expand training and shortgpt_prune code to support more model

Author: Li-Z-Q

slm/pipelines/examples/contrastive_training/README.md

@@ -18,10 +18,11 @@ pip install -r slm/pipelines/examples/contrastive_training/requirements.txt
 ```
 
 
-下载 DuReader-Retrieval 中文数据集：
+下载 DuReader-Retrieval 和 MMarco-Retrieval 中文数据集：
 ```
 cd data
 wget https://paddlenlp.bj.bcebos.com/datasets/dureader_dual.train.jsonl
+python download_mmarco.py
 ```
 
 ## 训练

slm/pipelines/examples/contrastive_training/data/download_mmarco.py

@@ -22,9 +22,7 @@
 print(len(dataset["train"]))
 
 
-fw = open(
-    "/141nfs/lizhuoqun/PaddleNLP_1022/PaddleNLP/slm/pipelines/examples/contrastive_training/data/mmarco.jsonl", "w"
-)
+fw = open("./mmarco.jsonl", "w")
 
 i = 0
 for data in tqdm.tqdm(dataset["train"]):