PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
mllm eval bug fix (#297)
n1ck-guo0bb70a64f684
diff --git a/auto_round/mllm/eval.py b/auto_round/mllm/eval.pyindex 8506b089f..ec1b2b118 100644--- a/auto_round/mllm/eval.py+++ b/auto_round/mllm/eval.py@@ -39,7 +39,7 @@ MODEL_TYPE_TO_VLMEVAL_MODEL = {#model_name"Qwen-VL": dict(cls="QwenVL"),"Qwen-VL-Chat": dict(cls="QwenVLChat"),- "Qwen2-VL": dict(cls="Qwen2VLChat", min_pixels=1280*28*28, max_pixels=16384*28*28),+ "Qwen2-VL": dict(cls="Qwen2VLChat", min_pixels=1280*28*28, max_pixels=16384*28*28, verbose=False),"Llama-3.2": dict(cls="llama_vision"),"Phi-3-vision": dict(cls="Phi3Vision"),"Phi-3.5-vision": dict(cls="Phi3_5Vision"),@@ -112,7 +112,9 @@ def mllm_eval(kwargs["model_path"] = pretrained_model_name_or_pathmodel_cls = kwargs.pop("cls")model_cls = getattr(vlmeval.vlm, model_cls)- vlmeval.config.supported_VLM[model_name] = partial(model_cls, verbose=verbose, **kwargs)+ if "verbose" in kwargs:+ kwargs["verbose"] = verbose+ vlmeval.config.supported_VLM[model_name] = partial(model_cls, **kwargs)pred_root = os.path.join(work_dir, model_name)os.makedirs(pred_root, exist_ok=True)diff --git a/auto_round/script/mllm.py b/auto_round/script/mllm.pyindex 6a79adcd3..3631d867e 100644--- a/auto_round/script/mllm.py+++ b/auto_round/script/mllm.py@@ -47,7 +47,7 @@ class BasicArgumentParser(argparse.ArgumentParser):self.add_argument("--asym", action='store_true',help=" asym quantization")- self.add_argument("--dataset", required=True, type=str,+ self.add_argument("--dataset", type=str, default=None,help="The dataset for quantization training. It can be a custom one.")self.add_argument("--lr", default=None, type=float,@@ -201,6 +201,7 @@ def tune(args):model_name = model_name[:-1]logger.info(f"start to quantize {model_name}")+ assert args.dataset is not None, "dataset should not be None."device_str = detect_device(args.device)torch_dtype = "auto"