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

[Bugfix] Add OOT backend option (#32471)

Browse Source 
Signed-off-by: Iryna Boiko <iboiko@habana.ai>
This commit is contained in:
Iryna Boiko
2026-01-18 23:20:39 +01:00
committed by GitHub
parent eebc58df0c
commit f5d1740030
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/fused_moe/oracle/unquantized.py
View File

@@ -33,6 +33,7 @@ class UnquantizedMoeBackend(Enum):
CPU = "CPU" CPU = "CPU"
XPU = "XPU" XPU = "XPU"
TPU = "TPU" TPU = "TPU"
OOT = "OOT"
# NOTE(zyongye): Unsupported backend means backend # NOTE(zyongye): Unsupported backend means backend
@@ -42,6 +43,7 @@ UNSUPPORTED_BACKEND = [
UnquantizedMoeBackend.CPU, UnquantizedMoeBackend.CPU,
UnquantizedMoeBackend.XPU, UnquantizedMoeBackend.XPU,
UnquantizedMoeBackend.TPU, UnquantizedMoeBackend.TPU,
UnquantizedMoeBackend.OOT,
] ]
@@ -95,6 +97,8 @@ def select_unquantized_moe_backend(
backend = UnquantizedMoeBackend.CPU backend = UnquantizedMoeBackend.CPU
if current_platform.is_tpu(): if current_platform.is_tpu():
backend = UnquantizedMoeBackend.TPU backend = UnquantizedMoeBackend.TPU
if current_platform.is_out_of_tree():
backend = UnquantizedMoeBackend.OOT
logger.info_once(_make_log_backend(backend), scope="local") logger.info_once(_make_log_backend(backend), scope="local")
return backend return backend