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

Drop 0.10.2 (#3284)

Browse Source 
Drop v0.10.2 support, we support vLLM 0.11.0rc3 now.
- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/releases/v0.11.0

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-10-09 10:28:38 +08:00
committed by GitHub
parent 2dde1268c7
commit f12f76d7ba
17 changed files with 202 additions and 653 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
vllm_ascend/ops/fused_moe.py
View File

@@ -43,8 +43,7 @@ from vllm_ascend.ops.moe.experts_selector import select_experts
from vllm_ascend.ops.moe.moe_comm_method import setup_moe_comm_method
from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_NZ,
get_all_reduce_merge_state,
get_rm_router_logits_state, is_310p,
vllm_version_is)
get_rm_router_logits_state, is_310p)
class AscendUnquantizedFusedMoEMethod(UnquantizedFusedMoEMethod):
@@ -275,25 +274,14 @@ class AscendFusedMoE(FusedMoE):
if self.scoring_func != "softmax" and not self.use_grouped_topk:
raise ValueError("Only softmax scoring function is supported for "
"non-grouped topk.")
if vllm_version_is("0.10.2"):
moe = FusedMoEConfig.make(
num_experts=self.global_num_experts,
experts_per_token=top_k,
hidden_dim=hidden_size,
num_local_experts=self.local_num_experts,
moe_parallel_config=self.moe_parallel_config,
# TODO (bnell): this needs to be fixed for quantized types.
in_dtype=params_dtype,
quant_config=quant_config)
else:
moe = FusedMoEConfig(
num_experts=self.global_num_experts,
experts_per_token=top_k,
hidden_dim=hidden_size,
num_local_experts=self.local_num_experts,
moe_parallel_config=self.moe_parallel_config,
in_dtype=params_dtype,
)
moe = FusedMoEConfig(
num_experts=self.global_num_experts,
experts_per_token=top_k,
hidden_dim=hidden_size,
num_local_experts=self.local_num_experts,
moe_parallel_config=self.moe_parallel_config,
in_dtype=params_dtype,
)
self.moe_config = moe
# TODO: The self.moe_config.tp_size here is not correct, fixme soon

10
vllm_ascend/ops/moe/fused_moe_prepare_and_finalize.py
View File

@@ -26,8 +26,6 @@ from vllm.distributed.parallel_state import (
from vllm.forward_context import get_forward_context
from vllm.model_executor.layers.fused_moe import FusedMoEConfig
from vllm_ascend.utils import vllm_version_is
class FusedMoEPrepareAndFinalize(ABC):
"""
@@ -416,12 +414,8 @@ class FusedMoEPrepareAndFinalizeWithNaiveMulticast(FusedMoEPrepareAndFinalize):
self.enable_shared_expert_dp = enable_shared_expert_dp
if self.moe_config.dp_size > 1:
if vllm_version_is("0.10.2"):
self.cu_tokens_across_dp_cpu = get_forward_context(
).dp_metadata.cu_tokens_across_dp_cpu
else:
self.cu_tokens_across_dp_cpu = get_forward_context(
).dp_metadata.cu_tokens_across_sp(1)
self.cu_tokens_across_dp_cpu = get_forward_context(
).dp_metadata.cu_tokens_across_sp(1)
hidden_states = self._naive_multicast(hidden_states,
self.cu_tokens_across_dp_cpu)
if rm_router_logits: