diff --git a/samples/transformers-auto-model/Qwen1.5-0.5B/graph_net.json b/samples/transformers-auto-model/Qwen1.5-0.5B/graph_net.json
index 1dc8f04d6..040a66cde 100644
--- a/samples/transformers-auto-model/Qwen1.5-0.5B/graph_net.json
+++ b/samples/transformers-auto-model/Qwen1.5-0.5B/graph_net.json
@@ -4,5 +4,5 @@
     "num_nodes_required": 1,
     "dynamic": false,
     "model_name": "Qwen/Qwen1.5-0.5B",
-    "heuristic_tag": "unknown"
-}
\ No newline at end of file
+    "heuristic_tag": "nlp"
+}
diff --git a/samples/transformers-auto-model/joeddav_xlm-roberta-large-xnli/graph_net.json b/samples/transformers-auto-model/joeddav_xlm-roberta-large-xnli/graph_net.json
new file mode 100644
index 000000000..bb6c06e2f
--- /dev/null
+++ b/samples/transformers-auto-model/joeddav_xlm-roberta-large-xnli/graph_net.json
@@ -0,0 +1,7 @@
+{
+    "framework": "torch",
+    "num_devices_required": 1,
+    "num_nodes_required": 1,
+    "source": "huggingface_hub",
+    "heuristic_tag": "nlp"
+}
diff --git a/samples/transformers-auto-model/joeddav_xlm-roberta-large-xnli/input_meta.py b/samples/transformers-auto-model/joeddav_xlm-roberta-large-xnli/input_meta.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/tools/check_and_count_samples.py b/tools/check_and_count_samples.py
new file mode 100644
index 000000000..f4ab5cc1c
--- /dev/null
+++ b/tools/check_and_count_samples.py
@@ -0,0 +1,130 @@
+import os
+import json
+
+
+def check_completeness(samples_dir):
+    samples_missing_hash = []
+    samples_missing_json = []
+    samples_missing_meta = []
+    for root, dirs, files in os.walk(samples_dir):
+        if "shape_patches_" not in root and "model.py" in files:
+            model_path = root
+            if not os.path.exists(os.path.join(model_path, "graph_hash.txt")):
+                samples_missing_hash.append(model_path)
+            if not os.path.exists(os.path.join(model_path, "graph_net.json")):
+                samples_missing_json.append(model_path)
+            if not os.path.exists(
+                os.path.join(model_path, "input_meta.py")
+            ) or not os.path.exists(os.path.join(model_path, "weight_meta.py")):
+                samples_missing_meta.append(model_path)
+
+    all_samples_complete = (
+        len(samples_missing_hash) == 0
+        and len(samples_missing_json) == 0
+        and len(samples_missing_meta) == 0
+    )
+
+    if not all_samples_complete:
+        print(f"Check completeness result for {samples_dir}:")
+        print(f"1. {len(samples_missing_hash)} samples missing graph_hash.txt")
+        for model_path in samples_missing_hash:
+            print(f"  - {model_path}")
+
+        print(f"2. {len(samples_missing_json)} samples missing graph_net.json")
+        for model_path in samples_missing_json:
+            print(f"  - {model_path}")
+
+        print(
+            f"3. {len(samples_missing_meta)} samples missing input_meta.py or weight_meta.py"
+        )
+        for model_path in samples_missing_meta:
+            print(f"  - {model_path}")
+        print()
+
+    return all_samples_complete
+
+
+def check_redandancy(samples_dir):
+    graph_hash2model_paths = {}
+    for root, dirs, files in os.walk(samples_dir):
+        if "graph_hash.txt" in files:
+            model_path = root
+            graph_hash_path = os.path.join(model_path, "graph_hash.txt")
+            graph_hash = open(graph_hash_path).read()
+            if graph_hash not in graph_hash2model_paths.keys():
+                graph_hash2model_paths[graph_hash] = [model_path]
+            else:
+                graph_hash2model_paths[graph_hash].append(model_path)
+
+    has_duplicates = False
+    print(f"Totally {len(graph_hash2model_paths)} unique graphs under {samples_dir}.")
