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

awq exporting bugfix (#456)

WeiweiZhang15ccdb0d119d7
diff --git a/auto_round/export/export_to_awq/export.py b/auto_round/export/export_to_awq/export.py
index 7f3a8d33f..0a6146bea 100644
--- a/auto_round/export/export_to_awq/export.py
+++ b/auto_round/export/export_to_awq/export.py
@@ -50,7 +50,7 @@ def pack_layer(name, model, layer_config, backend, pbar):
scale, zp = linear_layer.scale, linear_layer.zp
scale = scale.t().contiguous()
zp = zp.t().contiguous()
- config["zp"] = config["zp"].to(torch.float32)
+ # config["zp"] = config["zp"].to(torch.float32)
bits = config["bits"]
group_size = config["group_size"]