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 14, 05:30 UTC
[Docs] Fix griffe warnings in vllm/lora/ops (#25369)
Signed-off-by: windsonsea <haifeng.yao@daocloud.io>
Michael Yao committed
Sep 23, 2025 at 11:42 UTC
fafbe11af4f211abe02ef874f113bc28232eb0e8
2 files changed
+3
-4
vllm/lora/ops/triton_ops/lora_kernel_metadata.py
+3
-3
@@ -83,8 +83,8 @@ class LoRAKernelMeta:
83
Prepare kernel metadata tensors for the current forward pass.
84
85
Args:
86
- token_lora_tensor (torch.Tensor): Tensor containing lora indices
87
- for each input token.
86
+ token_lora_mapping (torch.Tensor): Tensor containing lora indices
87
+ for each input token.
88
"""
89
90
self._reset()
@@ -136,7 +136,7 @@ class LoRAKernelMeta:
136
137
Args:
138
token_nums (int): Number of input tokens in the current forward
139
- pass.
139
+ pass of the kernel.
140
"""
141
return (
142
self.token_lora_mapping[:token_nums],
vllm/lora/ops/xla_ops/lora_ops.py
-1
@@ -93,7 +93,6 @@ def bgmv_shrink(
93
inputs (torch.Tensor): Input tensor of shape [num_tokens, hidden_size].
94
lora_b_weights (torch.Tensor): LoRA weights of shape
95
[num_loras, lora_rank, hidden_size].
96
- output_tensor (torch.Tensor): (Unused) output tensor (placeholder).
96
lora_indices_tensor (torch.Tensor): Tensor of shape [num_tokens]
97
indicating which LoRA matrix to use for each token.
98
scaling (float, optional): Scalar multiplier applied to the output.