[Performance] Optimize Dots3 NOTE runtime (#53517)

Signed-off-by: KurodaKanbei <mistergalahad@gmail.com>
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
Co-authored-by: Jee Jee Li <jeejeelee@inferact.ai>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
范裕达
2026-08-31 00:05:18 -07:00
committed by GitHub
co-authored by Jee Jee Li Jee Jee Li
parent 7292ee2791
commit da0b2d8b17
6 changed files with 94 additions and 85 deletions
+7 -1
View File
@@ -7,6 +7,7 @@ varlen MHA. Decode-only batches use the Triton absorbed-MQA kernel.
"""
from dataclasses import dataclass
from typing import ClassVar
import torch
@@ -19,7 +20,11 @@ from vllm.model_executor.layers.attention.mla_attention import (
from vllm.platforms import current_platform
from vllm.triton_utils import tl, triton
from vllm.utils.torch_utils import is_quantized_kv_cache
from vllm.v1.attention.backend import AttentionLayer, CommonAttentionMetadata
from vllm.v1.attention.backend import (
AttentionCGSupport,
AttentionLayer,
CommonAttentionMetadata,
)
from vllm.v1.attention.backends.mla.flashattn_mla_sparse import (
FlashAttnMLASparseBackend,
FlashAttnMLASparseImpl,
@@ -307,6 +312,7 @@ class Dots3NoteFlashAttnPrefillBackend(FlashAttnPrefillBackend):
class Dots3NoteMLAMetadataBuilder(TritonMLAMetadataBuilder):
"""Keep decode on MQA and route prefill/mixed batches through FA3."""
_cudagraph_support: ClassVar[AttentionCGSupport] = AttentionCGSupport.UNIFORM_BATCH
query_len_support = QueryLenSupport.UNIFORM
def __init__(self, kv_cache_spec, layer_names, vllm_config, device):