[Bugfix] Detect all attention-spelling variants in ModelConfig.is_hybrid (#52161)
Signed-off-by: Michal Ganczarenko <michal.ganczarenko@intel.com> Signed-off-by: Michał Ganczarenko <michal.ganczarenko@intel.com> Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Michał Ganczarenko committed
Aug 18, 2026 at 20:43 UTC
6948a43fbbf427e69ca0d325cd7ab1daee8d131b
1 file changed
+1
-1
vllm/config/model.py
+1
-1
@@ -1850,7 +1850,7 @@ class ModelConfig:
1850
# actually contain any non-attention layers.
1851
layer_types = getattr(self.hf_config, "layer_types", None)
1852
return layer_types is None or not all(
1853
- layer == "attention" for layer in layer_types
1853
+ layer in ("attention", "full_attention") for layer in layer_types
1854
)
1855
1856
@property