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

Correct docstring (#1652)

Yi Liude0650e10623
diff --git a/auto_round/export/export_to_llmcompressor/export.py b/auto_round/export/export_to_llmcompressor/export.py
index 61685377c..9ffbd4ee7 100644
--- a/auto_round/export/export_to_llmcompressor/export.py
+++ b/auto_round/export/export_to_llmcompressor/export.py
@@ -100,9 +100,10 @@ def _get_quant_format(model):
def _compress_and_set_format(layer, scheme, device=None):
"""Compress a layer and set its quantization format, with backward compatibility.
- Supports both old (<0.9.0) and new (>=0.9.0) compressed_tensors APIs:
+ Supports both old and new compressed_tensors APIs:
- New API: compress_module() handles compression, state dict replacement, status, and format.
- Old API: NaiveQuantizationCompressor instance + manual param management + set_per_module_format.
+ The new API was introduced after compressed_tensors v0.14.1, may be available in v0.15+.
"""
try:
from compressed_tensors.compressors import compress_module # pylint: disable=E0401