dustyposa / rasa_ch_faq Goto Github PK

View Code? Open in Web Editor NEW

186.0 7.0 58.0 285 KB

用 rasa 实现 rasa demo 机器人，有一些惊奇的功能，faq，图谱，多轮等

License: MIT License

Python 99.04% Dockerfile 0.96%

rasa python3 dialogue

rasa_ch_faq's Introduction

rasa_ch_bot

用 RASA 实现 RASA Bot 后端。能够回答关于 RASA 的问题，以及一些例子。欢迎大家多提 RASA 相关的问题，或者想看的示例，我会补充在这里。

前端地址: 这里

功能更新

[2021-08-04] 增加 onnx + 量化（用于提高特征提取的速度）的 feature 提取组件 # 一些注意事项
[2021-06-24] 增加知图谱的接入(放在GRAPH分支了)
[2021-06-01] 增加文本纠错 pipeline (由于 demo 较慢，默认未开启，如何开启？)
[2021-05-20] AlbertFeaturizer （在dev分支）
[2021-05-13] 增加按钮 demo
[2021-05-11] 支持查看 BTC 行情
[2021-05-08] 支持 吸猫\狗\狐狸。支持根据图片搜索动漫出处。
[2021-05-06] 支持 找饭店 demo。
[2021-04-13] 实现追问demo，实现细节。

部分功能展示

FAQ

天气查询及BTC查询功能

吸动物

搜动漫

知识图谱

支持的问题列表

请参见: 问题列举

一些配置

分词使用的 bert，自定义了 tokenizers

如何运行

由于使用了 bert_chinese，所以需要下载 bert_chinese 模型。并放到 pre_models 文件夹中，重命名为 tf_model.h5 命令执行:

curl -L https://www.flyai.com/m/bert-base-chinese-tf_model.h5 -o pre_models/tf_model.h5
rasa train

一些文件说明

run.py  # 相当于运行 rasa run
train.py  # == rasa train
run_action_server.py  # == rasa run actions
load_model.py   # 直接加载并运行模型，与 server 无关。（需要先训练好一个模型）

一些工具

back_translation.py  # 回译脚本
# 使用方式
python back_translation.py 需要回译的文本

ONNX注意事项

1. 使用说明

config 更改为：

  - name: compoments.nlu.featurizer.lm_featurizer.OnnxLanguageModelFeaturizer
    cache_dir: ./tmp
    model_name: bert
    model_weights: pre_models
    onnx: false  # 是否开启 onnx
    quantize: true  # 是否使用量化

下载 torch 的模型

curl -L https://www.flyai.com/m/bert-base-chinese-pytorch_model.bin -o pre_models/pytorch_model.bin

2. 依赖安装

pip install torch==1.9.0 transformers==4.8.2 onnx==1.9.0 onnxruntime==1.8.0 onnxruntime-tools==1.7.0 psutil==5.8.0

3. 一些注释

速度能提升多少，可以参考这篇文章
量化后速度有额外提升，但是效果可能会变差，需要根据语料调整
如何测试效果并查看结果

rasa train nlu && rasa test nlu
cat results/intent_errors.json

为什么没有用 tensorflow 用来做 onnx 尝试多次，都失败了，暂时未找到解决办法（输入的纬度不匹配），如果有人成功了，可以告诉我！！感谢！！！🙏

从零开始搭建机器人

1. 下载项目并进入

git clone https://github.com/Dustyposa/rasa_ch_faq.git 
cd rasa_ch_faq

2. 安装依赖

pip install -r requirements.txt
curl -L https://www.flyai.com/m/bert-base-chinese-tf_model.h5 -o pre_models/tf_model.h5
rasa train

ps: 注意 python 版本 37+

3. 训练模型

rasa train

4. 运行机器人

需要开两个 shell/iterm 第一个:

rasa shell

第二个:

rasa run actions

然后就可以在第一个 shell 窗口对话了

从 1 开始搭建机器人

这个是干啥的?如果第一个你已经会了，我们加点前端展示的，效果参看这里

1. 下载前端项目并进入

git clone -#-depth 1 https://github.com/Dustyposa/rasa_bot_front
cd rasa_bot_front

2. 启动前端

参照文档

3. 启动 rasa

同样是两个 shell/iterm，第一个命令稍有不同:

rasa run --cors "*"

第二个:

rasa run actions

rasa_ch_faq's People

Contributors

Stargazers

Watchers

rasa_ch_faq's Issues

单独 python bert_tokenizer.py 报错，是不是huggingface加载模型方式有变化？

rasa train 和rasa shell 都没问题。

python train.py 报错，麻烦帮看下啥问题

Some layers from the model checkpoint at pre_models were not used when initializing TFBertModel: ['mlm___cls', 'nsp___cls']

