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
Fix lora tests failure in TPU CI due to the removal of LoRA bias (#26723)
Signed-off-by: Xiongfei Wei <isaacwxf23@gmail.com>
XiongfeiWei committed
Oct 13, 2025 at 22:04 UTC
4497c8f82114ef82bc3dd903b790e27c29b22de1
1 file changed
+1
-2
vllm/v1/worker/tpu_model_runner.py
+1
-2
@@ -2128,12 +2128,11 @@ def replace_set_lora(model):
2128
lora_a: torch.Tensor,
2129
lora_b: torch.Tensor,
2130
embeddings_tensor: torch.Tensor | None,
2131
- bias: torch.Tensor | None = None,
2131
):
2132
# TODO: The integer index leads to a recompilation, but converting it
2133
# to a tensor doesn't seem to work anymore. This might be fixed with a
2134
# later release of torch_xla.
2136
- self._original_set_lora(index, lora_a, lora_b, embeddings_tensor, bias)
2135
+ self._original_set_lora(index, lora_a, lora_b, embeddings_tensor)
2136
torch_xla.sync(wait=False)
2137
2138
def _tpu_reset_lora(self, index: int):