[FIX] fix mambaish model pp kv cache compute (#17334)
This commit is contained in:
@@ -124,7 +124,12 @@ class ModelRunnerKVCacheMixin:
|
|||||||
self.num_effective_layers,
|
self.num_effective_layers,
|
||||||
)
|
)
|
||||||
elif mambaish := self.mambaish_config:
|
elif mambaish := self.mambaish_config:
|
||||||
num_layers = len(mambaish.full_attention_layer_ids)
|
effective_layer_ids = [
|
||||||
|
i
|
||||||
|
for i in mambaish.full_attention_layer_ids
|
||||||
|
if self.start_layer <= i < self.end_layer
|
||||||
|
]
|
||||||
|
num_layers = len(effective_layer_ids)
|
||||||
else:
|
else:
|
||||||
num_layers = self.num_effective_layers
|
num_layers = self.num_effective_layers
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user