[Bugfix] Follow vLLM Qwen-Moe/VL and KV Connector change to fix broken CI (#2181)

### What this PR does / why we need it?
This pr fix broken CI:
1. Fix the
ee2eb6ecd8
changes, in this commit, they fused the gate and up projections in the
vision MLP, This can improve performance by reducing one matrix
multiplication. so, this pr do the following things:
- Specify that the two linear layers are fused as `mlp.gate_up_proj`
when loading the weights.
    - Use a SiluAndMul activation function.
2. Fix
aefeea0fde,
Update ModelRunnerOutput parameters to adapt to its changes
3. Fix
[vllm-commit](https://github.com/vllm-project/vllm/pull/20815/files#diff-3ffb829a39ab2b3e4706aa28f5e476815f36c3a87b98d6a66514ebedc8f3ffb4R354-R356),
fix qwen moe
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?


- vLLM version: v0.10.0
- vLLM main:
fed5849d3f

---------

Signed-off-by: wangli <wangli858794774@gmail.com>

tests/ut/kv_connector/test_remote_decode_lifecycle.py

@@ -25,6 +25,7 @@ from tests.ut.kv_connector.utils import (assert_scheduler_empty,
                                          create_model_runner_output,
                                          create_request, create_scheduler,
                                          create_vllm_config)
+from vllm_ascend.utils import vllm_version_is
 
 
 def test_basic_lifecycle():
@@ -102,7 +103,13 @@ def test_basic_lifecycle():
 
     # (3b): execute_model()
     model_runner_output = copy.deepcopy(EMPTY_MODEL_RUNNER_OUTPUT)
-    model_runner_output.finished_sending = [request_id]
+    if vllm_version_is("0.10.0"):
+        model_runner_output.finished_sending = [request_id]
+    else:
+        from vllm.v1.worker.kv_connector_model_runner_mixin import \
+            KVConnectorOutput  # type: ignore  # noqa
+        model_runner_output.kv_connector_output = KVConnectorOutput(
+            finished_sending=[request_id])
 
     # (3c): update_from_output()
     scheduler.update_from_output(scheduler_output, model_runner_output)
@@ -157,7 +164,13 @@ def test_prefix_cache_lifecycle():
     scheduler_output = scheduler.schedule()
     scheduler.schedule()
     model_runner_output = copy.deepcopy(EMPTY_MODEL_RUNNER_OUTPUT)
-    model_runner_output.finished_sending = [request_remote.request_id]
+    if vllm_version_is("0.10.0"):
+        model_runner_output.finished_sending = [request_remote.request_id]
+    else:
+        from vllm.v1.worker.kv_connector_model_runner_mixin import \
+            KVConnectorOutput  # noqa
+        model_runner_output.kv_connector_output = KVConnectorOutput(
+            finished_sending=[request_remote.request_id])
     scheduler.update_from_output(scheduler_output, model_runner_output)
     _ = scheduler.schedule()
     assert_scheduler_empty(scheduler)

tests/ut/kv_connector/test_remote_prefill_lifecycle.py

@@ -19,7 +19,7 @@
 import copy
 
 from vllm.v1.outputs import EMPTY_MODEL_RUNNER_OUTPUT
-from vllm.v1.request import FinishReason, RequestStatus
+from vllm.v1.request import RequestStatus
 
 from tests.ut.kv_connector.utils import (assert_scheduler_empty,
                                          create_model_runner_output,
@@ -55,10 +55,7 @@ def test_basic_lifecycle():
     # Nothing running and empty scheduler output.
     assert len(scheduler.running) == 0
     assert len(scheduler_output.scheduled_new_reqs) == 0
-    if vllm_version_is("0.9.1"):
-        assert len(scheduler_output.scheduled_cached_reqs) == 0
-    else:
-        assert scheduler_output.scheduled_cached_reqs.num_reqs == 0
+    assert scheduler_output.scheduled_cached_reqs.num_reqs == 0
     assert len(scheduler_output.num_scheduled_tokens) == 0
     assert scheduler_output.total_num_scheduled_tokens == 0
 
