[Bugfix] Fix missing tie_word_embeddings on Qwen3-VL text_config

tests/ut/test.py

@@ -1,9 +1,79 @@
+from types import SimpleNamespace
 from unittest.mock import MagicMock, Mock, patch
 
 import pytest
 import torch
 
 
+def _make_platform_config(hf_config):
+    from vllm.config import CUDAGraphMode
+
+    return SimpleNamespace(
+        parallel_config=SimpleNamespace(worker_cls="manual", data_parallel_size=1),
+        model_config=SimpleNamespace(
+            use_mla=False,
+            hf_config=hf_config,
+            enforce_eager=False,
+        ),
+        cache_config=SimpleNamespace(block_size=16),
+        speculative_config=None,
+        compilation_config=SimpleNamespace(
+            cudagraph_mode=CUDAGraphMode.NONE,
+            pass_config=SimpleNamespace(enable_fusion=True),
+            backend=None,
+            custom_ops=[],
+        ),
+    )
+
+
+def _check_platform_config(vllm_config):
+    import vllm.envs as envs
+
+    from vllm_kunlun.platforms.kunlun import KunlunPlatform
+
+    with patch.object(envs, "VLLM_ALL2ALL_BACKEND", None, create=True):
+        KunlunPlatform.check_and_update_config(vllm_config)
+
+
+def test_qwen3_vl_text_config_inherits_top_level_tie_word_embeddings():
+    text_config = SimpleNamespace()
+    hf_config = SimpleNamespace(
+        architectures=["Qwen3VLForConditionalGeneration"],
+        text_config=text_config,
+        tie_word_embeddings=False,
+    )
+
+    _check_platform_config(_make_platform_config(hf_config))
+
+    assert text_config.tie_word_embeddings is False
+
+
+def test_qwen3_vl_text_config_existing_tie_word_embeddings_is_preserved():
+    text_config = SimpleNamespace(tie_word_embeddings=True)
+    hf_config = SimpleNamespace(
+        architectures=["Qwen3VLForConditionalGeneration"],
+        text_config=text_config,
+        tie_word_embeddings=False,
+    )
+
+    _check_platform_config(_make_platform_config(hf_config))
+
+    assert text_config.tie_word_embeddings is True
+
+
+def test_non_qwen3_vl_text_config_is_not_modified():
+    text_config = SimpleNamespace()
+    hf_config = SimpleNamespace(
+        architectures=["Qwen3ForCausalLM"],
+        text_config=text_config,
+        tie_word_embeddings=False,
+    )
+
+    _check_platform_config(_make_platform_config(hf_config))
+
+    assert not hasattr(text_config, "tie_word_embeddings")
+
+
 def test_import():
     """Test that the module can be imported successfully."""
     from vllm_kunlun.compilation.wrapper import TorchCompileWrapperWithCustomDispatcher

vllm_kunlun/platforms/kunlun.py

@@ -18,6 +18,30 @@
 
 logger = init_logger(__name__)
 
+_QWEN3_VL_ARCHITECTURES = {"Qwen3VLForConditionalGeneration"}
+
+
+def _is_qwen3_vl_config(hf_config) -> bool:
+    config_type = type(hf_config).__name__
+    architectures = getattr(hf_config, "architectures", None) or ()
+    if isinstance(architectures, str):
+        architectures = (architectures,)
+
+    return config_type == "Qwen3VLConfig" or any(
+        architecture in _QWEN3_VL_ARCHITECTURES for architecture in architectures
+    )
+
+
+def _patch_qwen3_vl_text_config(hf_config) -> None:
+    if hf_config is None or not _is_qwen3_vl_config(hf_config):
+        return
+
+    text_config = getattr(hf_config, "text_config", None)
+    if text_config is None or hasattr(text_config, "tie_word_embeddings"):
+        return
+
+    text_config.tie_word_embeddings = getattr(hf_config, "tie_word_embeddings", False)
+
 
 class KunlunPlatform(Platform):
     """KunlunPlatform"""
@@ -179,6 +203,8 @@ def check_and_update_config(cls, vllm_config: "VllmConfig") -> None:
         parallel_config = vllm_config.parallel_config  # Not use scheduler_config
         # scheduler_config = vllm_config.scheduler_config
         model_config = vllm_config.model_config
+        if model_config is not None:
+            _patch_qwen3_vl_text_config(getattr(model_config, "hf_config", None))
 
         if parallel_config.worker_cls == "auto":
             # v0.15.1 do not support v0.15.1, remove the if condition