xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 08:04:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[FIX] Fix the case when input_is_parallel=False for ScaledActivation (#1737)

Browse Source
This commit is contained in:
Zhuohan Li 2023-11-20 23:56:48 -08:00 committed by GitHub
parent cf35d8f3d7
commit 7d761fe3c1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/layers/activation.py
View File

@ -61,6 +61,7 @@ class ScaledActivation(nn.Module):
):
super().__init__()
self.act = act_module
self.input_is_parallel = input_is_parallel
if input_is_parallel:
tp_size = get_tensor_model_parallel_world_size()
intermediate_size_per_partition = divide(intermediate_size,
@ -79,11 +80,12 @@ class ScaledActivation(nn.Module):
return self.act(x) / self.scales
def weight_loader(self, param: nn.Parameter, loaded_weight: torch.Tensor):
tp_rank = get_tensor_model_parallel_rank()
param_data = param.data
shard_size = param_data.shape[0]
start_idx = tp_rank * shard_size
loaded_weight = loaded_weight.narrow(0, start_idx, shard_size)
if self.input_is_parallel:
tp_rank = get_tensor_model_parallel_rank()
shard_size = param_data.shape[0]
start_idx = tp_rank * shard_size
loaded_weight = loaded_weight.narrow(0, start_idx, shard_size)
assert param_data.shape == loaded_weight.shape
param_data.copy_(loaded_weight)