Add native SmolVLA model integration

tests/test_eval_vla_execution.py

@@ -129,6 +129,23 @@ class EvalVLAExecutionTest(unittest.TestCase):
             ["r_vis", "top", "front"],
         )
 
+    def test_resolve_eval_image_resize_shape_prefers_agent_top_level_override(self):
+        cfg = OmegaConf.create(
+            {
+                "agent": {
+                    "eval_image_resize_shape": None,
+                    "condition_encoder": {
+                        "eval_image_resize_shape": [256, 256],
+                    },
+                },
+                "data": {
+                    "image_resize_shape": [224, 224],
+                },
+            }
+        )
+
+        self.assertIsNone(eval_vla._resolve_eval_image_resize_shape(cfg))
+
     def test_resolve_eval_image_resize_shape_prefers_condition_encoder_override(self):
         cfg = OmegaConf.create(
             {
@@ -270,6 +287,25 @@ class EvalVLAExecutionTest(unittest.TestCase):
         np.testing.assert_array_equal(queues["action"].popleft().numpy(), np.array([20.0], dtype=np.float32))
         np.testing.assert_array_equal(queues["action"].popleft().numpy(), np.array([30.0], dtype=np.float32))
 
+    def test_enqueue_predicted_actions_honors_explicit_chunk_start(self):
+        queues = eval_vla._new_local_policy_queues(obs_horizon=2)
+        predicted_actions = torch.tensor(
+            [[[10.0], [20.0], [30.0], [40.0]]],
+            dtype=torch.float32,
+        )
+
+        eval_vla._enqueue_predicted_actions(
+            queues,
+            predicted_actions=predicted_actions,
+            obs_horizon=2,
+            num_action_steps=2,
+            action_chunk_start=0,
+        )
+
+        self.assertEqual(len(queues["action"]), 2)
+        np.testing.assert_array_equal(queues["action"].popleft().numpy(), np.array([10.0], dtype=np.float32))
+        np.testing.assert_array_equal(queues["action"].popleft().numpy(), np.array([20.0], dtype=np.float32))
+
     def test_remote_policy_runner_only_requests_server_inference_when_local_action_queue_is_empty(self):
         request_queue = _FakeQueue()
         response_queue = _FakeQueue(
@@ -288,6 +324,7 @@ class EvalVLAExecutionTest(unittest.TestCase):
             camera_names=["front"],
             obs_horizon=2,
             num_action_steps=2,
+            action_chunk_start=0,
         )
         first_observation = {
             "qpos": torch.tensor([1.0, 2.0], dtype=torch.float32),
@@ -315,8 +352,46 @@ class EvalVLAExecutionTest(unittest.TestCase):
         self.assertEqual(request_queue.put_calls[0]["type"], "predict_chunk")
         self.assertEqual(request_queue.put_calls[0]["worker_index"], 3)
         self.assertEqual(request_queue.put_calls[0]["server_index"], 1)
-        np.testing.assert_array_equal(first_action.numpy(), np.array([20.0], dtype=np.float32))
-        np.testing.assert_array_equal(second_action.numpy(), np.array([30.0], dtype=np.float32))
+        np.testing.assert_array_equal(first_action.numpy(), np.array([10.0], dtype=np.float32))
+        np.testing.assert_array_equal(second_action.numpy(), np.array([20.0], dtype=np.float32))
+
+    def test_remote_eval_worker_passes_agent_action_chunk_start_to_remote_runner(self):
+        cfg = OmegaConf.create(
+            {
+                "agent": {
+                    "obs_horizon": 2,
+                    "num_action_steps": 2,
+                    "action_chunk_start": 0,
+                    "camera_names": ["front"],
+                },
+                "eval": {
+                    "obs_horizon": 2,
+                    "num_queries": 2,
+                    "response_timeout_s": 3.0,
+                    "camera_names": ["front"],
+                },
+            }
+        )
+        captured = {}
+
+        class CapturingRunner:
+            def __init__(self, **kwargs):
+                captured.update(kwargs)
+
+        with mock.patch.object(eval_vla, "_RemotePolicyRunner", CapturingRunner), \
+             mock.patch.object(eval_vla, "_run_eval_episode_plans", return_value={"ok": True}) as run_plans:
+            result = eval_vla._run_remote_eval_worker(
+                cfg,
+                episode_plans=[{"episode_index": 0}],
+                worker_index=1,
+                server_index=2,
+                request_queue=_FakeQueue(),
+                response_queue=_FakeQueue(),
+            )
+
+        self.assertEqual(result, {"ok": True})
+        self.assertEqual(captured["action_chunk_start"], 0)
+        run_plans.assert_called_once()
 
     def test_merge_worker_summaries_sorts_episodes_and_recomputes_aggregates(self):
         worker_summaries = [