@@ -94,7 +91,13 @@ def test_basic_lifecycle():
 
     # (2b): forward(): request finishes recv.
     model_runner_output = copy.deepcopy(EMPTY_MODEL_RUNNER_OUTPUT)
-    model_runner_output.finished_recving = [request_id]
+    if vllm_version_is("0.10.0"):
+        model_runner_output.finished_recving = [request_id]
+    else:
+        from vllm.v1.worker.kv_connector_model_runner_mixin import \
+            KVConnectorOutput  # type: ignore  # noqa
+        model_runner_output.kv_connector_output = KVConnectorOutput(
+            finished_recving=[request_id])
 
     # (2c): update_from_output():
     engine_core_outputs = scheduler.update_from_output(scheduler_output,
@@ -135,11 +138,6 @@ def test_basic_lifecycle():
                                                        model_runner_output)
     scheduler.schedule()
 
-    if vllm_version_is("0.9.1"):
-        outputs = engine_core_outputs[0].outputs
-        assert len(outputs) == 1
-        output = outputs[0]
-        assert output.finish_reason == FinishReason.STOP
     assert_scheduler_empty(scheduler)
 
 
@@ -213,7 +211,13 @@ def test_full_block_prompt():
     # # STEP (2): Recv.
     scheduler_output = scheduler.schedule()
     model_runner_output = copy.deepcopy(EMPTY_MODEL_RUNNER_OUTPUT)
-    model_runner_output.finished_recving = [request_id]
+    if vllm_version_is("0.10.0"):
+        model_runner_output.finished_recving = [request_id]
+    else:
+        from vllm.v1.worker.kv_connector_model_runner_mixin import \
+            KVConnectorOutput  # type: ignore  # noqa
+        model_runner_output.kv_connector_output = KVConnectorOutput(
+            finished_recving=[request_id])
     scheduler.update_from_output(scheduler_output, model_runner_output)
     assert len(scheduler.waiting) == 1
     assert (request_id in scheduler.finished_recving_kv_req_ids)
@@ -236,13 +240,6 @@ def test_full_block_prompt():
     # # Step (4): Hit EOS.
     scheduler_output = scheduler.schedule()
     model_runner_output = create_model_runner_output([request], use_eos=True)
-    engine_core_outputs = scheduler.update_from_output(scheduler_output,
-                                                       model_runner_output)
     scheduler.schedule()
 
-    if vllm_version_is("0.9.1"):
-        outputs = engine_core_outputs[0].outputs
-        assert len(outputs) == 1
-        output = outputs[0]
-        assert output.finish_reason == FinishReason.STOP
     assert_scheduler_empty(scheduler)

tests/ut/kv_connector/utils.py

@@ -186,6 +186,20 @@ def create_model_runner_output(
     sampled_token_ids = [[sampled_token] for _ in req_ids]
 
     # Make output data structure.
+    extra_args = {}
+    if not vllm_version_is("0.10.0"):
+        from vllm.v1.worker.kv_connector_model_runner_mixin import \
+            KVConnectorOutput  # type: ignore  # noqa
+        kv_connector_output = KVConnectorOutput(
+            finished_sending=finished_sending,
+            finished_recving=finished_recving)
+        extra_args = {"kv_connector_output": kv_connector_output}
+    else:
+        extra_args = {
+            "finished_sending": finished_sending,
+            "finished_recving": finished_recving,
+        }
+
     return ModelRunnerOutput(
         req_ids=req_ids,
         req_id_to_index=req_id_to_index,
@@ -193,9 +207,6 @@ def create_model_runner_output(
         spec_token_ids=None,
         logprobs=None,
         prompt_logprobs_dict={},
-        **({
-            "pooler_output": []
-        } if not vllm_version_is("0.9.1") else {}),
-        finished_sending=finished_sending,
-        finished_recving=finished_recving,
+        pooler_output=[],
+        **extra_args,
     )