[ce tests] Add trt mode and amp in ce test (#4174)

* [ce tests] add trt_mode in ppyolo

* [ce tests] set amp in tests.sh

tests/ppdet_params/ppyolo_mbv3_large_coco_params.txt

@@ -3,7 +3,7 @@ model_name:ppyolo_mbv3_large_coco
 python:python3.7
 gpu_list:0
 use_gpu:True
-auto_cast:null
+auto_cast:null|amp
 epoch:lite_train_infer=1|whole_train_infer=405
 save_dir:./output/
 TrainReader.batch_size:lite_train_infer=2|whole_train_infer=24
@@ -43,7 +43,7 @@ inference:deploy/python/infer.py
 --cpu_threads:1|4
 --batch_size:1|2
 --use_tensorrt:null
---run_mode:fluid
+--run_mode:fluid|trt_fp32|trt_fp16|trt_int8
 --model_dir:
 --image_dir:./dataset/coco/val2017/
 --save_log_path:null

tests/ppdet_params/ppyolo_r50vd_dcn_1x_coco_params.txt

@@ -3,7 +3,7 @@ model_name:ppyolo_r50vd_dcn_1x_coco
 python:python3.7
 gpu_list:0
 use_gpu:True
-auto_cast:null
+auto_cast:null|amp
 epoch:lite_train_infer=1|whole_train_infer=405
 save_dir:./output/
 TrainReader.batch_size:lite_train_infer=2|whole_train_infer=24
@@ -43,7 +43,7 @@ inference:deploy/python/infer.py
 --cpu_threads:1|4
 --batch_size:1|2
 --use_tensorrt:null
---run_mode:fluid
+--run_mode:fluid|trt_fp32|trt_fp16|trt_int8
 --model_dir:
 --image_dir:./dataset/coco/val2017/
 --save_log_path:null

tests/ppdet_params/ppyolov2_r50vd_dcn_365e_coco_params.txt

@@ -3,7 +3,7 @@ model_name:ppyolov2_r50vd_dcn_365e_coco
 python:python3.7
 gpu_list:0
 use_gpu:True
-auto_cast:null
+auto_cast:null|amp
 epoch:lite_train_infer=1|whole_train_infer=365
 save_dir:./output/
 TrainReader.batch_size:lite_train_infer=2|whole_train_infer=12

tests/ppdet_params/yolov3_darknet53_270e_coco_params.txt

@@ -3,7 +3,7 @@ model_name:yolov3_darknet53_270e_coco
 python:python3.7
 gpu_list:0
 use_gpu:True
-auto_cast:null
+auto_cast:null|amp
 epoch:lite_train_infer=1|whole_train_infer=270
 save_dir:./output/
 TrainReader.batch_size:lite_train_infer=2|whole_train_infer=8

tests/test.sh

@@ -304,7 +304,11 @@ else
                     continue
                 fi
 
+                if [ ${autocast} = "amp" ]; then
+                    set_autocast="--fp16"
+                else
                     set_autocast=$(func_set_params "${autocast_key}" "${autocast}")
+                fi
                 set_epoch=$(func_set_params "${epoch_key}" "${epoch_num}")
                 set_pretrain=$(func_set_params "${pretrain_model_key}" "${pretrain_model_value}")
                 set_batchsize=$(func_set_params "${train_batch_key}" "${train_batch_value}")
@@ -319,11 +323,11 @@ else
 
                 set_save_model=$(func_set_params "${save_model_key}" "${save_log}")
                 if [ ${#gpu} -le 2 ];then  # train with cpu or single gpu
-                    cmd="${python} ${run_train} ${set_use_gpu}  ${set_save_model} ${set_epoch} ${set_pretrain} ${set_autocast} ${set_batchsize} ${set_train_params1} "
+                    cmd="${python} ${run_train} ${set_use_gpu}  ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_train_params1} ${set_autocast} "
                 elif [ ${#gpu} -le 15 ];then  # train with multi-gpu
-                    cmd="${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_autocast} ${set_batchsize} ${set_train_params1}"
+                    cmd="${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} ${set_save_model} ${set_epoch} ${set_pretrain} ${set_batchsize} ${set_train_params1} ${set_autocast}"
                 else     # train with multi-machine
-                    cmd="${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} ${set_save_model} ${set_pretrain} ${set_epoch} ${set_autocast} ${set_batchsize} ${set_train_params1}"
+                    cmd="${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} ${set_save_model} ${set_pretrain} ${set_epoch} ${set_batchsize} ${set_train_params1} ${set_autocast}"
                 fi
                 # run train
                 eval "unset CUDA_VISIBLE_DEVICES"