chore: import upstream snapshot with attribution

2026-07-13 12:55:37 +08:00
commit 7ce4c8e27e
5900 changed files with 1668062 additions and 0 deletions
@@ -0,0 +1,78 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+from unittest.mock import Mock, patch
+
+
+def test_qwen3_5_lm_head_receives_quant_config():
+    from vllm.model_executor.models.qwen3_5 import Qwen3_5ForCausalLMBase
+
+    mock_quant_config = Mock()
+
+    mock_hf_config = Mock()
+    mock_hf_config.tie_word_embeddings = False
+    mock_hf_config.vocab_size = 128
+    mock_hf_config.hidden_size = 64
+
+    mock_vllm_config = Mock()
+    mock_vllm_config.model_config.hf_text_config = mock_hf_config
+    mock_vllm_config.cache_config.mamba_cache_mode = "align"
+    mock_vllm_config.scheduler_config = Mock()
+    mock_vllm_config.quant_config = mock_quant_config
+    mock_vllm_config.lora_config = None
+
+    mock_pp_group = Mock()
+    mock_pp_group.is_last_rank = True
+
+    with (
+        patch("vllm.model_executor.models.qwen3_5.Qwen3_5Model") as MockModel,
+        patch("vllm.model_executor.models.qwen3_5.ParallelLMHead") as MockLMHead,
+        patch("vllm.model_executor.models.qwen3_5.LogitsProcessor"),
+        patch(
+            "vllm.model_executor.models.qwen3_5.get_pp_group",
+            return_value=mock_pp_group,
+        ),
+    ):
+        MockModel.return_value.make_empty_intermediate_tensors = Mock()
+
+        Qwen3_5ForCausalLMBase(vllm_config=mock_vllm_config)
+
+        MockLMHead.assert_called_once()
+        call_kwargs = MockLMHead.call_args.kwargs
+        assert call_kwargs["quant_config"] is mock_quant_config
+
+
+def test_qwen3_5_mtp_lm_head_receives_quant_config():
+    from vllm.config import CompilationMode
+    from vllm.model_executor.models.qwen3_5_mtp import Qwen3_5MTP
+
+    mock_quant_config = Mock()
+
+    mock_hf_config = Mock()
+    mock_hf_config.tie_word_embeddings = False
+    mock_hf_config.vocab_size = 128
+    mock_hf_config.hidden_size = 64
+
+    mock_vllm_config = Mock()
+    mock_vllm_config.model_config.hf_text_config = mock_hf_config
+    mock_vllm_config.cache_config.mamba_cache_mode = "align"
+    mock_vllm_config.compilation_config.mode = CompilationMode.NONE
+    mock_vllm_config.quant_config = mock_quant_config
+
+    mock_pp_group = Mock()
+    mock_pp_group.is_last_rank = True
+
+    with (
+        patch("vllm.model_executor.models.qwen3_5_mtp.Qwen3_5MultiTokenPredictor"),
+        patch("vllm.model_executor.models.qwen3_5_mtp.ParallelLMHead") as MockLMHead,
+        patch("vllm.model_executor.models.qwen3_5_mtp.LogitsProcessor"),
+        patch(
+            "vllm.model_executor.models.qwen3_5_mtp.get_pp_group",
+            return_value=mock_pp_group,
+        ),
+    ):
+        Qwen3_5MTP(vllm_config=mock_vllm_config)
+
+        MockLMHead.assert_called_once()
+        call_kwargs = MockLMHead.call_args.kwargs
+        assert call_kwargs["quant_config"] is mock_quant_config