# kor schema，我们想要的输出格式
schema = Object(
    id="translate_result",
    description=(
        "任意文本的翻译结果。"
    ),
    attributes=[
        Text(
            id="chinese",
            description="中文翻译结果",
            examples=[], # Kor支持few-shot examples，但本例子比较简单，故不需要
            many=False, 
        ),
    ],
    many=False,
)

*# 运行结果*chain = create_extraction_chain(llm, schema, encoder_or_encoder_class='json')
text="We've trained a model, based on GPT-4, called CriticGPT to catch errors in ChatGPT's code output. We found that when people get help from CriticGPT to review ChatGPT code they outperform those without help 60% of the time. We are beginning the work to integrate CriticGPT-like models into our RLHF labeling pipeline, providing our trainers with explicit AI assistance. This is a step towards being able to evaluate outputs from advanced AI systems that can be difficult for people to rate without better tools."
print(chain.run(text)['data'])

{'translate_result': {'chinese': '我们训练了一个基于GPT-4的模型，称为CriticGPT，用于捕捉ChatGPT代码输出的错误。我们发现，当人们从CriticGPT那里获得帮助来审查ChatGPT代码时，他们比没有帮助的人高出60%的效率。我们正在开始将类似CriticGPT的模型集成到我们的RLHF标记流程中，为我们的训练师提供明确的AI辅助。这是朝着能够评估来自高级AI系统的输出迈出的一步，这些输出在没有更好的工具的情况下很难被人类评估。'}}

print(chain.prompt.format_prompt(text="[user input]").to_string())
Your goal is to extract structured information from the user's input that matches the form described below. When extracting information please make sure it matches the type information exactly. Do not add any attributes that do not appear in the schema shown below.

```TypeScript

translate_result: { // 任意文本的翻译结果。
 chinese: string // 中文翻译结果
}

### Example2

- 评价解析预期效果：输入一段用户评价，得到评价属性（口味、价格等）、评价极性（正向、负向、中立）、评价词（好吃、贵等）、参考片段。
- 结构化输出，第一步是定义schema，我们可以设置成这样的schema
- 在这个例子中，我们使用Pydantic Model来定义schema，Pydantic Model也能够支持few-shot examples，其额外好处是可以Validate
```python

# 评价解析的pydantic model
class Sentiment(enum.Enum):
    positive = "positive"
    negative = "negative"
    neural = "neural"

class Dianpin(BaseModel):
    aspect: str = Field(
        description="评价属性"
    )
    sentiment_word: str = Field(
        description='对评价属性的评价词，从原文中抽取',
    )
    sentiment: Optional[Sentiment] = Field(
        description='对评价属性的情感，positive\negative\neural中的一个',
    )
    reference: str = Field(
        description='评价的原文'
    ) 

# 运行kor
schema, validator = from_pydantic(
    Dianpin, 
    description='对评价的解析结果', 
    examples=[],  
    many=True #支持list of aspect
)
chain = create_extraction_chain(
    llm, schema, encoder_or_encoder_class="json", validator=validator
)

pprint(chain.run("整体来说，环境可以，味道的话也还不错，但价格有一点小贵。"))

{'data': {},
 'errors': [ParseError('The LLM has returned structured data which does not match the expected schema. Providing additional examples may help improve the parse.')],
 'raw': '\n'
        '<json>\n'
        '[\n'
        '  {\n'
        '    "aspect": "环境",\n'
        '    "sentiment_word": "可以",\n'
        '    "sentiment": "positive"\n'
        '  },\n'
        '  {\n'
        '    "aspect": "味道",\n'
        '    "sentiment_word": "还不错",\n'
        '    "sentiment": "positive"\n'
        '  },\n'
        '  {\n'
        '    "aspect": "价格",\n'
        '    "sentiment_word": "小贵",\n'
        '    "sentiment": "negative"\n'
        '  }\n'
        ']\n'
        '</json>',
 'validated_data': {}}

