diff --git a/tests/samplers/test_sampler.py b/tests/samplers/test_sampler.py
index 719254a398c0..19a5ca5e2750 100644
--- a/tests/samplers/test_sampler.py
+++ b/tests/samplers/test_sampler.py
@@ -418,6 +418,7 @@ def run_test_case(*, expected_penalization: List[bool],
                 prompt_len = seq_data.get_prompt_len()
                 seq_lens.append(prompt_len)
 
+                assert sgm.sampling_params is not None
                 if sgm.sampling_params.prompt_logprobs:
                     # with prompt_logprobs each token in the prompt has a row in
                     # logits
@@ -533,6 +534,8 @@ def test_sampling():
 
         for i, (sequence_output, metadata) in enumerate(
                 zip(sampler_output, seq_group_metadata_list)):
+            assert metadata.sampling_params is not None
+
             if metadata.sampling_params.use_beam_search:
                 continue
 
@@ -550,6 +553,8 @@ def test_sampling():
             assert expected_tokens_item is not None
 
             for n, nth_output in enumerate(sequence_output.samples):
+                assert metadata.sampling_params is not None
+
                 if (metadata.sampling_params.temperature == 0
                         or metadata.sampling_params.seed is not None):
                     # Ensure exact matches for greedy or random with seed
diff --git a/vllm/assets/audio.py b/vllm/assets/audio.py
index b00a61ebfec6..49bb6aeee90b 100644
--- a/vllm/assets/audio.py
+++ b/vllm/assets/audio.py
@@ -19,7 +19,9 @@ def audio_and_sample_rate(self) -> Tuple[np.ndarray, int]:
 
         audio_path = get_vllm_public_assets(filename=f"{self.name}.ogg",
                                             s3_prefix=ASSET_DIR)
-        return librosa.load(audio_path, sr=None)
+        y, sr = librosa.load(audio_path, sr=None)
+        assert isinstance(sr, int)
+        return y, sr
 
     @property
     def url(self) -> str:
diff --git a/vllm/entrypoints/openai/rpc/client.py b/vllm/entrypoints/openai/rpc/client.py
index dc316ca1160c..a472e12e8ca4 100644
--- a/vllm/entrypoints/openai/rpc/client.py
+++ b/vllm/entrypoints/openai/rpc/client.py
@@ -101,6 +101,7 @@ def __init__(self, rpc_path: str):
         # Maximum number of sockets that can be opened (typically 65536).
         # ZMQ_SOCKET_LIMIT (http://api.zeromq.org/4-2:zmq-ctx-get)
         socket_limit = self.context.get(zmq.constants.SOCKET_LIMIT)
+        assert isinstance(socket_limit, int)
         if socket_limit < VLLM_RPC_SOCKET_LIMIT_CUTOFF:
             raise ValueError(
                 f"Found zmq.constants.SOCKET_LIMIT={socket_limit}, which caps "
@@ -141,8 +142,8 @@ async def run_proxy(self, socket_from, socket_to):
         poller.register(socket_from, zmq.constants.POLLIN)
         poller.register(socket_to, zmq.constants.POLLIN)
         while True:
-            events = await poller.poll()
-            events = dict(events)
+            events_lst = await poller.poll()
+            events = dict(events_lst)
             if socket_from in events:
                 identity, msg = await socket_from.recv_multipart()
                 await socket_to.send_multipart([identity, msg])
diff --git a/vllm/multimodal/base.py b/vllm/multimodal/base.py
index 5b00117c64e5..f26e3292c264 100644
--- a/vllm/multimodal/base.py
+++ b/vllm/multimodal/base.py
@@ -14,7 +14,7 @@
 from vllm.config import ModelConfig
 from vllm.inputs import InputContext
 from vllm.logger import init_logger
-from vllm.utils import json_map_leaves
+from vllm.utils import JSONTree, is_list_of, json_map_leaves
 
 logger = init_logger(__name__)
 
@@ -54,13 +54,14 @@ def _try_stack(nested_tensors: NestedTensors) -> NestedTensors:
             return nested_tensors
 
         stacked = [MultiModalInputs._try_stack(t) for t in nested_tensors]
-        if any(isinstance(t, list) for t in stacked):
+        if is_list_of(stacked, list):
+            # Do not stack nested lists
             return stacked
 
         tensors_ = cast(List[torch.Tensor], stacked)
         if any(t.shape != tensors_[0].shape for t in tensors_):
             # The tensors have incompatible shapes and can't be stacked.
-            return tensors_
+            return stacked
 
         return torch.stack(tensors_)
 
@@ -101,8 +102,14 @@ def as_kwargs(
         *,
         device: torch.types.Device,
     ) -> BatchedTensorInputs:
-        return json_map_leaves(lambda x: x.to(device, non_blocking=True),
-                               batched_inputs)
+        json_inputs = cast(JSONTree[torch.Tensor], batched_inputs)
+
+        json_mapped = json_map_leaves(
+            lambda x: x.to(device, non_blocking=True),
+            json_inputs,
+        )
+
+        return cast(BatchedTensorInputs, json_mapped)
 
 
 _T = TypeVar("_T")
diff --git a/vllm/sequence.py b/vllm/sequence.py
index 964072dd7c8f..f289a9aec80c 100644
--- a/vllm/sequence.py
+++ b/vllm/sequence.py
@@ -883,7 +883,7 @@ class SequenceGroupMetadata(
     request_id: str
     is_prompt: bool
     seq_data: Dict[int, SequenceData]
-    sampling_params: SamplingParams
+    sampling_params: Optional[SamplingParams]
     block_tables: Dict[int, List[int]]
     do_sample: bool = True
     pooling_params: Optional[PoolingParams] = None