diff --git a/crates/ml-alpha/src/mamba2_block.rs b/crates/ml-alpha/src/mamba2_block.rs
index 79275b8cd..bde8edfba 100644
--- a/crates/ml-alpha/src/mamba2_block.rs
+++ b/crates/ml-alpha/src/mamba2_block.rs
@@ -547,6 +547,12 @@ impl Mamba2Block {
             )
             .result()
             .map_err(|e| anyhow!("Mamba2Block: cublasSetWorkspace_v2: {e:?}"))?;
+            cudarc::cublas::sys::cublasSetMathMode(
+                *cublas.handle(),
+                cudarc::cublas::sys::cublasMath_t::CUBLAS_TF32_TENSOR_OP_MATH,
+            )
+            .result()
+            .map_err(|e| anyhow!("Mamba2Block: cublasSetMathMode TF32: {e:?}"))?;
         }
 
         // ── Parameter allocation + initialisation. ──────────────────────
diff --git a/crates/ml-alpha/src/rl/dqn.rs b/crates/ml-alpha/src/rl/dqn.rs
index 0d7d40500..c725b6a69 100644
--- a/crates/ml-alpha/src/rl/dqn.rs
+++ b/crates/ml-alpha/src/rl/dqn.rs
@@ -311,6 +311,12 @@ impl DqnHead {
             )
             .result()
             .map_err(|e| anyhow::anyhow!("DqnHead: cublasSetWorkspace_v2: {e:?}"))?;
+            cudarc::cublas::sys::cublasSetMathMode(
+                *cublas.handle(),
+                cudarc::cublas::sys::cublasMath_t::CUBLAS_TF32_TENSOR_OP_MATH,
+            )
+            .result()
+            .map_err(|e| anyhow::anyhow!("DqnHead: cublasSetMathMode TF32: {e:?}"))?;
         }
 
         // Bias-add and reduce-sum-axis0 kernel handles from ml-core.