Adding "torch compile" annotations to moe models (#9758)
This commit is contained in:
@@ -5,6 +5,7 @@ import torch
|
||||
from torch import nn
|
||||
|
||||
from vllm.attention import Attention, AttentionMetadata
|
||||
from vllm.compilation.decorators import support_torch_compile
|
||||
from vllm.config import CacheConfig
|
||||
from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
|
||||
get_tensor_model_parallel_world_size,
|
||||
@@ -360,6 +361,7 @@ class ArcticDecoderLayer(nn.Module):
|
||||
return hidden_states
|
||||
|
||||
|
||||
@support_torch_compile
|
||||
class ArcticModel(nn.Module):
|
||||
|
||||
def __init__(
|
||||
|
||||
Reference in New Issue
Block a user