Problem on executing main_train

[772435284]

# 循环轮次数目
nums_round = 1
for i in range(1):
    # train
    main_train(logger, tokenizer, model, to_be_trained_entities, yanbao_texts)        
    
    model = model.to(DEVICE)
    model_predict_entities = test(model, test_dataloader, logger=logger, device=DEVICE)
    
    # 修复训练预测结果
    reviewed_entities = review_model_predict_entities(model_predict_entities)
    
    # 将训练预测结果再次放入训练集中， 重新训练或者直接出结果
    for ent_type, ents in reviewed_entities.items():
        to_be_trained_entities[ent_type] = list(set(to_be_trained_entities[ent_type] + ents))

# 创造出提交结果
submit_entities = extract_entities(to_be_trained_entities)

---

RuntimeError Traceback (most recent call last) in 3 for i in range(1): 4 # train ----> 5 main_train(logger, tokenizer, model, to_be_trained_entities, yanbao_texts) 6 7 model = model.to(DEVICE)

in main_train(logger, tokenizer, model, to_be_trained_entities, yanbao_texts) 24 param.requires_grad = False 25 else: ---> 26 param.requires_grad = True 27 optimizer = torch.optim.Adam([para for para in model.parameters() if para.requires_grad], 28 lr=0.001,

RuntimeError: you can only change requires_grad flags of leaf variables.

没有改动过源代码，一直卡在这个地方不知道怎么修

[wgwang]

可能是torch的版本问题：

torch==1.4.0 TorchCRF==1.0.4 torchvision==0.5.0

另外，可以试着运行提供的docker，docker内提供了完整的环境： nvidia-docker run --runtime=nvidia --name ccksenviron -p 12312:12312 -d wgwang/ccks2020-baseline:V1 然后运行下面命令进入docker内： docker exec -it ccksenviron /bin/bash

或者通过12312端口直接访问jupyter