[Quant] Enable humming w[2-7]a[4,8] inference with compressed-tensors (#46390)

Signed-off-by: HDCharles <charlesdavidhernandez@gmail.com>
This commit is contained in:
HDCharles
2026-07-14 20:22:31 -06:00
committed by GitHub
parent 3ad85e0de4
commit 9b2be4e9a5
8 changed files with 544 additions and 3 deletions
@@ -45,7 +45,10 @@ class HummingLinearKernel(MPLinearKernel):
quant_config["has_zero_point"] = True
convert_linear_layer_to_humming_standard(layer=layer, name_map=name_map)
prepare_humming_layer(layer, quant_config)
input_quant_config = getattr(layer, "_humming_input_quant_config", None)
prepare_humming_layer(
layer, quant_config, input_quant_config=input_quant_config
)
def apply_weights(
self,