EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Revert "[Bugfix] support mtp kv transfer and pp partition by hand in kv transfer (#4892)" (#4981)

Browse Source 
This reverts commit 332b547728.

This break deepseek3.2 in PD case.

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
This commit is contained in:
wangxiyuan
2025-12-13 18:58:55 +08:00
committed by GitHub
parent 31c94b7e7b
commit 5211e991ad
2 changed files with 14 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/ut/kv_connector/test_mooncake_connector.py
View File

@@ -242,8 +242,7 @@ class TestKVCacheRecvingThreadBasic(unittest.TestCase):
block_len=[1024, 2048],
ready_event=self.ready_event,
vllm_config=self.vllm_config,
kv_caches=self.kv_caches,
prefill_pp_layer_partition=None)
kv_caches=self.kv_caches)
def test_add_request(self):
test_req = {
@@ -296,8 +295,7 @@ class TestSocketManagement(unittest.TestCase):
block_len=[1024, 2048],
ready_event=self.ready_event,
vllm_config=self.vllm_config,
kv_caches=self.kv_caches,
prefill_pp_layer_partition=None)
kv_caches=self.kv_caches)
self.thread.remote_sockets = defaultdict(deque)
self.thread.remote_poller = MagicMock()
@@ -354,8 +352,7 @@ class TestCoreFunctionality(unittest.TestCase):
block_len=[1024, 2048],
ready_event=self.ready_event,
vllm_config=self.vllm_config,
kv_caches=self.kv_caches,
prefill_pp_layer_partition=None)
kv_caches=self.kv_caches)
self.thread.request_queue = self.mock_queue
self.test_req = {
"request_id": "req1",
@@ -437,8 +434,7 @@ class TestMetadataHandling(unittest.TestCase):
block_len=[1024, 2048],
ready_event=self.ready_event,
vllm_config=self.vllm_config,
kv_caches=self.kv_caches,
prefill_pp_layer_partition=None)
kv_caches=self.kv_caches)
self.test_metadata = MooncakeAgentMetadata(
engine_id="remote_engine",
te_rpc_port=9090,
@@ -502,8 +498,7 @@ class TestMainThreadLoop(unittest.TestCase):
block_len=[1024, 2048],
ready_event=self.ready_event,
vllm_config=self.vllm_config,
kv_caches=self.kv_caches,
prefill_pp_layer_partition=None)
kv_caches=self.kv_caches)
self.thread.request_queue = queue.Queue()
@patch.object(KVCacheRecvingThread, '_handle_request')
@@ -540,7 +535,6 @@ class MockVllmConfig:
self.parallel_config = MagicMock()
self.cache_config = MagicMock()
self.kv_transfer_config = MagicMock()
self.speculative_config = MagicMock()
self.model_config.use_mla = True
self.parallel_config.tensor_parallel_size = 2
self.parallel_config.data_parallel_rank = 0