x-plug / chatplug Goto Github PK

As shown in the figure, use XDPX-based one-click inference

我理解，这里在处理历史对话数据的时候，需要先对"\t"字符做替换

麻烦确认一下这里是不是有些问题~感谢

数据开源或者数据获取方式？

File "/home/xxx/ChatPLUG/XDPX/scripts/fidchat_new.py", line 21, in
from xdpx.utils.chat.openweb_search import Snippet
ModuleNotFoundError: No module named 'xdpx.utils.chat.openweb_search

File "/home/xxx/ChatPLUG/XDPX/xdpx/utils/chat/pipeline.py", line 26, in
from xdpx.utils.chat.openkg_retrieval import OpenKG
ModuleNotFoundError: No module named 'xdpx.utils.chat.openkg_retrieval'
或者是我少装了什么包？

• 用了哪些通用指令数据？比例是多少？
• 微调的对话数据是多少？
• 哪些数据导致了模型的对话的personalize上效果更好？
• 不使用FiD是否更好？现在LLM的长度应该是够的。

请问MacOS 13.3.1 M1 Max 是不是不支持？好像是找不到GPU

2023-04-20 14:48:06,856 - modelscope - INFO - PyTorch version 2.1.0.dev20230414 Found.
2023-04-20 14:48:06,856 - modelscope - INFO - Loading ast index from /Users/lewaylee/.cache/modelscope/ast_indexer
2023-04-20 14:48:06,888 - modelscope - INFO - No valid ast index found from /Users/lewaylee/.cache/modelscope/ast_indexer, generating ast index from prebuilt!
2023-04-20 14:48:06,926 - modelscope - INFO - Loading done! Current index file version is 1.5.0, with md5 77c50e7053fa0988be9fa0099573ada5 and a total number of 860 components indexed
2023-04-20 14:48:07,940 - modelscope - INFO - Use user-specified model revision: v1.0.1
2023-04-20 14:48:08,277 - modelscope - INFO - initiate model from /Users/lewaylee/.cache/modelscope/hub/damo/ChatPLUG-240M
2023-04-20 14:48:08,277 - modelscope - INFO - initiate model from location /Users/lewaylee/.cache/modelscope/hub/damo/ChatPLUG-240M.
2023-04-20 14:48:08,278 - modelscope - INFO - initialize model from /Users/lewaylee/.cache/modelscope/hub/damo/ChatPLUG-240M
| _IncompatibleKeys(missing_keys=[], unexpected_keys=['backbone.loss.generator.dense.weight', 'backbone.loss.generator.dense.bias', 'backbone.loss.criterion.one_hot'])
2023-04-20 14:48:14,821 - modelscope - INFO - cuda is not available, using cpu instead.
Traceback (most recent call last):
File "/Users/lewaylee/Projects/hub_runtime/ChatPLUG-240M.py", line 47, in
result = pipeline_ins(input,**kwargs)
File "/Users/lewaylee/miniconda3/envs/pytorch-gpu/lib/python3.10/site-packages/modelscope/pipelines/base.py", line 212, in call
output = self._process_single(input, *args, **kwargs)
File "/Users/lewaylee/miniconda3/envs/pytorch-gpu/lib/python3.10/site-packages/modelscope/pipelines/base.py", line 251, in _process_single
out = self.postprocess(out, **postprocess_params)
File "/Users/lewaylee/miniconda3/envs/pytorch-gpu/lib/python3.10/site-packages/modelscope/pipelines/nlp/fid_dialogue_pipeline.py", line 198, in postprocess
hypotheses[0], skip_special_tokens=self.is_t5)
UnboundLocalError: local variable 'hypotheses' referenced before assignment

Hello,

I'm currently studying your code and trying to understand how the full user dialogue history is divided into the short recent dialogue context and the long dialogue history, as described in the corresponding paper.

According to the paper:

"Besides, to avoid frequently encoding the full user history h, we divide it into a short recent dialogue context c and a long dialogue history l, i.e., h = [c, l], where only the short recent context c will interact with other side information in the encoder."

However, after examining the code, its not clear how this splitting process occurs. Could you provide more information on how this process works? or point to the part of the code that handles this division? Any help or guidance would be greatly appreciated.

Best regards

chatplug is very good at roleplaying, if I want to enrich the output of a custom role model, how should I organize the training data

1.请问为啥x-train chat_pipeline/chatplug_3.7B_train_sftv2.6.0_instruction.hjson 改成240M报错AttributeError: 'BertConfig' object has no attribute 'd_model'

2.请问在32G的显卡上用5K的数据微调3.7B的checkpoint爆显存不够，是我的参数没设置对吗？还是机器配置不够呢？

1.请问为啥x-train chat_pipeline/chatplug_3.7B_train_sftv2.6.0_instruction.hjson 改成240M报错AttributeError: 'BertConfig' object has no attribute 'd_model'

2.请问在32G的显卡上用5K的数据微调3.7B的checkpoint爆显存不够，是我的参数没设置对吗？还是机器配置不够呢？

hi，我在项目中找不到persona_benchmark.json。我已经读过Role-Play，但是仍有疑问。我是这样写的，似乎不起作用。

{"persona": "你是一个大学生，今年21岁，名字叫小婉。 我是你的男学长，你很喜欢我， 你的目标是让我愿意出来跟你约会。"} {"role_instruction": "请用萌妹子的风格回答， 请多多使用emoji表情。"}

help

在线的demo 中 https://www.modelscope.cn/studios/damo/role_play_chat/summary 可以公开rolejson 数据吗？
另外在线demo中使用的模型是chatplug 3.7B 吗？