diff --git a/crates/ml-dqn/src/branching.rs b/crates/ml-dqn/src/branching.rs
index 370eb75d9..6bf0c977d 100644
--- a/crates/ml-dqn/src/branching.rs
+++ b/crates/ml-dqn/src/branching.rs
@@ -297,7 +297,9 @@ impl BranchingDuelingQNetwork {
         }
 
         let vars = VarMap::new();
-        let vb = VarBuilder::from_varmap(&vars, candle_core::DType::BF16, &device);
+        // F32 weights: the fused CUDA trainer's Adam kernel and gpu_weights.rs
+        // fast-path extraction require F32. BF16 mirrors are maintained by GpuDqnTrainer.
+        let vb = VarBuilder::from_varmap(&vars, candle_core::DType::F32, &device);
 
         // Shared encoder (always standard Linear -- noise only in heads)
         let mut shared_layers = Vec::new();
diff --git a/crates/ml-dqn/src/noisy_layers.rs b/crates/ml-dqn/src/noisy_layers.rs
index 2dc2e2099..c497953e0 100644
--- a/crates/ml-dqn/src/noisy_layers.rs
+++ b/crates/ml-dqn/src/noisy_layers.rs
@@ -64,8 +64,8 @@ impl NoisyLinear {
     ) -> Result<Self, MLError> {
         let device = vb.device().clone();
 
-        // All init in F32, then cast — Candle's Init::Uniform CUDA kernel lacks BF16 PTX.
-        let dtype = candle_core::DType::BF16;
+        // F32 weights: fused CUDA trainer operates on F32, BF16 mirrors managed separately.
+        let dtype = candle_core::DType::F32;
 
         // Initialize μ_w ~ U(-1/√in, 1/√in) (Rainbow DQN standard)
         let mu_range = 1.0 / (in_features as f64).sqrt();
@@ -384,7 +384,7 @@ mod tests {
     fn test_noisy_linear_creation() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let _layer = NoisyLinear::new(64, 32, vb, 0.5)?;
         Ok(())
@@ -394,7 +394,7 @@ mod tests {
     fn test_noisy_linear_forward() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let mut layer = NoisyLinear::new(64, 32, vb, 0.5)?;
         layer.reset_noise()?; // Resample noise before forward
@@ -402,7 +402,7 @@ mod tests {
         // Create dummy input
         let input = Tensor::randn(0.0_f32, 1.0_f32, (4, 64), &device)
             .map_err(|e| MLError::ModelError(format!("Failed to create input: {}", e)))?
-            .to_dtype(candle_core::DType::BF16)
+            .to_dtype(candle_core::DType::F32)
             .map_err(|e| MLError::ModelError(format!("Failed to cast input: {}", e)))?;
 
         // Forward pass
@@ -418,12 +418,12 @@ mod tests {
     fn test_noise_reset() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let mut layer = NoisyLinear::new(64, 32, vb, 0.5)?;
         let input = Tensor::randn(0.0_f32, 1.0_f32, (4, 64), &device)
             .map_err(|e| MLError::ModelError(format!("Failed to create input: {}", e)))?
-            .to_dtype(candle_core::DType::BF16)
+            .to_dtype(candle_core::DType::F32)
             .map_err(|e| MLError::ModelError(format!("Failed to cast input: {}", e)))?;
 
         // First forward pass
@@ -467,12 +467,12 @@ mod tests {
     fn test_disable_noise() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let mut layer = NoisyLinear::new(64, 32, vb, 0.5)?;
         let input = Tensor::randn(0.0_f32, 1.0_f32, (4, 64), &device)
             .map_err(|e| MLError::ModelError(format!("Failed to create input: {}", e)))?
-            .to_dtype(candle_core::DType::BF16)
+            .to_dtype(candle_core::DType::F32)
             .map_err(|e| MLError::ModelError(format!("Failed to cast input: {}", e)))?;
 
         // Reset noise for first pass
@@ -511,7 +511,7 @@ mod tests {
     fn test_factorized_noise_dimensions() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let mut layer = NoisyLinear::new(128, 64, vb, 0.5)?;
         layer.reset_noise()?;
@@ -527,12 +527,12 @@ mod tests {
     fn test_reset_noise_with_sigma() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let mut layer = NoisyLinear::new(64, 32, vb, 0.5)?;
         let input = Tensor::randn(0.0_f32, 1.0_f32, (4, 64), &device)
             .map_err(|e| MLError::ModelError(format!("Failed to create input: {}", e)))?
-            .to_dtype(candle_core::DType::BF16)
+            .to_dtype(candle_core::DType::F32)
             .map_err(|e| MLError::ModelError(format!("Failed to cast input: {}", e)))?;
 
         // Test with high sigma (0.6)
@@ -571,7 +571,7 @@ mod tests {
     fn test_sigma_scaling_effect() -> Result<(), MLError> {
         let device = Device::new_cuda(0).expect("CUDA required");
         let varmap = VarMap::new();
-        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::BF16, &device);
+        let vb = VarBuilder::from_varmap(&varmap, candle_core::DType::F32, &device);
 
         let mut layer = NoisyLinear::new(64, 32, vb, 0.5)?;
 
@@ -579,7 +579,7 @@ mod tests {
         layer.disable_noise()?;
         let input = Tensor::randn(0.0_f32, 1.0_f32, (4, 64), &device)
             .map_err(|e| MLError::ModelError(format!("Failed to create input: {}", e)))?
-            .to_dtype(candle_core::DType::BF16)
+            .to_dtype(candle_core::DType::F32)
             .map_err(|e| MLError::ModelError(format!("Failed to cast input: {}", e)))?;
         let output_no_noise = layer.forward(&input)?;