Add zh_dureader_model_v2 && add train function (#31)

* 1. add zh_dureader_v2; 2. add train function

* update readme and example

* add train config

* Update README.md

* fix readme typo

* Fixed wrong config path, added train data format note

* fix zh_dureader models http address

* faiss example fit predict interator

* add dureader.para

Co-authored-by: sfwydyc <dingyuchen@baidu.com>
Co-authored-by: Xing Yiran <procedure2012@hotmail.com>
Co-authored-by: MrTuo <595983351@qq.com>

README.md

@@ -14,6 +14,8 @@ In recent years, the dense retrievers based on pre-trained language models have
 * ***Easy-to-use***: By integrating this toolkit with [JINA](https://jina.ai/), 🚀RocketQA can help developers build an end-to-end retrieval system and question answering system with several lines of code. <img src="https://github.com/PaddlePaddle/RocketQA/blob/main/RocketQA_flow.png" alt="" align=center />  
 
 ## News
+* April 29, 2022: Training function is added to RocketQA toolkit.
+* April 29, 2022: The baseline models of **DuReader<sub>retrieval</sub>** (both cross encoder and dual encoder) are available in RocketQA models now. You can use them directly by `rocketqa.load_model`.
 * March 30, 2022: The baseline of **DuReader<sub>retrieval</sub>** [leaderboard](https://aistudio.baidu.com/aistudio/competition/detail/157/0/introduction) was released. [[code/model]](https://github.com/PaddlePaddle/RocketQA/tree/main/research/DuReader-Retrieval-Baseline)
 * March 30, 2022: We released **DuReader<sub>retrieval</sub>**, a large-scale Chinese benchmark for passage retrieval. The dataset contains over 90K questions and 8M passages from Baidu Search. [[paper]](https://arxiv.org/abs/2203.10232) [[data]](https://github.com/baidu/DuReader/tree/master/DuReader-Retrieval)
 * December 3, 2021: The toolkit of dense retriever RocketQA was released, including the first chinese dense retrieval model trained on DuReader. 
@@ -117,13 +119,21 @@ Given a list of paragraphs and their corresponding titles (optional), returns th
 
 Given a list of queries and paragraphs (and titles), returns their matching scores (dot product between two representation vectors). 
 
+#### [`model.train(train_set: str, epoch: int, save_model_path: str, args)`](https://github.com/sfwydyc/my-rocketqa/blob/594877ca505053cb67c6b9b689dbbf237f074ac4/rocketqa/encoder/dual_encoder.py#L247)
+
+Given the hyperparameters `train_set`, `epoch` and `save_model_path`, you can train your own dual encoder model or finetune our models. Other settings like `save_steps` and `learning_rate` can also be set in `args`. Please see examples/example.py for detail.
+
 ### Cross encoder
 Cross-encoder returned by "load_model()" supports the following function:
 
 #### [`model.matching(query: List[str], para: List[str], title: List[str])`](https://github.com/PaddlePaddle/RocketQA/blob/3a99cf2720486df8cc54acc0e9ce4cbcee993413/rocketqa/predict/cross_encoder.py#L129)
 
 Given a list of queries and paragraphs (and titles), returns their matching scores (probability that the paragraph is the query's right answer).
 
+#### [`model.train(train_set: str, epoch: int, save_model_path: str, args)`](https://github.com/sfwydyc/my-rocketqa/blob/594877ca505053cb67c6b9b689dbbf237f074ac4/rocketqa/encoder/dual_encoder.py#L247)
+
+Given the hyperparameters `train_set`, `epoch` and `save_model_path`, you can train your own cross encoder model or finetune our models. Other settings like `save_steps` and `learning_rate` can also be set in `args`. Please see examples/example.py for detail.
+
 
 ### Examples
 
@@ -149,7 +159,48 @@ p_embs = dual_encoder.encode_para(para=para_list)
 dot_products = dual_encoder.matching(query=query_list, para=para_list)
 ```
 
+####  Train Your Own Model
+To train your own models, you can use 'train()' function with your dataset and parameters.The training data format refers to ./examples/data/cross.train.tsv, which contains 4 columns: query, title, para, label (0 or 1), separated by "\t"
+
+```python
+import rocketqa
+
+# init cross encoder, and set device and batch_size
+cross_encoder = rocketqa.load_model(model="zh_dureader_ce_v2", use_cuda=True, device_id=0, batch_size=32)
+
+# finetune cross encoder based on "zh_dureader_ce_v2"
+cross_encoder.train('./examples/data/cross.train.tsv', 2, 'ce_models', save_steps=1000, learning_rate=1e-5, log_folder='log_ce')
+
+```
 
+#### Run Your Own Model
+To run your own models, you should set parameter `model` in 'load_model()' with a JSON config file.
+
+```python
+import rocketqa
+
+# init cross encoder
+cross_encoder = rocketqa.load_model(model="./examples/ce_models/config.json", use_cuda=True, device_id=0, batch_size=16)
+
+# compute relevance of query and para
+relevance = cross_encoder.matching(query=query_list, para=para_list)
+```
+
+config is a JSON file like this
+```
+{
+    "model_type": "cross_encoder",
+    "max_seq_len": 384,
+    "model_conf_path": "zh_config.json",
+    "model_vocab_path": "zh_vocab.txt",
+    "model_checkpoint_path": ${YOUR_MODEL},
+    "for_cn": true,
+    "share_parameter": 0
+}
+```
+Folder `examples` provides more details.
+
+
 ## Citations
 
 If you find RocketQA v1 models helpful, feel free to cite our publication [RocketQA: An Optimized Training Approach to Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/pdf/2010.08191.pdf)

examples/ce_models/config.json

@@ -0,0 +1,9 @@
+{
+    "model_type": "cross_encoder",
+    "max_seq_len": 384,
+    "model_conf_path": "zh_config.json",
+    "model_vocab_path": "zh_vocab.txt",
+    "model_checkpoint_path": "step_36000",
+    "for_cn": true,
+    "share_parameter": 0
+}

examples/ce_models/zh_config.json

@@ -0,0 +1,12 @@
+{
+  "attention_probs_dropout_prob": 0.1, 
+  "hidden_act": "relu", 
+  "hidden_dropout_prob": 0.1, 
+  "hidden_size": 768, 
+  "initializer_range": 0.02, 
+  "max_position_embeddings": 513, 
+  "num_attention_heads": 12, 
+  "num_hidden_layers": 12, 
+  "type_vocab_size": 2, 
+  "vocab_size": 18000
+}