Adding "torch compile" annotations to moe models (#9758)

This commit is contained in:
Yongzao
2024-10-28 13:49:56 -07:00
committed by GitHub
parent 5f8d8075f9
commit aa0addb397
4 changed files with 8 additions and 0 deletions
+2
View File
@@ -28,6 +28,7 @@ from torch import nn
from transformers import MixtralConfig
from vllm.attention import Attention, AttentionMetadata
from vllm.compilation.decorators import support_torch_compile
from vllm.config import CacheConfig, LoRAConfig
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
from vllm.model_executor.layers.fused_moe import FusedMoE
@@ -245,6 +246,7 @@ class MixtralDecoderLayer(nn.Module):
return hidden_states, residual
@support_torch_compile
class MixtralModel(nn.Module):
def __init__(