diff --git a/torchrec/hybrid_torchrec/src/ids_process/bucketize.cpp b/torchrec/hybrid_torchrec/src/ids_process/bucketize.cpp
index 44526df01cb04a16c4bba0e4677d0b583ce9bf66..f657d8fef97233b3a27feaf2e6e9920aefc025c3 100644
--- a/torchrec/hybrid_torchrec/src/ids_process/bucketize.cpp
+++ b/torchrec/hybrid_torchrec/src/ids_process/bucketize.cpp
@@ -199,7 +199,7 @@ void BlockBucketizeSparseFeaturesCpuKernel(const at::Tensor& lengths, const at::
 
     // 去重逻辑 (需要时启用)
     if constexpr (DoUnique) {
-        auto* idsCountData = GetSafeDataPtr<IndexT>(idsCounts, "idsCounts");
+        auto* idsCountData = ReturnCount ? GetSafeDataPtr<IndexT>(idsCounts, "idsCounts") : nullptr;
         int64_t uniqueSize = Deduplicate<OffsetT, IndexT, ReturnCount>(
             newLengthsData, newOffsetsData, offsetsData, indicesData, newIndicesData,
             unbucketizePermuteData, numFeatures, batchSize, bucketSize, idsCountData);
diff --git a/torchrec/hybrid_torchrec/test/dt/test_bucketize_kjt.py b/torchrec/hybrid_torchrec/test/dt/test_bucketize_kjt.py
index 6256ab8fd868efc29ef02010fabdffea89d25263..1f00d35390210ee8759070db916355884a88c3ef 100644
--- a/torchrec/hybrid_torchrec/test/dt/test_bucketize_kjt.py
+++ b/torchrec/hybrid_torchrec/test/dt/test_bucketize_kjt.py
@@ -26,3 +26,29 @@ class TestBucketizeKJTBeforeAll2All:
         block_sizes = [kjt.lengths()[i * length: i * length + length].sum() for i in range(len(kjt.keys()))]
         block_sizes = torch.tensor(block_sizes)
         bucketize_kjt_before_all2all(kjt, world_size, block_sizes)
+
+    @pytest.mark.parametrize("world_size", [1, 2])
+    def test_do_unique_without_admit(self, world_size):
+        kjt = self.create_kjt()
+        length = len(kjt.lengths()) // len(kjt.keys())
+        block_sizes = [kjt.lengths()[i * length: i * length + length].sum() for i in range(len(kjt.keys()))]
+        block_sizes = torch.tensor(block_sizes)
+
+        bucketized, _ = bucketize_kjt_before_all2all(
+            kjt, world_size, block_sizes, output_permute=True, do_unique=True,
+        )
+
+        assert isinstance(bucketized, KeyedJaggedTensor)
+        # keys 应在分桶后按 bucket 重复
+        assert len(bucketized.keys()) == len(kjt.keys()) * world_size
+
+        # 校验：分桶且去重后，每个 key 段内不应存在重复值
+        b_vals = bucketized.values()
+        b_lens = bucketized.lengths().view(-1).tolist()
+        pos = 0
+        for length in b_lens:
+            if length == 0:
+                continue
+            seg = b_vals[pos:pos + length]
+            assert torch.unique(seg).numel() == seg.numel()
+            pos += length
diff --git a/torchrec/hybrid_torchrec/test/dt/test_ids_process.py b/torchrec/hybrid_torchrec/test/dt/test_ids_process.py
index 1b9a2bc1658dac894260cce157a4f5271b29c023..6d91a4a7ded56bcff7c4aa60c120f6e103b5661d 100644
--- a/torchrec/hybrid_torchrec/test/dt/test_ids_process.py
+++ b/torchrec/hybrid_torchrec/test/dt/test_ids_process.py
@@ -151,8 +151,10 @@ def test_ids2indices_out(input_size, pin_memory, num_mapper):
             unique_start = unique_offset[i].item()
             unique_end = unique_offset[i + 1].item()
             unique_this = unique[unique_start:unique_end]
+            unique_ids_this = unique_ids[unique_start:unique_end]
             unique_inverse_this = unique_inverse[start:end]
             verify_unique(indices, unique_this, unique_inverse_this)
+            assert torch.unique(unique_ids_this).numel() == unique_ids_this.numel(), "unique_ids_this is not unique"
 
