diff --git a/fastdeploy/model_executor/layers/linear.py b/fastdeploy/model_executor/layers/linear.py
index 2c8e9fe33..d17ab1be3 100644
--- a/fastdeploy/model_executor/layers/linear.py
+++ b/fastdeploy/model_executor/layers/linear.py
@@ -498,6 +498,7 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
             if weight_need_transpose:
                 loaded_weight = get_tensor(loaded_weight)
                 loaded_weight = loaded_weight.transpose([1, 0])
+                # Avoid redundant transpose of fused weights when weight_loader is called iteratively
                 param.weight_need_transpose = False
             # Loaded weight is already fused on disk.
             shard_offsets = [
@@ -638,6 +639,7 @@ class QKVParallelLinear(ColumnParallelLinear):
             if weight_need_transpose:
                 loaded_weight = get_tensor(loaded_weight)
                 loaded_weight = loaded_weight.transpose([1, 0])
+                # Avoid redundant transpose of fused weights when weight_loader is called iteratively
                 param.weight_need_transpose = False
             # Loaded weight is already fused on disk
             shard_offsets = [
diff --git a/fastdeploy/model_executor/layers/quantization/__init__.py b/fastdeploy/model_executor/layers/quantization/__init__.py
index 58f610ebf..6be178282 100644
--- a/fastdeploy/model_executor/layers/quantization/__init__.py
+++ b/fastdeploy/model_executor/layers/quantization/__init__.py
@@ -16,6 +16,8 @@ quantization module
 """
 from typing import Dict, List, Type
 
+from fastdeploy.utils import parse_quantization
+
 from .quant_base import QuantConfigBase
 
 QUANTIZATION_METHODS: List[str] = [
@@ -35,6 +37,8 @@ QUANTIZATION_METHODS: List[str] = [
 
 
 def parse_quant_config(args, model_config, is_ernie, is_v1_loader):
+    if args.quantization is not None and isinstance(args.quantization, str):
+        args.quantization = parse_quantization(args.quantization)
     # 1.model_config.is_quantized
     # TODO(bukejiyu)  model_config.is_quantized is v0 only need to be removed in future
     if model_config.model_format == "torch":
diff --git a/fastdeploy/model_executor/utils.py b/fastdeploy/model_executor/utils.py
index be66a5f27..0b226bf7b 100644
--- a/fastdeploy/model_executor/utils.py
+++ b/fastdeploy/model_executor/utils.py
@@ -240,7 +240,7 @@ def rename_offline_ckpt_suffix_to_fd_suffix(
     }
     moe_quant_type = ""
     dense_quant_type = ""
-    if fd_config.quant_config is None:
+    if fd_config.quant_config is not None:
         if fd_config.quant_config.name() == "mix_quant":
             moe_quant_type = fd_config.quant_config.moe_quant_type
             dense_quant_type = fd_config.quant_config.dense_quant_type
diff --git a/fastdeploy/worker/worker_process.py b/fastdeploy/worker/worker_process.py
index 4ef043f12..3cf6fe928 100644
--- a/fastdeploy/worker/worker_process.py
+++ b/fastdeploy/worker/worker_process.py
@@ -44,7 +44,7 @@ from fastdeploy.inter_communicator import EngineWorkerQueue as TaskQueue
 from fastdeploy.inter_communicator import IPCSignal
 from fastdeploy.model_executor.layers.quantization import parse_quant_config
 from fastdeploy.platforms import current_platform
-from fastdeploy.utils import get_logger, parse_quantization
+from fastdeploy.utils import get_logger
 from fastdeploy.worker.worker_base import WorkerBase
 
 logger = get_logger("worker_process", "worker_process.log")
@@ -655,8 +655,6 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
         FDConfig: Initialized FastDeploy configuration object
     """
     # RL rollout
-    if args.quantization is not None and isinstance(args.quantization, str):
-        args.quantization = parse_quantization(args.quantization)
     paddle.set_default_dtype(args.dtype)
     model_config = ModelConfig(vars(args))
     device_config = DeviceConfig(vars(args))