A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 23:45 UTC
support `torch.compile` for bailing moe (#21664)
Jinzhen Lin committed
Jul 27, 2025 at 07:54 UTC
c6573698414c88990c3ddcd5386d26fb5911c59d
1 file changed
+2
vllm/model_executor/models/bailing_moe.py
+2
@@ -32,6 +32,7 @@ from torch import nn
32
from transformers.configuration_utils import PretrainedConfig
33
34
from vllm.attention import Attention
35
+from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
38
get_tensor_model_parallel_world_size,
@@ -291,6 +292,7 @@ class BailingMoeBlock(nn.Module):
292
return hidden_states, residual
293
294
295
+@support_torch_compile
296
class BailingMoeModel(nn.Module):
297
298
def __init__(