diff --git a/docs/get_started/model_optimization.md b/docs/get_started/model_optimization.md
index f419eeb0f0baf256af077811e85f6923294e599f..d63ddb884d036642cf680607c657ce9bbc1f6628 100644
--- a/docs/get_started/model_optimization.md
+++ b/docs/get_started/model_optimization.md
@@ -48,6 +48,37 @@ tcmalloc（即Thread-Caching Malloc）是一个通用的内存分配器，通过
 | bev_pool_v3 | from mx_driving import bev_pool_v3 |
 | SparseConv3d | from mx_driving import SparseConv3d |
 | SubMConv3d | from mx_driving import SubMConv3d |
+#### 【算子替换示例】
+
+当采集完profiling后，查看算子耗时统计，分析耗时占比大的算子，进行替换，以BEVFormer模型为例，multi_scale_deformable_attn算子正反向在模型中占比很高，需要替换成mxDriving中的亲和算子：
+<img src="op_statistic.png" alt="算子耗时占比" width="800" align="center">
+
+
+替换过程如下，算子详细参数参考[算子清单](../api/README.md)，算子使用位置为projects/mmdet3d_plugin/bevformer/modules/decoder.py：
+- 头文件引入mx_driving
+```python
+import mx_driving
+```
+- 找到算子使用的位置，位置在325行，原始代码为：
+```python
+    if torch.cuda.is_available() and value.is_cuda:
+
+        # using fp16 deformable attention is unstable because it performs many sum operations
+        if value.dtype == torch.float16:
+            MultiScaleDeformableAttnFunction = MultiScaleDeformableAttnFunction_fp32
+        else:
+            MultiScaleDeformableAttnFunction = MultiScaleDeformableAttnFunction_fp32
+        output = MultiScaleDeformableAttnFunction.apply(
+            value, spatial_shapes, level_start_index, sampling_locations,
+            attention_weights, self.im2col_step)
+    else:
+        output = multi_scale_deformable_attn_pytorch(
+            value, spatial_shapes, sampling_locations, attention_weights)
+```
+替换为mx_driving仓中算子：
+```python
+    output = mx_driving.multi_scale_deformable_attn(value, spatial_shapes, level_start_index, sampling_locations, attention_weights)
+```
 
 ### 2.3 自驾模型host bound问题优化
 自动驾驶算法有很多slice、gather、sort等小算子，2D图像特征转换到BEV空间，涉及大量投影、插值、采样等操作，点云数据的体素化处理、稀疏卷积等。自动驾驶算法中涉及很多逻辑处理，如检测算法的target assign，规控类算法lovaz loss，以及一些对groundtruth处理的操作。Host bound问题：小算子下发多，CPU计算处理逻辑多、负载大，算子没有NPU高性能替换等造成host bound问题严重。
diff --git a/docs/get_started/op_statistic.png b/docs/get_started/op_statistic.png
new file mode 100644
index 0000000000000000000000000000000000000000..6ea78b1cc9a7167be9bf3a936500eb2ca34c494b
Binary files /dev/null and b/docs/get_started/op_statistic.png differ