[Feature 2.0.2] support top_k_top_p sampling (#2789)

* support top_k_top_p sampling

* fix

* add api param

* add api para

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* change func name

fastdeploy/worker/vl_gpu_model_runner.py

@@ -13,10 +13,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
+import argparse
 import json
 import os
 import random
-import argparse
 
 import numpy as np
 import paddle
@@ -24,6 +24,9 @@ import paddle.distributed.fleet as fleet
 from paddleformers.transformers.model_utils import load_tp_checkpoint
 from safetensors import safe_open
 
+from fastdeploy.config import (DeviceConfig, FDConfig, KVCacheConfig,
+                               LoadConfig, ModelConfig, MoEConfig, MoEPhase,
+                               ParallelConfig, SpeculativeConfig)
 from fastdeploy.input.ernie_tokenizer import ErnieBotTokenizer
 from fastdeploy.input.mm_processor import DataProcessor
 from fastdeploy.model_executor.layers.attention import get_attention_backend
@@ -44,9 +47,6 @@ from fastdeploy.platforms import current_platform
 from fastdeploy.worker.forward_meta import ForwardMeta
 from fastdeploy.worker.utils import check_safetensors_model
 from fastdeploy.worker.vl_model_runner_base import VLModelRunnerBase
-from fastdeploy.config import (DeviceConfig, FDConfig, KVCacheConfig,
-                                LoadConfig, ModelConfig, MoEConfig,
-                                MoEPhase, ParallelConfig, SpeculativeConfig)
 
 if current_platform.is_cuda() and current_platform.available():
     from fastdeploy.model_executor.layers.utils import (