xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 03:54:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

[FIX] Don't initialize parameter by default (#1067)

Browse Source
This commit is contained in:
Zhuohan Li 2023-09-17 17:15:38 -07:00 committed by GitHub
parent e21d7687a9
commit 90979c38f8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/model_executor/parallel_utils/tensor_parallel/layers.py
View File

@ -83,7 +83,7 @@ class VocabParallelEmbedding(torch.nn.Module):
init_method=init.xavier_normal_,
params_dtype: torch.dtype=None,
use_cpu_initialization: bool=False,
perform_initialization: bool=True):
perform_initialization: bool=False):
super(VocabParallelEmbedding, self).__init__()
assert not perform_initialization
assert not use_cpu_initialization
@ -113,7 +113,7 @@ class VocabParallelEmbedding(torch.nn.Module):
self.weight = Parameter(torch.empty(
self.num_embeddings_per_partition, self.embedding_dim,
device=torch.cuda.current_device(), dtype=params_dtype))
def forward(self, input_):
if self.tensor_model_parallel_size > 1:
# Build the mask.
@ -172,7 +172,7 @@ class ColumnParallelLinear(torch.nn.Module):
skip_bias_add=False,
params_dtype=None,
use_cpu_initialization=False,
perform_initialization=True,
perform_initialization=False,
quant_config=None,
):
super(ColumnParallelLinear, self).__init__()
@ -288,7 +288,7 @@ class RowParallelLinear(torch.nn.Module):
skip_bias_add=False,
params_dtype=None,
use_cpu_initialization=False,
perform_initialization=True,
perform_initialization=False,
reduce_results=True,
quant_config=None,
):