 
 @pytest.mark.parametrize("input_size", [10000])
diff --git a/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/embedding.py b/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/embedding.py
index 46597a728c9297305f64e5e83e87326f948d538e..15f9237bd2cb1621fdb25d1fdc504eb09e42d9f2 100644
--- a/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/embedding.py
+++ b/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/embedding.py
@@ -323,10 +323,12 @@ class EmbCacheShardedEmbeddingCollection(ShardedEmbeddingCollection):
             module.embedding_configs()
         )
         self._table_names: List[str] = [
-            config.name for config in self._embedding_configs
+            config.name 
+            for config in self._embedding_configs
         ]
         self._table_name_to_config: Dict[str, EmbCacheEmbeddingConfig] = {
-            config.name: config for config in self._embedding_configs
+            config.name: config 
+            for config in self._embedding_configs
         }
         self.module_sharding_plan: EmbeddingModuleShardingPlan = cast(
             EmbeddingModuleShardingPlan,
@@ -503,7 +505,8 @@ class EmbCacheShardedEmbeddingCollection(ShardedEmbeddingCollection):
             for emb_table in lookup.grouped_configs[0].embedding_tables
         ]
         emb_names: List[str] = [
-            emb_table.name for emb_table in lookup.grouped_configs[0].embedding_tables
+            emb_table.name 
+            for emb_table in lookup.grouped_configs[0].embedding_tables
         ]
         emb_not_admitted_default_value: List[float] = []
         for emb_name in emb_names:
diff --git a/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sequence_sharding.py b/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sequence_sharding.py
index b0b268dd7abc3c12b517ed4997547486bcd1c385..b6b040bf457b044f0a8624a387a143a796478866 100644
--- a/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sequence_sharding.py
+++ b/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sequence_sharding.py
@@ -70,6 +70,7 @@ class EmbCacheRwSequenceEmbeddingSharding(RwSequenceEmbeddingSharding):
             has_feature_processor=self._has_feature_processor,
             need_pos=self._need_pos,
             enable_admit=self._enable_admit,
+            is_ec=True,
         )
 
     def create_lookup(
diff --git a/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sharding.py b/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sharding.py
index 3df6adf78739b95b2e92f398acdb0e2ca6e7b45e..fe3ef5ee2067edf86356449b6dc36d6c2246dde9 100644
--- a/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sharding.py
+++ b/torchrec/torchrec_embcache/src/torchrec_embcache/distributed/sharding/rw_sharding.py
@@ -89,6 +89,7 @@ class EmbCacheRwSparseFeaturesDist(RwSparseFeaturesDist):
         need_pos: bool = False,
         keep_original_indices: bool = False,
         enable_admit: bool = False,
+        is_ec: bool = False,
     ) -> None:
         super().__init__(
             pg,
@@ -105,8 +106,7 @@ class EmbCacheRwSparseFeaturesDist(RwSparseFeaturesDist):
 
         # local unique只可用于EC(Embedding Collection / Sequence Embedding)
         yes_str = ("true", "1", "yes")
-        self._do_unique = os.environ.get("DO_EC_LOCAL_UNIQUE", "False").lower() in yes_str and \
-                          os.environ.get("USE_EC", "False").lower() in yes_str
+        self._do_unique = os.environ.get("DO_EC_LOCAL_UNIQUE", "False").lower() in yes_str and is_ec
 
         self._enable_admit = enable_admit
 
diff --git a/torchrec/torchrec_embcache/tests/acc_test/run_test.sh b/torchrec/torchrec_embcache/tests/acc_test/run_test.sh
index 9d046b90f3b7367d9d4149886fff11ba61ef61ea..653c0dca629101bc28a264f107453007e53730cd 100644
--- a/torchrec/torchrec_embcache/tests/acc_test/run_test.sh
+++ b/torchrec/torchrec_embcache/tests/acc_test/run_test.sh
@@ -48,5 +48,5 @@ export ASCEND_RT_VISIBLE_DEVICES=6,7
 pytest ./test_embedding_cache_pipeline.py
 pytest ./test_embedding_ec_cache_pipeline.py
 
-# export DO_EC_LOCAL_UNIQUE=1
-# pytest ./test_feature_filter.py
+export DO_EC_LOCAL_UNIQUE=1
+pytest ./test_embedding_ec_cache_pipeline.py