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, 09:30 UTC
[Model] [Config] Correctly identify granite-4.0-micro as non-hybrid model (#28563)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Thomas Parnell committed
Nov 12, 2025 at 19:17 UTC
64d57c3be7443137a6a117cf7f3fb9bd625a749f
1 file changed
+7
vllm/config/model.py
+7
@@ -1619,6 +1619,13 @@ class ModelConfig:
1619
1620
@property
1621
def is_hybrid(self) -> bool:
1622
+ # Handle granite-4.0-micro case which uses hybrid config but does not
1623
+ # actually contain any non-attention layers.
1624
+ layer_types = getattr(self.hf_config, "layer_types", None)
1625
+ if layer_types is not None and all(
1626
+ layer == "attention" for layer in layer_types
1627
+ ):
1628
+ return False
1629
return self._model_info.is_hybrid
1630
1631
@property