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, 06:00 UTC
[BugFix][FlashInfer] Fix attention backend interface mismatch with unexpected keyword `use_irope` (#19134)
Signed-off-by: Yunqiu Guo <guorachel@meta.com>
Rachel Guo committed
Jun 10, 2025 at 01:48 UTC
467bef18a353e8ca133a0a5611d6f4384ea351b6
1 file changed
+5
vllm/v1/attention/backends/flashinfer.py
+5
@@ -508,7 +508,12 @@ class FlashInferImpl(AttentionImpl):
508
logits_soft_cap: Optional[float] = None,
509
attn_type: AttentionType = AttentionType.DECODER,
510
kv_sharing_target_layer_name: Optional[int] = None,
511
+ use_irope: bool = False,
512
) -> None:
513
+ if use_irope:
514
+ logger.warning_once(
515
+ "Using irope in FlashInfer is not supported yet, it will fall"
516
+ " back to global attention for long context.")
517
self.num_heads = num_heads
518
self.head_size = head_size
519
self.scale = float(scale)