biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Attention] Implement universal BACKEND_MAP (#25900)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2025-10-08 15:00:25 -04:00
committed by GitHub
parent b25d7b5657
commit 76879cc160
12 changed files with 119 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/v1/spec_decode/test_mtp.py
View File

@@ -10,7 +10,7 @@ from tests.v1.attention.utils import (
BatchSpec,
create_common_attn_metadata,
create_standard_kv_cache_spec,
get_attention_backend,
try_get_attention_backend,
)
from vllm.attention.backends.registry import _Backend
from vllm.config import (
@@ -177,7 +177,7 @@ def test_mtp_propose(num_speculative_tokens, monkeypatch):
sampling_metadata = mock.MagicMock()
# Setup attention metadata
attn_metadata_builder_cls, _ = get_attention_backend(_Backend.FLASH_ATTN)
attn_metadata_builder_cls, _ = try_get_attention_backend(_Backend.FLASH_ATTN)
attn_metadata_builder = attn_metadata_builder_cls(
kv_cache_spec=create_standard_kv_cache_spec(proposer.vllm_config),