PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
Fix unsupported dtype (#1868)
Wenhua Cheng7138be82e6e9
diff --git a/auto_round/compressors/entry.py b/auto_round/compressors/entry.pyindex 73f3f26fe..d001aaf72 100644--- a/auto_round/compressors/entry.py+++ b/auto_round/compressors/entry.py@@ -584,11 +584,11 @@ class AutoRoundCompatible:# Check model type for logging (use warning_once to avoid repeating for every block# when called from LLM-Compressor which instantiates AutoRound per block)if is_mllm_model(model, platform=platform):- logger.warning_once("Using MLLM mode for multimodal model.")+ logger.info("Using MLLM mode for multimodal model.")elif is_diffusion_model(model):- logger.warning_once("Using Diffusion mode for diffusion model.")+ logger.info("Using Diffusion mode for diffusion model.")else:- logger.warning_once("Using LLM mode.")+ logger.info("Using LLM mode.")# Create AutoRound instance using new architecturecompressor = AutoRound(diff --git a/auto_round/data_type/utils.py b/auto_round/data_type/utils.pyindex bd4d74eaa..dbfd44425 100644--- a/auto_round/data_type/utils.py+++ b/auto_round/data_type/utils.py@@ -170,6 +170,9 @@ def get_quant_func(if data_type in QUANT_FUNC_WITH_DTYPE:return QUANT_FUNC_WITH_DTYPE[data_type], data_type+ raise ValueError(+ f"No quantization function found for dtype={dtype}, bits={bits}, sym={sym}, group_size={group_size}"+ )def round_ste(x: torch.Tensor):