diff --git a/fbgemm_gpu/src/ssd_split_embeddings_cache/embedding_rocksdb_wrapper.h b/fbgemm_gpu/src/ssd_split_embeddings_cache/embedding_rocksdb_wrapper.h
index 4ca404c157..80c2e5a967 100644
--- a/fbgemm_gpu/src/ssd_split_embeddings_cache/embedding_rocksdb_wrapper.h
+++ b/fbgemm_gpu/src/ssd_split_embeddings_cache/embedding_rocksdb_wrapper.h
@@ -213,6 +213,10 @@ class EmbeddingRocksDBWrapper : public torch::jit::CustomClassHolder {
     impl_->release_snapshot(handle);
   }
 
+  void delete_rocksdb_checkpoint_dir() {
+    impl_->delete_rocksdb_checkpoint_dir();
+  }
+
   int64_t get_snapshot_count() const {
     return impl_->get_snapshot_count();
   }
diff --git a/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper.h b/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper.h
index 1c14d3f809..2d1601941c 100644
--- a/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper.h
+++ b/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper.h
@@ -121,7 +121,7 @@ class KVTensorWrapper : public torch::jit::CustomClassHolder {
   void deserialize(const std::string& serialized);
 
   std::vector<std::string> get_kvtensor_serializable_metadata() const;
-
+  void delete_rocksdb_checkpoint_dir() const;
   friend void to_json(json& j, const KVTensorWrapper& kvt);
   friend void from_json(const json& j, KVTensorWrapper& kvt);
 
diff --git a/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper_cpu.cpp b/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper_cpu.cpp
index 2a91a1faa2..de1e78e9e6 100644
--- a/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper_cpu.cpp
+++ b/fbgemm_gpu/src/ssd_split_embeddings_cache/kv_tensor_wrapper_cpu.cpp
@@ -108,4 +108,10 @@ std::string KVTensorWrapper::layout_str() {
   oss << options_.layout();
   return oss.str();
 }
+
+std::vector<std::string> KVTensorWrapper::get_kvtensor_serializable_metadata()
+    const {
+  FBEXCEPTION("Not implemented");
+  return std::vector<std::string>{};
+}
 } // namespace ssd
diff --git a/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_split_table_batched_embeddings.cpp b/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_split_table_batched_embeddings.cpp
index 541ae73c9d..3a764e8d0f 100644
--- a/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_split_table_batched_embeddings.cpp
+++ b/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_split_table_batched_embeddings.cpp
@@ -412,9 +412,24 @@ std::string KVTensorWrapper::serialize() const {
   return json_serialized.dump();
 }
 
+void KVTensorWrapper::delete_rocksdb_checkpoint_dir() const {
+  if (readonly_db_) {
+    LOG(INFO) << "deleting checkpoint dir for " << checkpoint_uuid;
+    readonly_db_->delete_rocksdb_checkpoint_dir();
+  } else if (db_) {
+    auto* db = dynamic_cast<EmbeddingRocksDB*>(db_.get());
+    LOG(INFO) << "embedding delete";
+    db->delete_rocksdb_checkpoint_dir();
+  }
+}
+
 std::vector<std::string> KVTensorWrapper::get_kvtensor_serializable_metadata()
     const {
   std::vector<std::string> metadata;
+  // Return empty metadata if checkpoint_handle_ is not initialized yet
+  if (checkpoint_handle_ == nullptr) {
+    return metadata;
+  }
   auto* db = dynamic_cast<EmbeddingRocksDB*>(db_.get());
   auto checkpoint_paths = db->get_checkpoints(checkpoint_handle_->uuid);
   metadata.push_back(std::to_string(checkpoint_paths.size()));
@@ -931,6 +946,9 @@ static auto embedding_rocks_db_wrapper =
             &EmbeddingRocksDBWrapper::wait_util_filling_work_done)
         .def("create_snapshot", &EmbeddingRocksDBWrapper::create_snapshot)
         .def("release_snapshot", &EmbeddingRocksDBWrapper::release_snapshot)
+        .def(
+            "delete_rocksdb_checkpoint_dir",
+            &EmbeddingRocksDBWrapper::delete_rocksdb_checkpoint_dir)
         .def("get_snapshot_count", &EmbeddingRocksDBWrapper::get_snapshot_count)
         .def(
             "get_keys_in_range_by_snapshot",
@@ -1142,7 +1160,10 @@ static auto kv_tensor_wrapper =
         .def("logs", &KVTensorWrapper::logs, "")
         .def(
             "get_kvtensor_serializable_metadata",
-            &KVTensorWrapper::get_kvtensor_serializable_metadata);
+            &KVTensorWrapper::get_kvtensor_serializable_metadata)
+        .def(
+            "delete_rocksdb_checkpoint_dir",
+            &KVTensorWrapper::delete_rocksdb_checkpoint_dir);
 
 TORCH_LIBRARY_FRAGMENT(fbgemm, m) {
   m.def(
diff --git a/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_table_batched_embeddings.h b/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_table_batched_embeddings.h
index 050f046767..d1ed713f82 100644
--- a/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_table_batched_embeddings.h
+++ b/fbgemm_gpu/src/ssd_split_embeddings_cache/ssd_table_batched_embeddings.h
@@ -345,6 +345,13 @@ class EmbeddingRocksDB : public kv_db::EmbeddingKVDB {
     return result;
   }
 
+  void delete_rocksdb_checkpoint_dir() {
+    for (auto shard = 0; shard < dbs_.size(); ++shard) {
+      LOG(INFO) << "removing checkpoint directories: " << db_paths_[shard];
+      kv_db_utils::remove_dir(db_paths_[shard]);
+    }
+  }
+
   void initialize_dbs(
       int64_t num_shards,
       std::string path,