[Bugfix][Multimodal] Release Qwen2.5-VL and Qwen3-VL RoPE caches with the model (#54346)

Signed-off-by: waizuichougou <2082431897@qq.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>

vllm/model_executor/models/qwen2_5_vl.py

@@ -27,7 +27,7 @@
 """Inference-only Qwen2.5-VL model compatible with HuggingFace weights."""
 
 from collections.abc import Callable, Iterable, Iterator, Mapping, Sequence
-from functools import lru_cache, partial
+from functools import partial
 from typing import Annotated, Any, Literal, TypeAlias
 
 import einops
@@ -82,6 +82,7 @@ from vllm.multimodal.video_prune.evs import (
 )
 from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
+from vllm.utils.cache import LRUCache
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
 from vllm.utils.torch_utils import PIN_MEMORY, async_tensor_h2d
 from vllm.v1.attention.backends.registry import AttentionBackendEnum
@@ -647,6 +648,7 @@ class Qwen2_5_VisionTransformer(nn.Module):
         self.spatial_merge_size = vision_config.spatial_merge_size
         self.fullatt_block_indexes = vision_config.fullatt_block_indexes
         self.spatial_merge_unit = self.spatial_merge_size**2
+        self._rope_by_thw_cache = LRUCache(capacity=1024)
         use_data_parallel = is_vit_use_data_parallel()
         self.tp_size = (
             1
@@ -795,8 +797,11 @@ class Qwen2_5_VisionTransformer(nn.Module):
 
         return index_new, cu_seqlens_tmp
 
-    @lru_cache(maxsize=1024)  # noqa: B019
     def get_rope_by_thw(self, t, h, w):
+        cache_key = (t, h, w)
+        if cache_key in self._rope_by_thw_cache:
+            return self._rope_by_thw_cache[cache_key]
+
         window_index_thw, cu_seqlens_window_thw = self.get_window_index_thw(t, h, w)
         cos_thw, sin_thw = self.rotary_pos_emb_thw(t, h, w)
 
@@ -809,13 +814,15 @@ class Qwen2_5_VisionTransformer(nn.Module):
         cu_seqlens_thw = torch.repeat_interleave(
             torch.tensor([h * w], dtype=torch.int32), t
         )
-        return (
+        result = (
             cos_thw,
             sin_thw,
             window_index_thw,
             cu_seqlens_window_thw,
             cu_seqlens_thw,
         )
+        self._rope_by_thw_cache[cache_key] = result
+        return result
 
     def compute_attn_mask_seqlen(
         self,

vllm/model_executor/models/qwen3_vl.py

@@ -110,6 +110,7 @@ from vllm.sequence import IntermediateTensors
 from vllm.tokenizers.protocol import TokenizerLike
 from vllm.tokenizers.registry import cached_tokenizer_from_config
 from vllm.triton_utils import HAS_TRITON, tl, triton
+from vllm.utils.cache import LRUCache
 from vllm.utils.collection_utils import is_list_of
 from vllm.utils.math_utils import round_up
 from vllm.utils.torch_utils import PIN_MEMORY
@@ -581,6 +582,7 @@ class Qwen3_VisionTransformer(nn.Module):
             else []
         )
         self.num_grid_per_side = int(self.num_position_embeddings**0.5)
+        self._rot_pos_ids_cache = LRUCache(capacity=1024)
 
         use_data_parallel = is_vit_use_data_parallel()
         self.tp_size = (
@@ -672,9 +674,11 @@ class Qwen3_VisionTransformer(nn.Module):
     def device(self) -> torch.device:
         return self.patch_embed.proj.weight.device
 
-    @staticmethod
-    @lru_cache(maxsize=1024)
-    def rot_pos_ids(h: int, w: int, spatial_merge_size: int) -> torch.Tensor:
+    def rot_pos_ids(self, h: int, w: int, spatial_merge_size: int) -> torch.Tensor:
+        cache_key = (h, w, spatial_merge_size)
+        if cache_key in self._rot_pos_ids_cache:
+            return self._rot_pos_ids_cache[cache_key]
+
         hpos_ids = np.broadcast_to(np.arange(h).reshape(h, 1), (h, w))
         h_div = h // spatial_merge_size
         w_div = w // spatial_merge_size
@@ -697,7 +701,9 @@ class Qwen3_VisionTransformer(nn.Module):
         wpos_ids = wpos_ids.transpose(0, 2, 1, 3)
         wpos_ids = wpos_ids.flatten()
 
-        return torch.from_numpy(np.stack([hpos_ids, wpos_ids], axis=-1))
+        result = torch.from_numpy(np.stack([hpos_ids, wpos_ids], axis=-1))
+        self._rot_pos_ids_cache[cache_key] = result
+        return result
 
     def rot_pos_emb(self, grid_thw: list[list[int]]):
         max_grid_size = max(max(h, w) for _, h, w in grid_thw)