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

Drop 0.12.0 support (#5146)

Browse Source 
We decided to release v0.13.0 soon. So no need to support 0.12.0 now.
Let's drop it.

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-12-20 09:38:53 +08:00
committed by GitHub
parent 243ab7d720
commit 758d81dcb1
21 changed files with 63 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
vllm_ascend/worker/model_runner_v1.py
View File

@@ -116,8 +116,7 @@ from vllm_ascend.spec_decode.interface import SpecDcodeType
from vllm_ascend.spec_decode.mtp_proposer import MtpProposer
from vllm_ascend.utils import (AscendDeviceType, ProfileExecuteDuration,
enable_sp, get_ascend_device_type, is_moe_model,
lmhead_tp_enable, maybe_trans_nz,
vllm_version_is)
lmhead_tp_enable, maybe_trans_nz)
from vllm_ascend.worker.npu_input_batch import NPUInputBatch
from vllm_ascend.ascend_forward_context import ( # isort: skip
@@ -243,24 +242,15 @@ class NPUModelRunner(GPUModelRunner):
# Set up Attention
self.use_sparse = hasattr(self.vllm_config.model_config.hf_config,
"index_topk")
if vllm_version_is('0.12.0'):
self.attn_backend = get_attn_backend(
0,
self.dtype,
None,
self.block_size,
use_mla=self.model_config.use_mla,
use_sparse=self.use_sparse)
else:
self.attn_backend = get_attn_backend(
0,
self.dtype,
None,
self.block_size,
use_mla=self.model_config.use_mla,
use_sparse=self.use_sparse,
use_mm_prefix=self.model_config is not None
and self.model_config.is_mm_prefix_lm)
self.attn_backend = get_attn_backend(
0,
self.dtype,
None,
self.block_size,
use_mla=self.model_config.use_mla,
use_sparse=self.use_sparse,
use_mm_prefix=self.model_config is not None
and self.model_config.is_mm_prefix_lm)
self.attn_mask_builder = AttentionMaskBuilder(self.device)
self._set_up_drafter()
@@ -1877,36 +1867,19 @@ class NPUModelRunner(GPUModelRunner):
self.speculative_config.method == "mtp":
attn_state = AscendAttentionState.SpecDecoding
if vllm_version_is("0.12.0"):
common_metadata = CommonAttentionMetadata(
query_start_loc=self.query_start_loc.gpu[:num_reqs +
common_metadata = CommonAttentionMetadata(
query_start_loc=self.query_start_loc.gpu[:num_reqs + 1],
query_start_loc_cpu=self.query_start_loc.cpu[:num_reqs +
1],
query_start_loc_cpu=self.query_start_loc.
cpu[:num_reqs + 1],
seq_lens_cpu=self.seq_lens.cpu[:num_reqs],
seq_lens=self.seq_lens.cpu[:num_reqs],
num_reqs=num_reqs,
num_actual_tokens=num_tokens,
block_table_tensor=block_table_tensor[:num_reqs],
slot_mapping=slot_mapping.gpu,
num_computed_tokens_cpu=num_computed_tokens_cpu,
max_query_len=max_query_len,
max_seq_len=seq_lens)
else:
common_metadata = CommonAttentionMetadata(
query_start_loc=self.query_start_loc.gpu[:num_reqs +
1],
query_start_loc_cpu=self.query_start_loc.
cpu[:num_reqs + 1],
_seq_lens_cpu=self.seq_lens.cpu[:num_reqs],
seq_lens=self.seq_lens.cpu[:num_reqs],
num_reqs=num_reqs,
num_actual_tokens=num_tokens,
block_table_tensor=block_table_tensor[:num_reqs],
slot_mapping=slot_mapping.gpu,
_num_computed_tokens_cpu=num_computed_tokens_cpu,
max_query_len=max_query_len,
max_seq_len=seq_lens)
_seq_lens_cpu=self.seq_lens.cpu[:num_reqs],
seq_lens=self.seq_lens.cpu[:num_reqs],
num_reqs=num_reqs,
num_actual_tokens=num_tokens,
block_table_tensor=block_table_tensor[:num_reqs],
slot_mapping=slot_mapping.gpu,
_num_computed_tokens_cpu=num_computed_tokens_cpu,
max_query_len=max_query_len,
max_seq_len=seq_lens)
for attn_group in self.attn_groups[kv_cache_group_id]:
builder = attn_group.get_metadata_builder()