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

Support quant for `meituan-longcat/LongCat-Flash-Lite` (#1388)

Tang Kaihuia12e3f5a575a
diff --git a/auto_round/utils/model.py b/auto_round/utils/model.py
index fe416cd97..94fb7198b 100644
--- a/auto_round/utils/model.py
+++ b/auto_round/utils/model.py
@@ -625,6 +625,8 @@ def get_block_names(model, quant_vision=False):
return [(name, module)]
target_modules = []
for n, m in module.named_children():
+ if hasattr(type(m), "__name__") and "NgramEmbedding" in type(m).__name__:
+ continue
if hasattr(type(m), "__name__") and "ModuleList" in type(m).__name__:
target_modules.append((".".join(filter(None, (name, n))), m))
else: