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

fix awq exporting bugs (#365)

Wenhua Chengd080ea0af5f5
diff --git a/README.md b/README.md
index 8ee725206..e9afcffa5 100644
--- a/README.md
+++ b/README.md
@@ -296,7 +296,7 @@ in [Gaudi Guide](https://docs.habana.ai/en/latest/).
from transformers import AutoModelForCausalLM, AutoTokenizer
from auto_round import AutoRoundConfig
-backend = "auto" ##cpu, hpu, cuda, cuda:marlin(supported in auto_round>0.3.1 and 'pip install -v gptqmodel --no-build-isolation')
+backend = "auto" ##cpu, hpu, cuda
quantization_config = AutoRoundConfig(
backend=backend
)
diff --git a/auto_round/export/export_to_awq/export.py b/auto_round/export/export_to_awq/export.py
index b82297c60..1921de446 100644
--- a/auto_round/export/export_to_awq/export.py
+++ b/auto_round/export/export_to_awq/export.py
@@ -113,8 +113,8 @@ def save_quantized_as_autoawq(output_dir, inplace=True, **kwargs):
quantization_config["modules_to_not_convert"] = modules_to_not_convert
##check module quantized in block, this may have bug for mixed precision quantization
- if hasattr(model, "config"):
- model.config.quantization_config = quantization_config
+ if hasattr(compressed_model, "config"):
+ compressed_model.config.quantization_config = quantization_config
save(compressed_model, output_dir, safe_serialization=True)
return compressed_model
diff --git a/auto_round/utils.py b/auto_round/utils.py
index cce826468..fc4189ef3 100644
--- a/auto_round/utils.py
+++ b/auto_round/utils.py
@@ -1074,7 +1074,7 @@ def set_layer_config_by_fp_layers(model, fp_layers):
if fp_layer in all_layer_names:
not_to_quantized_layers.append(fp_layer)
continue
- if fp_layer[-1] != ".":
+ if fp_layer[-1].isdigit():
fp_layer = fp_layer + "." ##ticky setting
for name in all_layer_names:
if fp_layer in name: