chore(pt): use more warmup steps (#4761)

The SiLUT activation function is using JIT compile. Hence, we require
more warm-up steps to avoid JIT compiling time interfering the profiling
result. This commit also adds logging info of the saved profiling
results.

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **Enhancements**
- Increased the profiling warmup duration during training for improved
profiling accuracy.
- Added a log message to notify users when the profiling trace is saved
to the TensorBoard log directory.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Author: caic99

deepmd/pt/train/training.py

@@ -684,7 +684,7 @@ def run(self) -> None:
             writer = SummaryWriter(log_dir=self.tensorboard_log_dir)
         if self.enable_profiler or self.profiling:
             prof = torch.profiler.profile(
-                schedule=torch.profiler.schedule(wait=1, warmup=1, active=3, repeat=1),
+                schedule=torch.profiler.schedule(wait=1, warmup=15, active=3, repeat=1),
                 on_trace_ready=torch.profiler.tensorboard_trace_handler(
                     self.tensorboard_log_dir
                 )
@@ -1058,10 +1058,14 @@ def log_loss_valid(_task_key="Default"):
             writer.close()
         if self.enable_profiler or self.profiling:
             prof.stop()
+            if self.enable_profiler:
+                log.info(
+                    f"The profiling trace has been saved under {self.tensorboard_log_dir}"
+                )
             if self.profiling:
                 prof.export_chrome_trace(self.profiling_file)
                 log.info(
-                    f"The profiling trace have been saved to: {self.profiling_file}"
+                    f"The profiling trace has been saved to: {self.profiling_file}"
                 )
 
     def save_model(self, save_path, lr=0.0, step=0) -> None: