[Bugfix][V1] Sync mamba_block_size via EngineCoreReadyResponse (#50809)
Signed-off-by: “新颐” <lxy537258@antgroup.com> Co-authored-by: “新颐” <lxy537258@antgroup.com>
This commit is contained in:
@@ -343,6 +343,42 @@ def test_apply_ready_response_syncs_block_size():
|
||||
assert client.vllm_config.cache_config.block_size == 1056
|
||||
|
||||
|
||||
def test_apply_ready_response_syncs_mamba_block_size():
|
||||
import msgspec
|
||||
|
||||
client = object.__new__(MPClient)
|
||||
client.vllm_config = SimpleNamespace(
|
||||
cache_config=SimpleNamespace(block_size=16, num_gpu_blocks=0),
|
||||
model_config=SimpleNamespace(max_model_len=8192),
|
||||
)
|
||||
client.stats_update_address = None
|
||||
|
||||
payload = msgspec.msgpack.encode(
|
||||
EngineCoreReadyResponse(
|
||||
max_model_len=8192,
|
||||
num_gpu_blocks=100,
|
||||
block_size=1056,
|
||||
dp_stats_address=None,
|
||||
dtype="bfloat16",
|
||||
vllm_version="test",
|
||||
world_size=1,
|
||||
data_parallel_size=1,
|
||||
tensor_parallel_size=1,
|
||||
pipeline_parallel_size=1,
|
||||
decode_context_parallel_size=1,
|
||||
data_parallel_rank=0,
|
||||
max_num_seqs=256,
|
||||
max_num_batched_tokens=8192,
|
||||
instance_id="test-instance",
|
||||
supports_lora=False,
|
||||
max_loras=0,
|
||||
mamba_block_size=1056,
|
||||
)
|
||||
)
|
||||
client._apply_ready_response(payload)
|
||||
assert client.vllm_config.cache_config.mamba_block_size == 1056
|
||||
|
||||
|
||||
def loop_until_done(client: EngineCoreClient, outputs: dict):
|
||||
while True:
|
||||
engine_core_outputs = client.get_output().outputs
|
||||
|
||||
Reference in New Issue
Block a user