Fix compilation

examples/rtc/eval_dataset.py

@@ -19,6 +19,7 @@ Usage:
         --device=mps
 
     # With torch.compile for faster inference (PyTorch 2.0+)
+    # Note: CUDA graphs disabled by default due to in-place ops in denoising loop
     uv run python examples/rtc/eval_dataset.py \
         --policy.path=helper2424/smolvla_check_rtc_last3 \
         --dataset.repo_id=helper2424/check_rtc \
@@ -27,7 +28,7 @@ Usage:
         --use_torch_compile=true \
         --torch_compile_mode=max-autotune
 
-    # With torch.compile on CUDA
+    # With torch.compile on CUDA (CUDA graphs disabled by default)
     uv run python examples/rtc/eval_dataset.py \
         --policy.path=helper2424/smolvla_check_rtc_last3 \
         --dataset.repo_id=helper2424/check_rtc \
@@ -36,13 +37,14 @@ Usage:
         --use_torch_compile=true \
         --torch_compile_mode=reduce-overhead
 
-    # With custom compile settings
+    # Enable CUDA graphs (advanced - may cause tensor aliasing errors)
     uv run python examples/rtc/eval_dataset.py \
         --policy.path=helper2424/smolvla_check_rtc_last3 \
         --dataset.repo_id=helper2424/check_rtc \
         --use_torch_compile=true \
         --torch_compile_backend=inductor \
-        --torch_compile_mode=max-autotune
+        --torch_compile_mode=max-autotune \
+        --torch_compile_disable_cudagraphs=false
 """
 
 import gc
@@ -142,6 +144,14 @@ class RTCEvalConfig(HubMixin):
         metadata={"help": "Compilation mode (default, reduce-overhead, max-autotune)"},
     )
 
+    torch_compile_disable_cudagraphs: bool = field(
+        default=True,
+        metadata={
+            "help": "Disable CUDA graphs in torch.compile. Required due to in-place tensor "
+            "operations in denoising loop (x_t += dt * v_t) which cause tensor aliasing issues."
+        },
+    )
+
     def __post_init__(self):
         # Parse policy path
         policy_path = parser.get_path_arg("policy")
@@ -265,17 +275,23 @@ class RTCEvaluator:
             logging.info(f"  [{policy_name}] Applying torch.compile to predict_action_chunk...")
             logging.info(f"    Backend: {self.cfg.torch_compile_backend}")
             logging.info(f"    Mode: {self.cfg.torch_compile_mode}")
+            logging.info(f"    Disable CUDA graphs: {self.cfg.torch_compile_disable_cudagraphs}")
             logging.info("    Note: Debug tracker excluded from compilation via @torch._dynamo.disable")
 
             # Compile the predict_action_chunk method
-            # The debug tracker is excluded from compilation via @torch._dynamo.disable decorator
+            # - Debug tracker is excluded from compilation via @torch._dynamo.disable
-            # on the Tracker.track() method, so it won't cause graph breaks
+            # - CUDA graphs disabled to prevent tensor aliasing from in-place ops (x_t += dt * v_t)
+            compile_kwargs = {
+                "backend": self.cfg.torch_compile_backend,
+                "mode": self.cfg.torch_compile_mode,
+            }
+
+            # Disable CUDA graphs if requested (prevents tensor aliasing issues)
+            if self.cfg.torch_compile_disable_cudagraphs:
+                compile_kwargs["options"] = {"triton.cudagraphs": False}
+
             original_method = policy.predict_action_chunk
-            compiled_method = torch.compile(
+            compiled_method = torch.compile(original_method, **compile_kwargs)
-                original_method,
-                backend=self.cfg.torch_compile_backend,
-                mode=self.cfg.torch_compile_mode,
-            )
             policy.predict_action_chunk = compiled_method
             logging.info(f"  ✓ [{policy_name}] Successfully compiled predict_action_chunk")
 

src/lerobot/policies/rtc/modeling_rtc.py

@@ -270,7 +270,7 @@ class RTCProcessor:
             execution_horizon=execution_horizon,
         )
 
-        return result, x_t
+        return result
 
     def get_prefix_weights(self, start, end, total):
         start = min(start, end)

src/lerobot/policies/smolvla/modeling_smolvla.py

@@ -806,7 +806,7 @@ class VLAFlowMatching(nn.Module):
                 prev_chunk_left_over = kwargs.get("prev_chunk_left_over")
                 execution_horizon = kwargs.get("execution_horizon")
 
-                v_t, x_t = self.rtc_processor.denoise_step(
+                v_t = self.rtc_processor.denoise_step(
                     x_t=x_t,
                     prev_chunk_left_over=prev_chunk_left_over,
                     inference_delay=inference_delay,