[Intel HPU] Support intel hpu platform (#4161)

* [Intel HPU] Support intel hpu platform

* fix some issues

* apply precommit and move AttentionBackend_HPU

* fix format issue

* correct ops import

* fix ci issue

* update code in layers

* fix code style issue

* remove dense tp moe ep mode

* fix enc_dec_block_num

* fix rebase issue

* rename hpu to gaudi in readme

* rename ForwardMeta_HPU to HPUForwardMeta

fastdeploy/distributed/communication.py

@@ -66,3 +66,26 @@ try:
 
 except:
     tensor_model_parallel_all_reduce = None
+
+from paddle.distributed.communication import stream
+from paddle.distributed.communication.reduce import ReduceOp
+
+
+def all_reduce(
+    tensor,
+    op,
+    group,
+    sync_op: bool = True,
+):
+    return stream.all_reduce(tensor, op=op, group=group, sync_op=sync_op, use_calc_stream=True)
+
+
+@paddle.jit.marker.unified
+def tensor_model_parallel_all_reduce_custom(input_: paddle.Tensor) -> paddle.Tensor:
+    """All-reduce the input tensor across model parallel group on calc stream."""
+    if paddle.in_dynamic_mode():
+        hcg = dist.fleet.get_hybrid_communicate_group()
+        mp_group = hcg.get_model_parallel_group()
+        all_reduce(input_, op=ReduceOp.SUM, group=mp_group)
+    else:
+        dist.all_reduce(input_)