Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

move packeding res to cpu (#1104)

Yi Liudaa151d495c3
diff --git a/auto_round/compressors/base.py b/auto_round/compressors/base.py
index a3309bc41..8bf788c24 100644
--- a/auto_round/compressors/base.py
+++ b/auto_round/compressors/base.py
@@ -1332,6 +1332,8 @@ class BaseCompressor(object):
self._immediate_pack(name)
if to_cpu:
m = m.to("cpu")
+ packed_m = get_module(self.model, name)
+ set_module(self.model, name, packed_m.to("cpu"))
else:
if to_cpu:
m = m.to("cpu")