diff --git a/crates/ml/examples/hyperopt_baseline_rl.rs b/crates/ml/examples/hyperopt_baseline_rl.rs
index a337ff8c5..b9ff75327 100644
--- a/crates/ml/examples/hyperopt_baseline_rl.rs
+++ b/crates/ml/examples/hyperopt_baseline_rl.rs
@@ -104,8 +104,8 @@ struct Args {
     #[arg(long, default_value = "1.0")]
     spread_ticks: f64,
 
-    /// Optimizer to use: "pso" (Particle Swarm) or "tpe" (Tree-Parzen Estimator)
-    #[arg(long, default_value = "pso")]
+    /// Optimizer to use: "tpe" (Tree-Parzen Estimator) or "pso" (Particle Swarm)
+    #[arg(long, default_value = "tpe")]
     optimizer: String,
 
     /// Number of parallel trial evaluations (PSO only; TPE is always sequential).
diff --git a/crates/ml/src/hyperopt/adapters/dqn.rs b/crates/ml/src/hyperopt/adapters/dqn.rs
index 7d938fb68..405cf1e94 100644
--- a/crates/ml/src/hyperopt/adapters/dqn.rs
+++ b/crates/ml/src/hyperopt/adapters/dqn.rs
@@ -2950,7 +2950,7 @@ impl HyperparameterOptimizable for DQNTrainer {
             if let Ok(sync_tensor) =
                 candle_core::Tensor::zeros(1, candle_core::DType::F32, &cleanup_device)
             {
-                let _ = sync_tensor.to_vec0::<f32>();
+                drop(sync_tensor.to_vec0::<f32>());
             }
             // Brief pause to allow CUDA memory allocator to reclaim freed blocks
             std::thread::sleep(std::time::Duration::from_millis(50));