Skip to content

Synthetic

Generator

Generator class for synthetic data generation.

Source code in mic_toolkit/synthetic/generation.py 
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
class Generator:
    """Generator class for synthetic data generation."""

    def __init__(
        self,
        endpoint: str,
        model: str,
    ):
        """Initializes the LLM Client and model.

        Args:
            endpoint (str): Endpoint for the LLM API. For Ollama it is usually "http://localhost:11434".
            model (str): Name of the model to use for generation. Find it using 'ollama list'.
        """
        self.client = Client(endpoint)
        self.model = model

    def generate_text(
        self,
        data: pd.DataFrame,
        system_prompt: str = "You are a helpful AI assistant. Please provide a response to the following user query:",
        max_tokens: int = None,
    ) -> pd.DataFrame:
        """_summary_

        Args:
            data (pd.DataFrame): Dataframe with a single column of text data.
            system_prompt (_type_, optional): optional System prompt. Defaults to "You are a helpful AI assistant. Please provide a response to the following user query:".
            max_tokens (int, optional): max output tokens. Defaults to None.

        Returns:
            pd.DataFrame: Output dataframe with generated text.
        """

        def generate_response(text):
            options = {}
            if max_tokens is not None:
                options["num_predict"] = max_tokens
            return self.client.chat(
                model=self.model,
                messages=[
                    {"role": "assistant", "content": system_prompt},
                    {"role": "user", "content": text},
                ],
                options=options,
            )["message"]["content"]

        data["output"] = data[data.columns[0]].progress_apply(generate_response)

        return data

    def create_system_prompt(self, labels: list[str], query: str = "") -> str:
        labels_str = ", ".join(labels)
        if query:
            return f"Classify the following text into one of the following categories: {labels_str} based on {query}. Just answer with the label. Absolutely no context is needed."
        else:
            return f"Classify the following text into one of the following categories: {labels_str}. Just answer with the label. Absolutely no context is needed."

    def generate_labels(
        self,
        labels: list[str],
        data: pd.DataFrame,
        query: str = "",
        max_tokens: int = None,
        max_tries: int = 5,
    ) -> pd.DataFrame:
        """_summary_

        Args:
            labels (list[str]): List of labels to classify the data into.
            data (pd.DataFrame): Dataframe with a single column of text data.
            query (str, optional): Classification query. Defaults to "".
            max_tokens (int, optional): max output tokens. Defaults to None.
            max_tries (int, optional): max tries to get the correct label. Defaults to 5.

        Returns:
            pd.DataFrame: _description_
        """
        system_prompt = self.create_system_prompt(labels, query)

        def classify_text(text):
            options = {}
            if max_tokens is not None:
                options["num_predict"] = max_tokens
            response = self.client.chat(
                model=self.model,
                messages=[
                    {"role": "system", "content": system_prompt},
                    {"role": "user", "content": text},
                ],
            )["message"]["content"]
            tries = max_tries
            while response not in labels and tries > 0:
                response = self.client.chat(
                    model=self.model,
                    messages=[
                        {
                            "role": "system",
                            "content": "You did not respond with just the label please respond again with the label only. Without any context or explanation"
                            + system_prompt,
                        },
                        {"role": "user", "content": text},
                    ],
                    options=options,
                )["message"]["content"]
                tries -= 1
            return response

        data["label"] = data[data.columns[0]].progress_apply(classify_text)
        return data

__init__(endpoint, model)

Initializes the LLM Client and model.

Parameters:

Name Type Description Default
endpoint str

Endpoint for the LLM API. For Ollama it is usually "http://localhost:11434".

 required
model str

Name of the model to use for generation. Find it using 'ollama list'.

 required
Source code in mic_toolkit/synthetic/generation.py 
11
12
13
14
15
16
17
18
19
20
21
22
23
def __init__(
    self,
    endpoint: str,
    model: str,
):
    """Initializes the LLM Client and model.

    Args:
        endpoint (str): Endpoint for the LLM API. For Ollama it is usually "http://localhost:11434".
        model (str): Name of the model to use for generation. Find it using 'ollama list'.
    """
    self.client = Client(endpoint)
    self.model = model

generate_labels(labels, data, query='', max_tokens=None, max_tries=5)

summary

Parameters:

Name Type Description Default
labels list[str]

List of labels to classify the data into.

 required
data DataFrame

Dataframe with a single column of text data.

 required
query str

Classification query. Defaults to "".

 ''
max_tokens int

max output tokens. Defaults to None.

 None
max_tries int

max tries to get the correct label. Defaults to 5.

 5

Returns:

Type Description
DataFrame

pd.DataFrame: description
Source code in mic_toolkit/synthetic/generation.py 
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
def generate_labels(
    self,
    labels: list[str],
    data: pd.DataFrame,
    query: str = "",
    max_tokens: int = None,
    max_tries: int = 5,
) -> pd.DataFrame:
    """_summary_

    Args:
        labels (list[str]): List of labels to classify the data into.
        data (pd.DataFrame): Dataframe with a single column of text data.
        query (str, optional): Classification query. Defaults to "".
        max_tokens (int, optional): max output tokens. Defaults to None.
        max_tries (int, optional): max tries to get the correct label. Defaults to 5.

    Returns:
        pd.DataFrame: _description_
    """
    system_prompt = self.create_system_prompt(labels, query)

    def classify_text(text):
        options = {}
        if max_tokens is not None:
            options["num_predict"] = max_tokens
        response = self.client.chat(
            model=self.model,
            messages=[
                {"role": "system", "content": system_prompt},
                {"role": "user", "content": text},
            ],
        )["message"]["content"]
        tries = max_tries
        while response not in labels and tries > 0:
            response = self.client.chat(
                model=self.model,
                messages=[
                    {
                        "role": "system",
                        "content": "You did not respond with just the label please respond again with the label only. Without any context or explanation"
                        + system_prompt,
                    },
                    {"role": "user", "content": text},
                ],
                options=options,
            )["message"]["content"]
            tries -= 1
        return response

    data["label"] = data[data.columns[0]].progress_apply(classify_text)
    return data

generate_text(data, system_prompt='You are a helpful AI assistant. Please provide a response to the following user query:', max_tokens=None)

summary

Parameters:

Name Type Description Default
data DataFrame

Dataframe with a single column of text data.

 required
system_prompt _type_

optional System prompt. Defaults to "You are a helpful AI assistant. Please provide a response to the following user query:".

 'You are a helpful AI assistant. Please provide a response to the following user query:'
max_tokens int

max output tokens. Defaults to None.

 None

Returns:

Type Description
DataFrame

pd.DataFrame: Output dataframe with generated text.
Source code in mic_toolkit/synthetic/generation.py 
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
def generate_text(
    self,
    data: pd.DataFrame,
    system_prompt: str = "You are a helpful AI assistant. Please provide a response to the following user query:",
    max_tokens: int = None,
) -> pd.DataFrame:
    """_summary_

    Args:
        data (pd.DataFrame): Dataframe with a single column of text data.
        system_prompt (_type_, optional): optional System prompt. Defaults to "You are a helpful AI assistant. Please provide a response to the following user query:".
        max_tokens (int, optional): max output tokens. Defaults to None.

    Returns:
        pd.DataFrame: Output dataframe with generated text.
    """

    def generate_response(text):
        options = {}
        if max_tokens is not None:
            options["num_predict"] = max_tokens
        return self.client.chat(
            model=self.model,
            messages=[
                {"role": "assistant", "content": system_prompt},
                {"role": "user", "content": text},
            ],
            options=options,
        )["message"]["content"]

    data["output"] = data[data.columns[0]].progress_apply(generate_response)

    return data