+    for graph_hash, model_paths in graph_hash2model_paths.items():
+        graph_hash2model_paths[graph_hash] = sorted(model_paths)
+        if len(model_paths) > 1:
+            has_duplicates = True
+            print(f"Redundant models detected for grap_hash {graph_hash}:")
+            for model_path in model_paths:
+                print(f"    {model_path}")
+    return has_duplicates, graph_hash2model_paths
+
+
+def count_samples(samples_dir, framework):
+    model_sources = os.listdir(samples_dir)
+
+    graph_net_count = 0
+    graph_net_dict = {}
+    model_names_set = set()
+    for source in model_sources:
+        source_dir = os.path.join(samples_dir, source)
+        if os.path.isdir(source_dir):
+            graph_net_dict[source] = 0
+            for root, dirs, files in os.walk(source_dir):
+                if "graph_net.json" in files:
+                    with open(os.path.join(root, "graph_net.json"), "r") as f:
+                        data = json.load(f)
+                        model_name = data.get("model_name", None)
+                    if model_name is not None and model_name != "NO_VALID_MATCH_FOUND":
+                        if model_name not in model_names_set:
+                            model_names_set.add(model_name)
+                            graph_net_count += 1
+                            graph_net_dict[source] += 1
+                    else:
+                        graph_net_count += 1
+                        graph_net_dict[source] += 1
+
+    print(f"Number of {framework} samples: {graph_net_count}")
+    for name, number in graph_net_dict.items():
+        print(f"- {name:24}: {number}")
+    print()
+
+
+def main():
+    filename = os.path.abspath(__file__)
+    root_dir = os.path.dirname(os.path.dirname(filename))
+
+    framework2dirname = {
+        "torch": "samples",
+        "paddle": "paddle_samples",
+    }
+
+    all_samples_complete = True
+    for samples_dirname in framework2dirname.values():
+        samples_dir = os.path.join(root_dir, samples_dirname)
+        all_samples_complete = all_samples_complete and check_completeness(samples_dir)
+    assert all_samples_complete, "Please fix the incompleted samples!"
+
+    all_samples_has_duplicates = False
+    for samples_dirname in framework2dirname.values():
+        samples_dir = os.path.join(root_dir, samples_dirname)
+        has_duplicates, graph_hash2model_paths = check_redandancy(samples_dir)
+        all_samples_has_duplicates = all_samples_has_duplicates or has_duplicates
+    print()
+    assert not all_samples_has_duplicates, "Please remove the redundant samples!"
+
+    for framework in framework2dirname.keys():
+        samples_dir = os.path.join(root_dir, framework2dirname[framework])
+        count_samples(samples_dir, framework)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/tools/ci/check_validate.sh b/tools/ci/check_validate.sh
index f443cd9ea..64e45119a 100644
--- a/tools/ci/check_validate.sh
+++ b/tools/ci/check_validate.sh
@@ -41,7 +41,6 @@ function prepare_torch_env() {
     pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu126 > /dev/null
     [ $? -ne 0 ] && LOG "[FATAL] Install torch2.9.0 failed!" && exit -1
   else
-    python ${GRAPH_NET_EXTRACT_WORKSPACE}/tools/count_sample.py
     LOG "[INFO] This pull request doesn't change any torch samples, skip the CI."
   fi
 }
@@ -62,7 +61,6 @@ function prepare_paddle_env() {
     [ $? -ne 0 ] && LOG "[FATAL] Install paddlepaddle-develop failed!" && exit -1
     python -c "import paddle; print('[PaddlePaddle Commit]', paddle.version.commit)"
   else
-    python ${GRAPH_NET_EXTRACT_WORKSPACE}/tools/count_sample.py
     LOG "[INFO] This pull request doesn't change any paddle samples, skip the CI."
   fi
 }
@@ -165,7 +163,8 @@ function main() {
   check_validation_info=$(check_paddle_validation)
   check_validation_code=$?
   summary_problems $check_validation_code "$check_validation_info"
-  python ${GRAPH_NET_EXTRACT_WORKSPACE}/tools/count_sample.py
+  python ${GRAPH_NET_EXTRACT_WORKSPACE}/tools/check_and_count_samples.py >&2
+  [ $? -ne 0 ] && LOG "[FATAL] Check completeness or redundancy failed!" && exit -1
   LOG "[INFO] check_validation run success and no error!"
 }
 
diff --git a/tools/count_sample.py b/tools/count_sample.py
deleted file mode 100644
index 3addb4d5a..000000000
--- a/tools/count_sample.py
+++ /dev/null
@@ -1,40 +0,0 @@
-import os
-import json
-
-
-filename = os.path.abspath(__file__)
-root_dir = os.path.dirname(os.path.dirname(filename))
-framework2dirname = {
-    "torch": "samples",
-    "paddle": "paddle_samples",
-}
-
-for framework in ["torch", "paddle"]:
-    samples_dir = os.path.join(root_dir, framework2dirname[framework])
-    model_categories = os.listdir(samples_dir)
-
-    graph_net_count = 0
-    graph_net_dict = {}
-    model_names_set = set()
-    for category in model_categories:
-        category_dir = os.path.join(samples_dir, category)
-        if os.path.isdir(category_dir):
-            graph_net_dict[category] = 0
-            for root, dirs, files in os.walk(category_dir):
-                if "graph_net.json" in files:
-                    with open(os.path.join(root, "graph_net.json"), "r") as f:
-                        data = json.load(f)
-                        model_name = data.get("model_name", None)
-                    if model_name is not None:
-                        if model_name not in model_names_set:
-                            model_names_set.add(model_name)
-                            graph_net_count += 1
-                            graph_net_dict[category] += 1
-                    else:
-                        graph_net_count += 1
-                        graph_net_dict[category] += 1
-
-    print(f"Number of {framework} samples: {graph_net_count}")
-    for name, number in graph_net_dict.items():
-        print(f"- {name:24}: {number}")
-    print()