This IS expected if you are initializing TFBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
This IS NOT expected if you are initializing TFBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).
All the layers of TFBertModel were initialized from the model checkpoint at pre_models.
If your task is similar to the task the model of the checkpoint was trained on, you can already use TFBertModel for predictions without further training.
2021-01-29 19:33:57.271548: W tensorflow/core/framework/op_kernel.cc:1767] OP_REQUIRES failed at einsum_op_impl.h:714 : Resource exhausted: OOM when allocating tensor with shape[64,12,218,218] and type float on /job:localhost/replica:0/task:0/device:CPU:0 by allocator cpu
Traceback (most recent call last):
File "train.py", line 6, in
training_files="data",
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/train.py", line 109, in train
loop,
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/utils/common.py", line 308, in run_in_loop
result = loop.run_until_complete(f)
File "uvloop/loop.pyx", line 1456, in uvloop.loop.Loop.run_until_complete
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/train.py", line 174, in train_async
finetuning_epoch_fraction=finetuning_epoch_fraction,
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/train.py", line 353, in _train_async_internal
finetuning_epoch_fraction=finetuning_epoch_fraction,
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/train.py", line 396, in _do_training
finetuning_epoch_fraction=finetuning_epoch_fraction,
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/train.py", line 818, in _train_nlu_with_validated_data
**additional_arguments,
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/nlu/train.py", line 116, in train
interpreter = trainer.train(training_data, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/nlu/model.py", line 209, in train
updates = component.train(working_data, self.config, **context)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/nlu/featurizers/dense_featurizer/lm_featurizer.py", line 814, in train
batch_docs = self._get_docs_for_batch(batch_messages, attribute)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/nlu/featurizers/dense_featurizer/lm_featurizer.py", line 767, in _get_docs_for_batch
batch_token_ids, batch_tokens, batch_examples, attribute, inference_mode
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/nlu/featurizers/dense_featurizer/lm_featurizer.py", line 685, in _get_model_features_for_batch
batch_attention_mask, padded_token_ids
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/rasa/nlu/featurizers/dense_featurizer/lm_featurizer.py", line 537, in _compute_batch_sequence_features
np.array(padded_token_ids), attention_mask=np.array(batch_attention_mask)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in call
outputs = call_fn(inputs, *args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/transformers/models/bert/modeling_tf_bert.py", line 901, in call
training=inputs["training"],
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in call
outputs = call_fn(inputs, *args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/transformers/models/bert/modeling_tf_bert.py", line 694, in call
training=inputs["training"],
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in call
outputs = call_fn(inputs, *args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/transformers/models/bert/modeling_tf_bert.py", line 452, in call
hidden_states, attention_mask, head_mask[i], output_attentions, training=training
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in call
outputs = call_fn(inputs, *args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/transformers/models/bert/modeling_tf_bert.py", line 418, in call
hidden_states, attention_mask, head_mask, output_attentions, training=training
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in call
outputs = call_fn(inputs, *args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/transformers/models/bert/modeling_tf_bert.py", line 354, in call
input_tensor, attention_mask, head_mask, output_attentions, training=training
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in call
outputs = call_fn(inputs, *args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/transformers/models/bert/modeling_tf_bert.py", line 288, in call
attention_scores = tf.einsum("aecd,abcd->acbe", key_layer, query_layer)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/util/dispatch.py", line 201, in wrapper
return target(*args, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/ops/special_math_ops.py", line 684, in einsum
return _einsum_v2(equation, *inputs, **kwargs)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/ops/special_math_ops.py", line 1113, in _einsum_v2
return gen_linalg_ops.einsum(inputs, resolved_equation)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/ops/gen_linalg_ops.py", line 1088, in einsum
_ops.raise_from_not_ok_status(e, name)
File "/root/anaconda3/envs/rasa/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 6843, in raise_from_not_ok_status
six.raise_from(core._status_to_exception(e.code, message), None)
File "", line 3, in raise_from
tensorflow.python.framework.errors_impl.ResourceExhaustedError: OOM when allocating tensor with shape[64,12,218,218] and type float on /job:localhost/replica:0/task:0/device:CPU:0 by allocator cpu [Op:Einsum]

RASA提问，增加RASA FAQ的语料库

想到一个加一个吧。

1. rasa run和rasa run actions的区别？
2. Rasa和RasaX的区别？
3. Rasa在何处调用Bert?
4. 回译文本工具的作用是什么？
5. API的形式调用Rasa的地址是多少以及需要传递什么参数？(答案)

训练时间过长

当nlu的语料>20000条，训练的时候，如果不用ONNX模型花费的时间为15分钟，使用ONNX模型需要花费2个多小时，即使用GPU也是这样，这个有什么优化方法吗？

