liminfei-amd 516f3ee922 model_patcher: filter quantized weight state keys precisely ... Collect the state-dict pieces emitted by each QuantizedTensor weight instead of filtering globally by suffix. This preserves real module parameters such as input_scale, keeps shared module aliases working, and leaves the real weight dequantization path intact. Signed-off-by: liminfei-amd <91481003+liminfei-amd@users.noreply.github.com>		2026-07-12 20:36:30 +08:00
..
test_mixed_precision.py	model_patcher: filter quantized weight state keys precisely	2026-07-12 20:36:30 +08:00