[CI] Fix UT (#2452)

Make UT CI happy 

- vLLM version: v0.10.0
- vLLM main:
d983769c41

---------

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: MengqingCao <cmq0113@163.com>

tests/ut/core/test_scheduler.py

@@ -8,6 +8,8 @@ from vllm.config import (CacheConfig, KVTransferConfig, ModelConfig,
                          SchedulerConfig, SpeculativeConfig, VllmConfig)
 from vllm.multimodal.inputs import PlaceholderRange
 from vllm.sampling_params import SamplingParams
+from vllm.v1.core.kv_cache_utils import (get_request_block_hasher,
+                                         init_none_hash)
 from vllm.v1.core.sched.output import SchedulerOutput
 from vllm.v1.kv_cache_interface import (FullAttentionSpec, KVCacheConfig,
                                         KVCacheGroupSpec)
@@ -36,7 +38,10 @@ def create_requests(
     mm_positions: Optional[list[PlaceholderRange]] = None,
     max_tokens: int = 16,
     stop_token_ids: Optional[list[int]] = None,
+    block_size: int = 3,
+    hash_fn=hash,
 ):
+    init_none_hash(hash_fn)
     prompt_logprobs = PROMPT_LOGPROBS
     sampling_params = SamplingParams(ignore_eos=False,
                                      max_tokens=max_tokens,
@@ -46,16 +51,16 @@ def create_requests(
     for i in range(num_requests):
         mm_position = None
         mm_inputs = None
-        request = Request(
-            request_id=f"{i}",
-            prompt_token_ids=[i] * num_tokens,
-            sampling_params=sampling_params,
-            multi_modal_kwargs=mm_inputs,
-            multi_modal_placeholders=mm_position,
-            multi_modal_hashes=None,
-            eos_token_id=EOS_TOKEN_ID,
-            pooling_params=None,
-        )
+        request = Request(request_id=f"{i}",
+                          prompt_token_ids=[i] * num_tokens,
+                          sampling_params=sampling_params,
+                          multi_modal_kwargs=mm_inputs,
+                          multi_modal_placeholders=mm_position,
+                          multi_modal_hashes=None,
+                          eos_token_id=EOS_TOKEN_ID,
+                          pooling_params=None,
+                          block_hasher=get_request_block_hasher(
+                              block_size, hash_fn))
         requests.append(request)
     return requests
 

tests/ut/kv_connector/test_mooncake_connector.py

@@ -1152,8 +1152,6 @@ class TestMooncakeConnectorWorker(unittest.TestCase):
                   MagicMock()),
             patch.dict('sys.modules',
                        {'vllm_ascend.envs': self.envs_ascend_mock}),
-            patch('vllm_ascend.distributed.mooncake_connector.envs_ascend',
-                  self.envs_ascend_mock),
         ]
 
         for p in self.patches:

tests/ut/kv_connector/utils.py

@@ -55,7 +55,6 @@ def assert_scheduler_empty(scheduler: Scheduler):
 
 
 def create_vllm_config(
-    model: str = "facebook/opt-125m",
     max_num_seqs: int = 16,
     max_num_batched_tokens: int = 1024,
     block_size: int = 128,
@@ -66,14 +65,11 @@ def create_vllm_config(
         max_num_batched_tokens=max_num_batched_tokens,
         max_model_len=max_num_batched_tokens,
     )
+    fake_weight_path = os.path.join(os.path.dirname(__file__), "..",
+                                    "fake_weight")
     model_config = ModelConfig(
-        model=model,
-        task="auto",
-        tokenizer=model,
-        tokenizer_mode="auto",
-        trust_remote_code=True,
-        dtype="float16",
-        seed=42,
+        model=fake_weight_path,
+        skip_tokenizer_init=True,
     )
     # Cache config, optionally force APC
     cache_config = CacheConfig(

tests/ut/torchair/test_utils.py

@@ -51,7 +51,7 @@ class TestTorchairUtils(TestBase):
         mock_model_registry.return_value = mock_registry
         utils.register_torchair_model()
 
-        self.assertEqual(mock_model_registry.register_model.call_count, 3)
+        self.assertEqual(mock_model_registry.register_model.call_count, 5)
         call_args_list = mock_model_registry.register_model.call_args_list
 
         expected_registrations = [
@@ -63,7 +63,11 @@ class TestTorchairUtils(TestBase):
              ),
             ("DeepseekV3ForCausalLM",
              "vllm_ascend.torchair.models.torchair_deepseek_v3:TorchairDeepseekV3ForCausalLM"
-             )
+             ),
+            ("Qwen2ForCausalLM",
+             "vllm_ascend.torchair.models.qwen2:CustomQwen2ForCausalLM"),
+            ("Qwen3ForCausalLM",
+             "vllm_ascend.torchair.models.qwen3_moe:CustomQwen3MoeForCausalLM")
         ]
 
         for i, (expected_name,

tests/ut/worker/test_input_batch.py

@@ -20,7 +20,7 @@ import numpy as np
 import pytest
 import torch
 from vllm.sampling_params import SamplingParams
-from vllm.utils import is_pin_memory_available, make_tensor_with_pad
+from vllm.utils import make_tensor_with_pad
 from vllm.v1.pool.metadata import PoolingMetadata
 from vllm.v1.sample.logits_processor import LogitsProcessors
 from vllm.v1.sample.metadata import SamplingMetadata
@@ -237,7 +237,7 @@ def test_sampling_metadata_in_input_batch(device: str, batch_size: int):
         max_model_len=1024,
         max_num_batched_tokens=1024,
         device=torch.device(device),
-        pin_memory=is_pin_memory_available(),
+        pin_memory=False,
         vocab_size=1024,
         block_sizes=[1],
     )
@@ -298,7 +298,7 @@ def test_sampling_metadata_in_input_batch(device: str, batch_size: int):
     assert (expected_sampling_metadata.output_token_ids ==
             sampling_metadata.output_token_ids)
     assert expected_sampling_metadata.no_penalties == \
-           sampling_metadata.no_penalties
+        sampling_metadata.no_penalties
     if sampling_metadata.allowed_token_ids_mask:
         assert torch.allclose(
             expected_sampling_metadata.allowed_token_ids_mask,
@@ -328,7 +328,7 @@ def test_swap_states_in_input_batch(device: str, batch_size: int,
         max_model_len=1024,
         max_num_batched_tokens=1024,
         device=torch.device(device),
-        pin_memory=is_pin_memory_available(),
+        pin_memory=False,
         vocab_size=1024,
         block_sizes=[1],
     )
@@ -337,7 +337,7 @@ def test_swap_states_in_input_batch(device: str, batch_size: int,
         max_model_len=1024,
         max_num_batched_tokens=1024,
         device=torch.device(device),
-        pin_memory=is_pin_memory_available(),
+        pin_memory=False,
         vocab_size=1024,
         block_sizes=[1],
     )