PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
fix log error (#843)
Heng Guo4b4182607760
diff --git a/auto_round/compressors/base.py b/auto_round/compressors/base.pyindex abfc47366..be42b824a 100644--- a/auto_round/compressors/base.py+++ b/auto_round/compressors/base.py@@ -823,7 +823,7 @@ class BaseCompressor(object):if isinstance(self.scheme, str) and self.scheme.lower().startswith("gguf"):for i in range(len(formats)):- if formats[i] != "fake" and formats[i] != self.scheme.lower().startswith("gguf"):+ if formats[i] != "fake" and formats[i] != self.scheme.lower():logger.warning(f"reset format {formats[i]} to {self.scheme.lower()} "f"since scheme {self.scheme} can only be exported to format {self.scheme.lower()}"diff --git a/auto_round/utils.py b/auto_round/utils.pyindex 80bea06c4..004bfd9cc 100644--- a/auto_round/utils.py+++ b/auto_round/utils.py@@ -1385,7 +1385,6 @@ def check_and_mark_fp8_model(model: torch.nn.Module) -> bool:if _is_fp8_linear(m):m.is_fp8_linear = Trueif not hasattr(model, "is_fp8"):- logger.warning("the support for fp8 model as input is experimental, please use with caution.")model.is_fp8 = Trueif hasattr(model, "is_fp8"):return True