A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 09:45 UTC
[Doc] fix arg docstring in linear layers (#18410)
Signed-off-by: giantcroc <1204449533@qq.com>
GiantCroc committed
May 21, 2025 at 21:45 UTC
c154d89306bf9978ff013659c5b1244af78386cc
1 file changed
+9
vllm/model_executor/layers/linear.py
+9
@@ -261,6 +261,7 @@ class ReplicatedLinear(LinearBase):
261
quant_config: Quantization configure.
262
prefix: The name of the layer in the state dict, including all parents
263
(e.g. model.layers.0.qkv_proj)
264
+ return_bias: If true, return bias together with outputs in forward pass.
265
"""
266
267
def __init__(
@@ -523,6 +524,7 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
524
quant_config: Quantization configure.
525
prefix: The name of the layer in the state dict, including all parents
526
(e.g. model.layers.0.qkv_proj)
527
+ return_bias: If true, return bias together with outputs in forward pass.
528
"""
529
530
def __init__(
@@ -805,6 +807,7 @@ class QKVParallelLinear(ColumnParallelLinear):
807
quant_config: Quantization configure.
808
prefix: The name of the layer in the state dict, including all parents
809
(e.g. model.layers.0.qkv_proj)
810
+ return_bias: If true, return bias together with outputs in forward pass.
811
"""
812
813
def __init__(
@@ -1155,7 +1158,13 @@ class RowParallelLinear(LinearBase):
1158
bias can be fused with other element-wise operations.
1159
We skip adding bias but instead return it.
1160
params_dtype: Data type for the parameters.
1161
+ reduce_results: If true, call all-reduce on output and make Y available
1162
+ to all GPUs, otherwise, every GPU will have its output
1163
+ which is Y = X_iA_i
1164
quant_config: Quantization configure.
1165
+ prefix: The name of the layer in the state dict, including all parents
1166
+ (e.g. model.layers.0.down_proj)
1167
+ return_bias: If true, return bias together with outputs in forward pass.
1168
"""
1169
1170
def __init__(