biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] Update compressed tensors lifecycle to remove prefix from create_weights (#7825)

Browse Source
This commit is contained in:
Dipika Sikka
2024-08-26 20:09:34 -04:00
committed by GitHub
parent 760e9f71a8
commit 015e6cc252
4 changed files with 17 additions and 75 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/layers/linear.py
View File

@@ -208,8 +208,7 @@ class ReplicatedLinear(LinearBase):
self.input_size,
self.output_size,
self.params_dtype,
weight_loader=self.weight_loader,
prefix=prefix)
weight_loader=self.weight_loader)
if bias:
self.bias = Parameter(
@@ -307,8 +306,7 @@ class ColumnParallelLinear(LinearBase):
params_dtype=self.params_dtype,
weight_loader=(
self.weight_loader_v2 if self.quant_method.__class__.__name__
in WEIGHT_LOADER_V2_SUPPORTED else self.weight_loader),
prefix=prefix)
in WEIGHT_LOADER_V2_SUPPORTED else self.weight_loader))
if bias:
self.bias = Parameter(
torch.empty(self.output_size_per_partition,
@@ -976,8 +974,7 @@ class RowParallelLinear(LinearBase):
params_dtype=self.params_dtype,
weight_loader=(
self.weight_loader_v2 if self.quant_method.__class__.__name__
in WEIGHT_LOADER_V2_SUPPORTED else self.weight_loader),
prefix=prefix)
in WEIGHT_LOADER_V2_SUPPORTED else self.weight_loader))
if not reduce_results and (bias and not skip_bias_add):
raise ValueError("When not reduce the results, adding bias to the "
"results can lead to incorrect results")