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 13, 08:15 UTC
Support all interleaved layer types (#28485)
Signed-off-by: Yong Hoon Shin <yhshin@meta.com>
Yong Hoon Shin committed
Nov 12, 2025 at 22:57 UTC
11ac9ddd037c63a8c9404cd1f62f9f81a5f38652
1 file changed
+1
-2
vllm/transformers_utils/config.py
+1
-2
@@ -472,8 +472,7 @@ def is_interleaved(config: PretrainedConfig) -> bool:
472
"""
473
text_config = config.get_text_config()
474
if layer_types := getattr(text_config, "layer_types", None):
475
- interleaved_types = {"full_attention", "sliding_attention"}
476
- return interleaved_types.issubset(layer_types)
475
+ return len(set(layer_types)) > 1
476
return False
477
478