Skip to content

openai

OpenAILLM

Bases: LLM

Source code in src/distilabel/llm/openai.py 
 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
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
class OpenAILLM(LLM):
    def __init__(
        self,
        task: "Task",
        model: str = "gpt-3.5-turbo",
        client: Union["OpenAI", None] = None,
        openai_api_key: Union[str, None] = None,
        max_new_tokens: int = 128,
        frequency_penalty: float = 0.0,
        presence_penalty: float = 0.0,
        temperature: float = 1.0,
        top_p: float = 1.0,
        num_threads: Union[int, None] = None,
        prompt_format: Union["SupportedFormats", None] = None,
        prompt_formatting_fn: Union[Callable[..., str], None] = None,
    ) -> None:
        """Initializes the OpenAILLM class.

        Args:
            task (Task): the task to be performed by the LLM.
            model (str, optional): the model to be used for generation. Defaults to "gpt-3.5-turbo".
            client (Union[OpenAI, None], optional): an OpenAI client to be used for generation.
                If `None`, a new client will be created. Defaults to `None`.
            openai_api_key (Union[str, None], optional): the OpenAI API key to be used for generation.
                If `None`, the `OPENAI_API_KEY` environment variable will be used. Defaults to `None`.
            max_new_tokens (int, optional): the maximum number of tokens to be generated.
                Defaults to 128.
            frequency_penalty (float, optional): the frequency penalty to be used for generation.
                Defaults to 0.0.
            presence_penalty (float, optional): the presence penalty to be used for generation.
                Defaults to 0.0.
            temperature (float, optional): the temperature to be used for generation.
                Defaults to 1.0.
            top_p (float, optional): the top-p value to be used for generation.
                Defaults to 1.0.
            num_threads (Union[int, None], optional): the number of threads to be used
                for parallel generation. If `None`, no parallel generation will be performed.
                Defaults to `None`.
            prompt_format (Union[SupportedFormats, None], optional): the format to be used
                for the prompt. If `None`, the default format of the task will be used, available
                formats are `openai`, `chatml`, `llama2`, `zephyr`, and `default`. Defaults to `None`,
                but `default` (concatenation of `system_prompt` and `formatted_prompt` with a line-break)
                will be used if no `prompt_formatting_fn` is provided.
            prompt_formatting_fn (Union[Callable[..., str], None], optional): a function to be
                applied to the prompt before generation. If `None`, no formatting will be applied.
                Defaults to `None`.

        Raises:
            AssertionError: if the provided `model` is not available in your OpenAI account.

        Examples:
            >>> from distilabel.tasks.text_generation import TextGenerationTask as Task
            >>> from distilabel.llm import OpenAILLM
            >>> task = Task()
            >>> llm = OpenAILLM(model="gpt-3.5-turbo", task=task)
        """
        super().__init__(
            task=task,
            num_threads=num_threads,
            prompt_format=prompt_format,
            prompt_formatting_fn=prompt_formatting_fn,
        )

        if not _OPENAI_AVAILABLE:
            raise ImportError(
                "`OpenAILLM` cannot be used as `openai` is not installed, please "
                " install it with `pip install openai`."
            )

        self.max_tokens = max_new_tokens
        self.frequency_penalty = frequency_penalty
        self.presence_penalty = presence_penalty
        self.temperature = temperature
        self.top_p = top_p

        self.client = client or OpenAI(api_key=openai_api_key, max_retries=6)

        assert (
            model in self.available_models
        ), f"Provided `model` is not available in your OpenAI account, available models are {self.available_models}"
        self.model = model

    def __rich_repr__(self) -> Generator[Any, None, None]:
        yield from super().__rich_repr__()
        yield (
            "parameters",
            {
                "max_tokens": self.max_tokens,
                "frequency_penalty": self.frequency_penalty,
                "presence_penalty": self.presence_penalty,
                "temperature": self.temperature,
                "top_p": self.top_p,
            },
        )

    @cached_property
    def available_models(self) -> List[str]:
        """Returns the list of available models in your OpenAI account."""
        return [model.id for model in self.client.models.list().data]

    @property
    def model_name(self) -> str:
        """Returns the name of the OpenAI model."""
        return self.model

    def _generate(
        self,
        inputs: List[Dict[str, Any]],
        num_generations: int = 1,
    ) -> List[List[LLMOutput]]:
        """Generates `num_generations` for each input in `inputs`.

        Args:
            inputs (List[Dict[str, Any]]): the inputs to be used for generation.
            num_generations (int, optional): the number of generations to be performed for each
                input. Defaults to 1.

