Allow configuring pytorch compilation mode (#905)

This PR allows users to configure the compilation mode used by
`Pi0Pytorch` to compile the `sample_actions` method.

This was specifically motivated by needing to use the
`max-autotune-no-cudagraphs` mode, due to threading issues. By default
the compilation mode is `max-autotune`, which preserves existing
behavior.

Author: kvablack

src/openpi/models/pi0_config.py

@@ -32,11 +32,20 @@ class Pi0Config(_model.BaseModelConfig):
     # This config option is not used directly by the model, but it is read by the ModelTransformFactory.
     discrete_state_input: bool = None  # type: ignore
 
+    pytorch_compile_mode: str | None = "max-autotune"
+
     def __post_init__(self):
         if self.max_token_len is None:
             object.__setattr__(self, "max_token_len", 200 if self.pi05 else 48)
         if self.discrete_state_input is None:
             object.__setattr__(self, "discrete_state_input", self.pi05)
+        if self.pytorch_compile_mode is not None:
+            assert self.pytorch_compile_mode in [
+                "default",
+                "reduce-overhead",
+                "max-autotune",
+                "max-autotune-no-cudagraphs",
+            ]
 
     @property
     @override

src/openpi/models_pytorch/pi0_pytorch.py

@@ -109,7 +109,8 @@ def __init__(self, config):
             self.action_time_mlp_out = nn.Linear(action_expert_config.width, action_expert_config.width)
 
         torch.set_float32_matmul_precision("high")
-        self.sample_actions = torch.compile(self.sample_actions, mode="max-autotune")
+        if config.pytorch_compile_mode is not None:
+            self.sample_actions = torch.compile(self.sample_actions, mode=config.pytorch_compile_mode)
 
         # Initialize gradient checkpointing flag
         self.gradient_checkpointing_enabled = False