# 运行kor
schema, validator = from_pydantic(
    Dianpin, 
    description='对评价的解析结果', 
    examples=[
        ('味道真不错，下次还来！', [{"aspect":"味道", "sentiment_word": "真不错", "sentiment": "positive", "reference": "味道真不错"}])
    ],
    many=True #支持list of aspect
)
chain = create_extraction_chain(
    llm, schema, encoder_or_encoder_class="json", validator=validator
)

pprint(chain.run("整体来说，环境可以，味道的话也还不错，但价格有一点小贵。"))

{'data': {'dianpin': [{'aspect': '环境',
                       'reference': '整体来说，环境可以',
                       'sentiment': 'positive',
                       'sentiment_word': '可以'},
                      {'aspect': '味道',
                       'reference': '味道的话也还不错',
                       'sentiment': 'positive',
                       'sentiment_word': '还不错'},
                      {'aspect': '价格',
                       'reference': '但价格有一点小贵',
                       'sentiment': 'negative',
                       'sentiment_word': '小贵'}]},
 'errors': [],
 'raw': '\n'
        '<json>\n'
        '{\n'
        '  "dianpin": [\n'
        '    {\n'
        '      "aspect": "环境",\n'
        '      "sentiment_word": "可以",\n'
        '      "sentiment": "positive",\n'
        '      "reference": "整体来说，环境可以"\n'
        '    },\n'
        '    {\n'
        '      "aspect": "味道",\n'
        '      "sentiment_word": "还不错",\n'
        '      "sentiment": "positive",\n'
        '      "reference": "味道的话也还不错"\n'
        '    },\n'
        '    {\n'
        '      "aspect": "价格",\n'
        '      "sentiment_word": "小贵",\n'
        '      "sentiment": "negative",\n'
        '      "reference": "但价格有一点小贵"\n'
        '    }\n'
        '  ]\n'
        '}\n'
        '</json>',
 'validated_data': [Dianpin(aspect='环境', sentiment_word='可以', sentiment=<Sentiment.positive: 'positive'>, reference='整体来说，环境可以'),
                    Dianpin(aspect='味道', sentiment_word='还不错', sentiment=<Sentiment.positive: 'positive'>, reference='味道的话也还不错'),
                    Dianpin(aspect='价格', sentiment_word='小贵', sentiment=<Sentiment.negative: 'negative'>, reference='但价格有一点小贵')]}

print(chain.prompt.format_prompt(text="[user input]").to_string())

Your goal is to extract structured information from the user's input that matches the form described below. When extracting information please make sure it matches the type information exactly. Do not add any attributes that do not appear in the schema shown below.

```TypeScript

dianpin: Array<{ // 对评价的解析结果
 aspect: string // 评价属性
 sentiment_word: string // 对评价属性的评价词，从原文中抽取
 sentiment: "positive" | "negative" | "neural" // 对评价属性的情感，positive
egative
eural中的一个
 reference: string // 评价的原文
}>

## **小结**

Kor主要基于prompt，是对LLM的一层封装；Kor的设计理念使其便于进行数据处理（raw data -> schema），但Kor的最大限制是，**并不能保证所抽取内容的结构稳定性**，而这点将会被guided decoding类技术解决。

# Function Calling

这里以mistral发布的12B模型 -- **Mistral-Nemo-Instruct-2407**为例，初步研究其实现方式。

从一个简单的FC例子入手：

- tool是经典的`get_current_weather`，schema如下
```python
Function(
  name="get_current_weather",
  description="Get the current weather",
  parameters={
    "type": "object",
    "properties": {
      "location": {
          "type": "string",
          "description": "The city and state, e.g. San Francisco, CA"},
        "format": {
           "type": "string",
           "enum": ["celsius", "fahrenheit"],
           "description": "The temperature unit to use. Infer this from the users location."},
      },
     "required": ["location", "format"],
})

from enum import Enum
from pydantic import BaseModel

class Name(str, Enum):
    john = "John"
    paul = "Paul"

class Age(int, Enum):
    twenty = 20
    thirty = 30

class Character(BaseModel):
    name: Name
    age: Age

from outlines import models, generate

model = models.transformers("mistralai/Mistral-7B-v0.1")
generator = generate.json(model, Character)

char = generator("Generate a young character named Paul.")
print(char)
# Character(name:"Paul", age:20)