwd_hook[1].remove()
           delattr(p, "_shard_bwd_hook")
+      # Free the full params with `requires_grad==False`
+      if frozen_params:
+        fsdp_module._free_full_params(
+            frozen_params,
+            apply_opt_barrier=self.optimization_barrier_in_backward)
 
     # Update root and nested FSDP's hooks and flags.
     for m in self.modules():  # includes self