Add FLASHINFER_MLA to test_mla_backends and add B200 CI run (#27663)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
@@ -6,7 +6,7 @@ from typing import ClassVar
|
||||
import torch
|
||||
from flashinfer.decode import trtllm_batch_decode_with_kv_cache_mla
|
||||
|
||||
from vllm.attention.backends.abstract import AttentionLayer, AttentionType
|
||||
from vllm.attention.backends.abstract import AttentionLayer, AttentionType, MultipleOf
|
||||
from vllm.logger import init_logger
|
||||
from vllm.v1.attention.backends.mla.common import (
|
||||
MLACommonBackend,
|
||||
@@ -40,6 +40,10 @@ class FlashInferMLABackend(MLACommonBackend):
|
||||
def get_builder_cls() -> type["FlashInferMLAMetadataBuilder"]:
|
||||
return FlashInferMLAMetadataBuilder
|
||||
|
||||
@classmethod
|
||||
def get_supported_kernel_block_size(cls) -> list[int | MultipleOf]:
|
||||
return [32, 64]
|
||||
|
||||
|
||||
g_fi_workspace = torch.zeros(
|
||||
FLASHINFER_MLA_WORKSPACE_BUFFER_SIZE,
|
||||
|
||||
Reference in New Issue
Block a user