Update to new version of base image

vllm/model_executor/models/internvl.py

@@ -4,7 +4,7 @@
 # Copyright (c) 2023 OpenGVLab
 # Licensed under The MIT License [see LICENSE for details]
 # --------------------------------------------------------
-import re
+import regex as re
 from functools import cached_property, partial
 from typing import (Iterable, List, Literal, Mapping, Optional, Tuple,
                     TypedDict, Union)

vllm/model_executor/models/minicpmv.py

@@ -22,7 +22,7 @@
 # limitations under the License.
 """Inference-only MiniCPM-V model compatible with HuggingFace weights."""
 import math
-import re
+import regex as re
 from functools import partial
 from typing import (Any, Callable, Iterable, List, Literal, Mapping, Optional,
                     Tuple, TypedDict, Union)

vllm/model_executor/models/molmo.py

@@ -1,6 +1,6 @@
 import logging
 import math
-import re
+import regex as re
 from array import array
 from dataclasses import dataclass
 from functools import lru_cache, partial

vllm/model_executor/models/phi3v.py

@@ -14,7 +14,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import itertools
-import re
+import regex as re
 from functools import cached_property, lru_cache
 from typing import (Any, Dict, Iterable, List, Literal, Mapping, Optional,
                     Tuple, TypedDict, Union)

vllm/model_executor/models/qwen.py

@@ -6,7 +6,7 @@
 """Inference-only QWen model compatible with HuggingFace weights."""
 
 import math
-import re
+import regex as re
 from functools import partial
 from typing import (Any, Callable, Dict, Iterable, List, Literal, Mapping,
                     Optional, Tuple, TypedDict, Union)

vllm/model_executor/models/qwen3.py

@@ -136,11 +136,11 @@ class Qwen3Attention(nn.Module):
         # Add qk-norm
         q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim,
                            self.head_dim)
-        q_by_head = self.q_norm.forward_native(q_by_head)
+        q_by_head = self.q_norm.forward_cuda(q_by_head.contiguous())
         q = q_by_head.view(q.shape)
         k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim,
                            self.head_dim)
-        k_by_head = self.k_norm.forward_native(k_by_head)
+        k_by_head = self.k_norm.forward_cuda(k_by_head.contiguous())
         k = k_by_head.view(k.shape)
         q, k = self.rotary_emb(positions, q, k)
         attn_output = self.attn(q, k, v, kv_cache, attn_metadata)

vllm/model_executor/models/registry.py

@@ -114,6 +114,7 @@ _MULTIMODAL_MODELS = {
     "PixtralForConditionalGeneration": ("pixtral", "PixtralForConditionalGeneration"),  # noqa: E501
     "QWenLMHeadModel": ("qwen", "QWenLMHeadModel"),
     "Qwen2VLForConditionalGeneration": ("qwen2_vl", "Qwen2VLForConditionalGeneration"),  # noqa: E501
+    "Qwen2_5_VLForConditionalGeneration": ("qwen2_5_vl", "Qwen2_5_VLForConditionalGeneration"),  # noqa: E501
     "UltravoxModel": ("ultravox", "UltravoxModel"),
     # [Encoder-decoder]
     "MllamaForConditionalGeneration": ("mllama", "MllamaForConditionalGeneration"),  # noqa: E501