qq_16940641: 大佬，请教一下，attention那里的计算，qk的缩放系数一开是缩小了layer_id倍，最后76行又放大了layer_id倍，这一来一回没什么变化啊，那为什么还需要这么操作呢？

Byyyi耀: 大佬互关吗？非常不错的文章，解决了我大问题！赞一个！

qq_38402407: 请问一下，这种该如何训练呢，以及如何构造训练数据，请问有相关资料吗，谢谢

qq_20105917: 在这之前的日志有警告。如下： 2024-06-05 15:35:05,203 - modelscope - WARNING - No preprocessor field found in cfg. 2024-06-05 15:35:05,203 - modelscope - WARNING - No val key and type key found in preprocessor domain of configuration.json file. 2024-06-05 15:35:05,203 - modelscope - WARNING - Cannot find available config to build preprocessor at mode inference, current config: {'model_dir': 'models/iic/nlp_gte_sentence-embedding_chinese-base'}. trying to build by task and model information. 2024-06-05 15:35:05,225 - modelscope - WARNING - No preprocessor field found in cfg. 2024-06-05 15:35:05,225 - modelscope - WARNING - No val key and type key found in preprocessor domain of configuration.json file. 2024-06-05 15:35:05,225 - modelscope - WARNING - Cannot find available config to build preprocessor at mode inference, current config: {'model_dir': 'models/iic/nlp_gte_sentence-embedding_chinese-base', 'sequence_length': 128}. trying to build by task and model information.

qq_20105917: response没有输出。模型Qwen-7b-Chat.最后的日志 /opt/python3.10/python3/lib/python3.10/site-packages/transformers/generation/configuration_utils.py:492: UserWarning: `do_sample` is set to `False`. However, `temperature` is set to `0.0` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `temperature`. warnings.warn( /opt/python3.10/python3/lib/python3.10/site-packages/transformers/generation/configuration_utils.py:497: UserWarning: `do_sample` is set to `False`. However, `top_p` is set to `0.8` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `top_p`. warnings.warn( /opt/python3.10/python3/lib/python3.10/site-packages/transformers/generation/configuration_utils.py:509: UserWarning: `do_sample` is set to `False`. However, `top_k` is set to `0` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `top_k`. warnings.warn(