[Bugfix] Restore weight_dtype in QuarkW8A8Fp8MoEMethod to fix GPT-OSS FP8 MoE weight loading (#54824)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
This commit is contained in:
@@ -243,6 +243,7 @@ class QuarkW8A8Fp8MoEMethod(QuarkMoEMethod):
|
|||||||
activation_quant_key: QuantKey | None,
|
activation_quant_key: QuantKey | None,
|
||||||
):
|
):
|
||||||
super().__init__(moe, weight_quant_key, activation_quant_key)
|
super().__init__(moe, weight_quant_key, activation_quant_key)
|
||||||
|
self.weight_dtype = "fp8"
|
||||||
self.weight_qscheme = (
|
self.weight_qscheme = (
|
||||||
"per_channel" if weight_quant_key == kFp8StaticChannelSym else "per_tensor"
|
"per_channel" if weight_quant_key == kFp8StaticChannelSym else "per_tensor"
|
||||||
)
|
)
|
||||||
|
|||||||
Reference in New Issue
Block a user