diff --git a/dynolog_npu/README.md b/dynolog_npu/README.md
index 86a23b7f82925079c26623b070936538768d9b8c..72d1664752cb0526e67a6a2f55af334c06d6efff 100644
--- a/dynolog_npu/README.md
+++ b/dynolog_npu/README.md
@@ -97,6 +97,7 @@ nputrace子命令支持的参数选项
 | profile-start-time | u64 | 用于同步采集的Unix时间戳，单位毫秒，默认值0，dynolog原生参数 |
 | duration-ms | u64 | 采集的周期，单位毫秒，默认值500，dynolog原生参数 |
 | iterations | i64 | 采集总迭代数，默认值-1，dynolog原生参数 |
+| warmup | u64 | 采集前预热的迭代数，默认值0，默认不预热 |
 | log-file | String | 采集落盘的路径，必选值 |
 | start-step | u64 | 开始采集的迭代数，默认值0 |
 | record-shapes | action | 是否采集算子的InputShapes和InputTypes，设置参数采集，默认不采集 |
@@ -154,6 +155,9 @@ dyno nputrace --start-step 10 --iterations 2 --activities NPU --analyse --data-s
 
 # 示例3：从第10个step开始采集，采集2个step，只采集CANN和device数据，只采集不解析，落盘路径为/tmp/profile_data
 dyno nputrace --start-step 10 --iterations 2 --activities NPU --log-file /tmp/profile_data
+
+# 示例4：从第10个step开始预热1个step，采集2个step，只采集CANN和device数据，只采集不解析，落盘路径为/tmp/profile_data
+dyno nputrace --start-step 10 --iterations 2 --warmup 1 --activities NPU --log-file /tmp/profile_data
 ```
 
 ### NPU Monitor功能
diff --git a/dynolog_npu/dynolog_npu/cli/src/commands/nputrace.rs b/dynolog_npu/dynolog_npu/cli/src/commands/nputrace.rs
index f70923bca4cc5ce29a8855a464c411b63a930ef0..cfc53a624d962664d1c2f3732b4873914ecf0953 100644
--- a/dynolog_npu/dynolog_npu/cli/src/commands/nputrace.rs
+++ b/dynolog_npu/dynolog_npu/cli/src/commands/nputrace.rs
@@ -15,6 +15,7 @@ pub enum NpuTraceTriggerConfig {
     IterationBased {
         start_step: u64,
         iterations: i64,
+        warmup: u64,
     },
 }
 
@@ -31,11 +32,13 @@ impl NpuTraceTriggerConfig {
             NpuTraceTriggerConfig::IterationBased {
                 start_step,
                 iterations,
+                warmup,
             } => format!(
                 r#"PROFILE_START_ITERATION=0
 PROFILE_START_STEP={}
-ACTIVITIES_ITERATIONS={}"#,
-                start_step, iterations
+ACTIVITIES_ITERATIONS={}
+WARMUP_ITERATIONS={}"#,
+                start_step, iterations, warmup
             ),
         }
     }
@@ -185,14 +188,16 @@ ACTIVITIES_DURATION_MSECS=1000"#
         );
 
         let trigger_config = NpuTraceTriggerConfig::IterationBased {    
-            profile_start_step: 1000,
+            start_step: 1000,
             iterations: 1000,
+            warmup: 1000,
         };
         assert_eq!(
             trigger_config.config(),
             r#"PROFILE_START_ITERATION=0
 PROFILE_START_STEP=1000
-ACTIVITIES_ITERATIONS=1000"#
+ACTIVITIES_ITERATIONS=1000
+WARMUP_ITERATIONS=1000"#
         );
     }
 
diff --git a/dynolog_npu/dynolog_npu/cli/src/main.rs b/dynolog_npu/dynolog_npu/cli/src/main.rs
index 9fdea3d1254467081356b2e0daeb8ed3ca05a16d..d1a485ded1ba9169d07547f337ee14737bd71157 100644
--- a/dynolog_npu/dynolog_npu/cli/src/main.rs
+++ b/dynolog_npu/dynolog_npu/cli/src/main.rs
@@ -136,6 +136,9 @@ enum Command {
         /// Number of steps to start profile.
         #[clap(long, default_value_t = 0)]
         start_step: u64,
+        /// Number of steps to warmup.
+        #[clap(long, default_value_t = 0)]
+        warmup: u64,
         /// Max number of processes to profile.
         #[clap(long, default_value_t = 3)]
         process_limit: u32,
@@ -281,6 +284,7 @@ fn main() -> Result<()> {
             iterations,
             profile_start_time,
             start_step,
+            warmup,
             process_limit,
             record_shapes,
             profile_memory,
@@ -302,6 +306,7 @@ fn main() -> Result<()> {
                 NpuTraceTriggerConfig::IterationBased {
                     start_step,
                     iterations,
+                    warmup,
                 }
             } else {
                 NpuTraceTriggerConfig::DurationBased {