michaelxu1107

> > 你好， 由于我现在暂时沒有计算资源，无法支撑模型训练， 但我想看一下这个模型能达到的效果。因为我看网上中文多篇章MRC的代码实现较少，若是您的模型效果比较理想，希望能借鉴下您的框架。请问您方便提供您已经训练好的模型参数文件让我稍微尝试一下么？谢谢。 > > 感谢关注。项目近期会迁移到 tf2.x 版本，同时会发布模型文件，敬请期待。 你好，有tf2.x版本项目代码吗？

hello, thanks for reply, but when I use script show above to convert model, it shows error below $ python convert_bert_from_uer_to_huggingface.py --input_model_path "./sdcup_base_model.bin-50000" --output_model_path "./pytorch_model.bin" Traceback (most recent call last):...

这个问题后面有解决么，我这边使用的langchain版本是0.0.193，使用的是单个tool问答，也碰到了 类似File "E:\workspace_shgl\BMTools-main\bmtools\agent\executor.py", line 99, in call self.callback_manager.on_chain_start( AttributeError: 'Executor' object has no attribute 'callback_manager'的问题

''' Traceback (most recent call last): File "/root/venv/lib/python3.10/site-packages/gradio/routes.py", line 408, in run_predict output = await app.get_blocks().process_api( File "/root/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1315, in process_api result = await self.call_function( File "/root/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1059,...

针对python 3.6 types.CodeType该如何适配呢 目前的做法是修改write_pyc函数，把 s += pack32(0) + pack32(self.timestamp) + pack32(self.source_size) 修改为 s += pack32(self.timestamp) + pack32(self.source_size)可以支持python3.6 但是当处理大文件时，执行到_gen_obs37_opcode_from_offset函数return bytes((opcode.opmap['JUMP_ABSOLUTE'], offset + 4)) + os.urandom(2)语句时会报错，原因为bytes只支持（0,255），这个该如何解决呢？

hello, do you have any solution for convert predicted vector back to natural language?

> They are the index of the vocabulary i.e. ivocab. Simply get the value of that key. > ivocab[pred[0]], which should give the answer in natural language. thanks

another question, i have trained model, but i find that the model only support answers which are one word, how to optimize model to support answers which are more than...