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

[CI] upgrade to vllm 0.9.0 (#959)

Browse Source 
Upgrade to vllm 0.9.0.
0.8.5 will not be supported any more.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-05-28 21:18:41 +08:00
committed by GitHub
parent e2a0c19cea
commit f6e5decc10
16 changed files with 79 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm_ascend/attention/attention_v1.py
View File

@@ -30,7 +30,6 @@ from vllm.v1.core.sched.output import SchedulerOutput
from vllm.v1.worker.gpu_input_batch import InputBatch
from vllm_ascend.ops.attention import vanilla_chunked_prefill
from vllm_ascend.utils import vllm_version_is
class AscendAttentionBackend(AttentionBackend):
@@ -142,14 +141,11 @@ class AscendAttentionMetadataBuilder:
def build(self, num_reqs, num_actual_tokens, max_query_len,
common_prefix_len):
if vllm_version_is("0.8.5") or vllm_version_is("0.8.5.post1"):
block_table = (self.runner.input_batch.block_table.
get_device_tensor()[:num_reqs])
else:
block_table = self.runner.input_batch.block_table[
0].get_device_tensor()
block_table[:num_reqs, :self.runner.max_num_blocks_per_req] = (
block_table[:num_reqs])
block_table = self.runner.input_batch.block_table[0].get_device_tensor(
)
block_table[:num_reqs, :self.runner.max_num_blocks_per_req] = (
block_table[:num_reqs])
query_lens = self.runner.query_lens
seq_lens = self.runner.seq_lens_cpu[:num_reqs]

14
vllm_ascend/attention/mla_v1.py
View File

@@ -16,7 +16,6 @@ from vllm.model_executor.layers.rotary_embedding import RotaryEmbedding
from vllm_ascend.attention.attention_v1 import AscendAttentionState
from vllm_ascend.ops.attention import vanilla_chunked_prefill_mla
from vllm_ascend.utils import vllm_version_is
from vllm_ascend.worker.model_runner_v1 import NPUModelRunner
if TYPE_CHECKING:
@@ -239,14 +238,11 @@ class AscendMLAMetadataBuilder:
# function. We should avoid GPU -> CPU sync as much as possible because
# it blocks on all previous kernels.
device = self.runner.device
if vllm_version_is("0.8.5") or vllm_version_is("0.8.5.post1"):
block_table = (self.runner.input_batch.block_table.
get_device_tensor()[:num_reqs])
else:
block_table = self.runner.input_batch.block_table[
0].get_device_tensor()
block_table[:num_reqs, :self.runner.max_num_blocks_per_req] = (
block_table[:num_reqs])
block_table = self.runner.input_batch.block_table[0].get_device_tensor(
)
block_table[:num_reqs, :self.runner.max_num_blocks_per_req] = (
block_table[:num_reqs])
slot_mapping = self.runner.slot_mapping_cpu[:num_actual_tokens].to(
device, non_blocking=True)
input_positions = self.runner.positions_cpu[:num_actual_tokens].to(