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

[Attention] Clean up iRoPE in V1 (#21188)

Browse Source 
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
Lucas Wilkinson
2025-07-21 12:10:30 -04:00
committed by GitHub
parent 6ece16c4fe
commit 304dce7ec0
9 changed files with 14 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/attention/backends/triton_attn.py
View File

@@ -72,9 +72,6 @@ class TritonAttentionMetadataBuilder(
vllm_config.parallel_config)
self.headdim = model_config.get_head_size()
self.attention_chunk_size = getattr(vllm_config.scheduler_config,
'attention_chunk_size', None)
def build_for_cudagraph_capture(
self, common_attn_metadata: CommonAttentionMetadata
) -> TritonAttentionMetadata:
@@ -208,7 +205,6 @@ class TritonAttentionImpl(AttentionImpl):
logits_soft_cap: Optional[float] = None,
attn_type: AttentionType = AttentionType.DECODER,
kv_sharing_target_layer_name: Optional[int] = None,
use_irope: bool = False,
) -> None:
self.num_heads = num_heads
self.head_size = head_size
@@ -228,8 +224,6 @@ class TritonAttentionImpl(AttentionImpl):
self.logits_soft_cap = logits_soft_cap
self.kv_sharing_target_layer_name = kv_sharing_target_layer_name
self.use_irope = use_irope
self.num_queries_per_kv = self.num_heads // self.num_kv_heads
TritonAttentionBackend.validate_head_size(head_size)