[BE] remove old pytorch version warning on strided sharding since 2.5…

… is official released (#665)

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at
bottom):
* __->__ #665

#507 added a PyTorch version check when users try to use FSDP+TP, to
make sure the right PT version includes DTensor strided sharding which
assures correct DTensor checkpoint. Since PyTorch 2.5 is official
released and strided sharding is included in 2.5, we can safely remove
this warning.

torchtitan/parallelisms/parallelize_llama.py

@@ -34,7 +34,6 @@
 from torchtitan.config_manager import JobConfig, TORCH_DTYPE_MAP
 from torchtitan.logging import logger
 from torchtitan.parallelisms.parallel_dims import ParallelDims
-from torchtitan.parallelisms.utils import check_strided_sharding_enabled
 
 
 def parallelize_llama(
@@ -330,12 +329,6 @@ def apply_fsdp(
     if cpu_offload:
         fsdp_config["offload_policy"] = CPUOffloadPolicy()
 
-    # TODO: remove this check once PyTorch 2.5 is released. We can safely assume
-    # that users won't use a nightly build which is older than 20240809 by then.
-    if tp_enabled:
-        # check if strided sharding is enabled, which is necessary for 2D/3D DCP
-        check_strided_sharding_enabled()
-
     for layer_id, transformer_block in model.layers.items():
         if pp_enabled:
             # For PP, do not reshard after forward to avoid per-microbatch