在前端中无法正常问答，该排查哪一部分呢？

运行截图如下

后端已经开启：

actions也是开启的：

而且使用rasa shell的话是可以正常问答的，现在就是前端无法回答，想问一下您之前跑的时候遇到过类似的问题吗？

多轮对话确定一个推荐，怎么实现？

多轮对话确定一个推荐，怎么实现？
人：你好
AI：你好
人：我要找大学
AI：你什么学历
人：高中
AI：你什么年龄
人：18岁
AI：你想读什么专业
人：计算机
AI：你想再什么地区
人：北京
AI：推荐你北京大学

如果要实现多轮条件合并才能推出结果，怎么实现呢?

关于前端文件的使用

我使用了您开源出来的前端项目，另外我没有使用 npm 部署前端服务, 而是直接使用了您给出的命令行: --cors "*" 加载了这个前端文件。但是我有一个问题：

rasa 的 core 服务是如何和这个前端服务打通的，也就是说，打开前端后，我每次从前端输入的一句话是如何进入到 rasa 内部中，并且 rasa 可以维持住这个对话的状态的。
这个问题基础上面的第一点，因为我现在有一个训练好的 rasa 服务，但是我需要把这个服务添加（或者说引入）到一个目前现有的聊天对话页面中，我应该如何做呢？

还请您不吝赐教。期待得到您的回复，我的 VX：18336304089
如果您愿意提供帮助，我愿意有偿获得您的指导与帮助。

运行rasa train时报错，报错信息如下：
Traceback (most recent call last):
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/transformers/modeling_tf_utils.py", line 1232, in from_pretrained
missing_keys, unexpected_keys = load_tf_weights(model, resolved_archive_file)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/transformers/modeling_tf_utils.py", line 459, in load_tf_weights
with h5py.File(resolved_archive_file, "r") as f:
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/h5py/_hl/files.py", line 406, in init
fid = make_fid(name, mode, userblock_size,
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/h5py/_hl/files.py", line 173, in make_fid
fid = h5f.open(name, flags, fapl=fapl)
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "h5py/h5f.pyx", line 88, in h5py.h5f.open
OSError: Unable to open file (file signature not found)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/Users/anaconda3/envs/rasa/bin/rasa", line 8, in
sys.exit(main())
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/main.py", line 117, in main
cmdline_arguments.func(cmdline_arguments)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/cli/train.py", line 59, in
train_parser.set_defaults(func=lambda args: run_training(args, can_exit=True))
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/cli/train.py", line 91, in run_training
training_result = train_all(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/api.py", line 109, in train
return rasa.utils.common.run_in_loop(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/utils/common.py", line 296, in run_in_loop
result = loop.run_until_complete(f)
File "uvloop/loop.pyx", line 1456, in uvloop.loop.Loop.run_until_complete
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/model_training.py", line 108, in train_async
return await _train_async_internal(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/model_training.py", line 289, in _train_async_internal
await _do_training(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/model_training.py", line 335, in _do_training
model_path = await _train_nlu_with_validated_data(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/model_training.py", line 759, in _train_nlu_with_validated_data
await rasa.nlu.train.train(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/train.py", line 97, in train
trainer = Trainer(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/model.py", line 167, in init
self.pipeline = self._build_pipeline(cfg, component_builder)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/model.py", line 178, in _build_pipeline
component = component_builder.create_component(component_cfg, cfg)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/components.py", line 938, in create_component
component = registry.create_component_by_config(component_config, cfg)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/registry.py", line 193, in create_component_by_config
return component_class.create(component_config, config)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/featurizers/dense_featurizer/lm_featurizer.py", line 105, in create
return cls(component_config, hf_transformers_loaded=hf_transformers_loaded)
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/rasa/nlu/featurizers/dense_featurizer/lm_featurizer.py", line 88, in init
self._load_model_instance(skip_model_load)
File "/Users/wenhk/Downloads/rasa_ch_faq/compoments/nlu/featurizer/lm_featurizer.py", line 83, in _load_model_instance
self.model = model_class_dict[self.model_name].from_pretrained(
File "/Users/anaconda3/envs/rasa/lib/python3.8/site-packages/transformers/modeling_tf_utils.py", line 1234, in from_pretrained
raise OSError(
OSError: Unable to load weights from h5 file. If you tried to load a TF 2.0 model from a PyTorch checkpoint, please set from_pt=True.

请问这是什么问题？

执行rasa命令提示命未找到命令

[root@localhost rasa_ch_faq]# rasa train
-bash: rasa: 未找到命令

大佬执行rasa命令提示这个是啥情况