mirror of
https://github.com/vladmandic/automatic
synced 2026-09-20 01:31:13 +02:00
SDNQ don't divide group_size
This commit is contained in:
@@ -271,8 +271,6 @@ def sdnq_quantize_layer_weight(weight, layer_class_name=None, weights_dtype="int
|
||||
group_size = 2 ** ((2 if svd_up is None else 3) + dtype_dict[weights_dtype]["num_bits"])
|
||||
else:
|
||||
group_size = 2 ** ((1 if svd_up is None else 2) + dtype_dict[weights_dtype]["num_bits"])
|
||||
elif group_size != -1 and not is_linear_type:
|
||||
group_size = max(group_size // 2, 1)
|
||||
|
||||
if group_size > 0:
|
||||
if group_size >= channel_size:
|
||||
|
||||
Reference in New Issue
Block a user