diff --git a/docker/Dockerfile.ascend b/docker/Dockerfile.ascend
index 911ee3f9..f80dbdd2 100644
--- a/docker/Dockerfile.ascend
+++ b/docker/Dockerfile.ascend
@@ -86,13 +86,13 @@ RUN pip uninstall ms-swift modelscope -y && pip install --no-cache-dir pip==23.*
 if [ "$INSTALL_MS_DEPS" = "True" ]; then \
     pip install --no-cache-dir omegaconf==2.0.6 && \
     pip install 'editdistance==0.8.1' && \
-    pip install --no-cache-dir 'cython<=0.29.36' versioneer 'numpy<2.0' && \
+    pip install --no-cache-dir 'cython>=3.0' versioneer 'numpy>=2.0' && \
     pip install --no-cache-dir -r /var/modelscope/framework.txt && \
     pip install --no-cache-dir -r /var/modelscope/audio.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \
     pip install --no-cache-dir -r /var/modelscope/tests.txt && \
     pip install --no-cache-dir -r /var/modelscope/server.txt && \
     pip install --no-cache-dir https://modelscope.oss-cn-beijing.aliyuncs.com/packages/imageio_ffmpeg-0.4.9-py3-none-any.whl --no-dependencies --force && \
-    pip install --no-cache-dir 'scipy<1.13.0' && \
+    pip install --no-cache-dir 'scipy' && \
     pip install --no-cache-dir funtextprocessing typeguard==2.13.3 scikit-learn && \
     pip install --no-cache-dir 'decord>=0.6.0' mpi4py paint_ldm ipykernel fasttext && \
     pip install --no-cache-dir 'blobfile>=1.0.5' && \
diff --git a/docker/Dockerfile.ubuntu b/docker/Dockerfile.ubuntu
index 7be6731c..d61d428e 100644
--- a/docker/Dockerfile.ubuntu
+++ b/docker/Dockerfile.ubuntu
@@ -25,13 +25,13 @@ RUN pip uninstall ms-swift modelscope -y && pip --no-cache-dir install pip==23.*
 if [ "$INSTALL_MS_DEPS" = "True" ]; then \
     pip --no-cache-dir install omegaconf==2.0.6 && \
     pip install 'editdistance==0.8.1' && \
-    pip install --no-cache-dir 'cython<=0.29.36' versioneer 'numpy<2.0' && \
+    pip install --no-cache-dir 'cython>=3.0' versioneer 'numpy>=2.0' && \
     pip install --no-cache-dir -r /var/modelscope/framework.txt && \
     pip install --no-cache-dir -r /var/modelscope/audio.txt -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \
     pip install --no-cache-dir -r /var/modelscope/tests.txt && \
     pip install --no-cache-dir -r /var/modelscope/server.txt && \
     pip install --no-cache-dir https://modelscope.oss-cn-beijing.aliyuncs.com/packages/imageio_ffmpeg-0.4.9-py3-none-any.whl --no-dependencies --force && \
-    pip install --no-cache-dir 'scipy<1.13.0' && \
+    pip install --no-cache-dir 'scipy' && \
     pip install --no-cache-dir funtextprocessing typeguard==2.13.3 scikit-learn -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \
     pip install --no-cache-dir 'decord>=0.6.0' mpi4py paint_ldm ipykernel fasttext -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html && \
     pip install --no-cache-dir 'blobfile>=1.0.5' && \
diff --git a/modelscope/models/cv/video_depth_estimation/models/model_checkpoint.py b/modelscope/models/cv/video_depth_estimation/models/model_checkpoint.py
index 761990ff..b577ca60 100644
--- a/modelscope/models/cv/video_depth_estimation/models/model_checkpoint.py
+++ b/modelscope/models/cv/video_depth_estimation/models/model_checkpoint.py
@@ -40,7 +40,7 @@ class ModelCheckpoint:
         self.kth_best_model = ''
         self.best = 0
         # Monitoring modes
-        torch_inf = torch.tensor(np.Inf)
+        torch_inf = torch.tensor(np.inf)
         mode_dict = {
             'min': (torch_inf, 'min'),
             'max': (-torch_inf, 'max'),
diff --git a/modelscope/pipelines/multi_modal/soonet_video_temporal_grounding_pipeline.py b/modelscope/pipelines/multi_modal/soonet_video_temporal_grounding_pipeline.py
index 0251e745..ff71c000 100644
--- a/modelscope/pipelines/multi_modal/soonet_video_temporal_grounding_pipeline.py
+++ b/modelscope/pipelines/multi_modal/soonet_video_temporal_grounding_pipeline.py
@@ -1,5 +1,6 @@
 # Copyright 2022-2023 The Alibaba Fundamental Vision  Team Authors. All rights reserved.
 
+import math
 import os
 from typing import Any, Dict
 
@@ -144,12 +145,12 @@ class SOONetVideoTemporalGroundingPipeline(Pipeline):
         start_ts, end_ts, scale_boundaries = list(), list(), [0]
         ori_video_length = len(imgs)
         pad_video_length = int(
-            np.math.ceil(ori_video_length / self.max_anchor_length)
+            math.ceil(ori_video_length / self.max_anchor_length)
             * self.max_anchor_length)
         for i in range(self.config.nscales):
             anchor_length = self.config.snippet_length * (2**i)
             pad_feat_length = pad_video_length // anchor_length
-            nfeats = np.math.ceil(ori_video_length / anchor_length)
+            nfeats = math.ceil(ori_video_length / anchor_length)
             s_times = np.arange(0, nfeats).astype(np.float32) * (
                 anchor_length // self.fps)
             e_times = np.arange(1, nfeats + 1).astype(np.float32) * (
@@ -180,7 +181,7 @@ class SOONetVideoTemporalGroundingPipeline(Pipeline):
             #
             ori_video_length, feat_dim = video_feats.shape
             pad_video_length = int(
-                np.math.ceil(ori_video_length / self.max_anchor_length)
+                math.ceil(ori_video_length / self.max_anchor_length)
                 * self.max_anchor_length)
             pad_video_feats = torch.zeros((pad_video_length, feat_dim),
                                           dtype=float)