ParallelDecoderBlock: Fix regression

turboderp · turboderp · commit 232cc1d8ea22 · 2025-11-16T14:27:56.000+01:00
diff --git a/exllamav3/modules/transformer.py b/exllamav3/modules/transformer.py
@@ -245,6 +245,7 @@ def allocate_q(self, quant_args: dict, surplus_bits: int):
             self.mlp.gates if any(isinstance(self.mlp, x) for x in [GatedMLP, BlockSparseMLP]) else None,
             self.mlp.ups if self.mlp else None,
             self.mlp.downs if self.mlp else None,
+            None
         )
 
 

Original file line number	Diff line number	Diff line change
`@@ -245,6 +245,7 @@ def allocate_q(self, quant_args: dict, surplus_bits: int):`
`245`	`245`	`self.mlp.gates if any(isinstance(self.mlp, x) for x in [GatedMLP, BlockSparseMLP]) else None,`
`246`	`246`	`self.mlp.ups if self.mlp else None,`
`247`	`247`	`self.mlp.downs if self.mlp else None,`
	`248`	`+ None`
`248`	`249`	`)`
`249`	`250`
`250`	`251`