[CPU] add CPU support for Voxtral (#53921)
Signed-off-by: Liron Kesem <52330564+LironKesem@users.noreply.github.com>
This commit is contained in:
@@ -32,6 +32,7 @@ from vllm.v1.attention.backend import (
|
||||
CommonAttentionMetadata,
|
||||
subclass_attention_backend_with_overrides,
|
||||
)
|
||||
from vllm.v1.attention.backends.cpu_attn import CPUAttentionBackend
|
||||
from vllm.v1.attention.backends.flash_attn import FlashAttentionBackend
|
||||
|
||||
try:
|
||||
@@ -225,6 +226,7 @@ def create_whisper_attention_backend_with_block_pooling(
|
||||
b
|
||||
for b in (
|
||||
AiterFlashAttentionBackend,
|
||||
CPUAttentionBackend,
|
||||
FlashAttentionBackend,
|
||||
RocmAttentionBackend,
|
||||
TritonAttentionBackend,
|
||||
@@ -239,7 +241,9 @@ def create_whisper_attention_backend_with_block_pooling(
|
||||
"appreciated."
|
||||
)
|
||||
|
||||
if not issubclass(underlying_attn_backend, FlashAttentionBackend):
|
||||
if not issubclass(
|
||||
underlying_attn_backend, (CPUAttentionBackend, FlashAttentionBackend)
|
||||
):
|
||||
logger.info(
|
||||
"Using %s for Whisper causal attention with block pooling. "
|
||||
"This backend was recently enabled for this model. "
|
||||
|
||||
Reference in New Issue
Block a user