        Returns:
            List[List[LLMOutput]]: the generated outputs.
        """
        prompts = self._generate_prompts(
            inputs, default_format="openai", expected_output_type=list
        )
        outputs = []
        for prompt in prompts:
            chat_completions = self.client.chat.completions.create(
                messages=prompt,
                model=self.model,
                n=num_generations,
                max_tokens=self.max_tokens,
                frequency_penalty=self.frequency_penalty,
                presence_penalty=self.presence_penalty,
                temperature=self.temperature,
                top_p=self.top_p,
                timeout=50,
            )

            output = []
            for chat_completion in chat_completions.choices:
                try:
                    parsed_response = self.task.parse_output(
                        chat_completion.message.content.strip()
                    )
                except Exception as e:
                    logger.error(f"Error parsing OpenAI response: {e}")
                    parsed_response = None
                output.append(
                    LLMOutput(
                        model_name=self.model_name,
                        prompt_used=prompt,
                        raw_output=chat_completion.message.content,
                        parsed_output=parsed_response,
                    )
                )
            outputs.append(output)
        return outputs

available_models: List[str] cached property

Returns the list of available models in your OpenAI account.

model_name: str property

Returns the name of the OpenAI model.

__init__(task, model='gpt-3.5-turbo', client=None, openai_api_key=None, max_new_tokens=128, frequency_penalty=0.0, presence_penalty=0.0, temperature=1.0, top_p=1.0, num_threads=None, prompt_format=None, prompt_formatting_fn=None)

Initializes the OpenAILLM class.

Parameters:

Name Type Description Default
task Task

the task to be performed by the LLM.

 required
model str

the model to be used for generation. Defaults to "gpt-3.5-turbo".

 'gpt-3.5-turbo'
client Union[OpenAI, None]

an OpenAI client to be used for generation. If None, a new client will be created. Defaults to None.

 None
openai_api_key Union[str, None]

the OpenAI API key to be used for generation. If None, the OPENAI_API_KEY environment variable will be used. Defaults to None.

 None
max_new_tokens int

the maximum number of tokens to be generated. Defaults to 128.

 128
frequency_penalty float

the frequency penalty to be used for generation. Defaults to 0.0.

 0.0
presence_penalty float

the presence penalty to be used for generation. Defaults to 0.0.

 0.0
temperature float

the temperature to be used for generation. Defaults to 1.0.

 1.0
top_p float

the top-p value to be used for generation. Defaults to 1.0.

 1.0
num_threads Union[int, None]

the number of threads to be used for parallel generation. If None, no parallel generation will be performed. Defaults to None.

 None
prompt_format Union[SupportedFormats, None]

the format to be used for the prompt. If None, the default format of the task will be used, available formats are openai, chatml, llama2, zephyr, and default. Defaults to None, but default (concatenation of system_prompt and formatted_prompt with a line-break) will be used if no prompt_formatting_fn is provided.

 None
prompt_formatting_fn Union[Callable[..., str], None]

a function to be applied to the prompt before generation. If None, no formatting will be applied. Defaults to None.

 None

Raises:

Type Description
AssertionError

if the provided model is not available in your OpenAI account.

Examples:

>>> from distilabel.tasks.text_generation import TextGenerationTask as Task
>>> from distilabel.llm import OpenAILLM
>>> task = Task()
>>> llm = OpenAILLM(model="gpt-3.5-turbo", task=task)
Source code in src/distilabel/llm/openai.py 
 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
def __init__(
    self,
    task: "Task",
    model: str = "gpt-3.5-turbo",
    client: Union["OpenAI", None] = None,
    openai_api_key: Union[str, None] = None,
    max_new_tokens: int = 128,
    frequency_penalty: float = 0.0,
    presence_penalty: float = 0.0,
    temperature: float = 1.0,
    top_p: float = 1.0,
    num_threads: Union[int, None] = None,
    prompt_format: Union["SupportedFormats", None] = None,
    prompt_formatting_fn: Union[Callable[..., str], None] = None,
) -> None:
    """Initializes the OpenAILLM class.

    Args:
        task (Task): the task to be performed by the LLM.
        model (str, optional): the model to be used for generation. Defaults to "gpt-3.5-turbo".
        client (Union[OpenAI, None], optional): an OpenAI client to be used for generation.
            If `None`, a new client will be created. Defaults to `None`.
        openai_api_key (Union[str, None], optional): the OpenAI API key to be used for generation.
            If `None`, the `OPENAI_API_KEY` environment variable will be used. Defaults to `None`.
        max_new_tokens (int, optional): the maximum number of tokens to be generated.
            Defaults to 128.
        frequency_penalty (float, optional): the frequency penalty to be used for generation.
            Defaults to 0.0.
        presence_penalty (float, optional): the presence penalty to be used for generation.
            Defaults to 0.0.
        temperature (float, optional): the temperature to be used for generation.
            Defaults to 1.0.
        top_p (float, optional): the top-p value to be used for generation.
            Defaults to 1.0.
        num_threads (Union[int, None], optional): the number of threads to be used
            for parallel generation. If `None`, no parallel generation will be performed.
            Defaults to `None`.
        prompt_format (Union[SupportedFormats, None], optional): the format to be used
            for the prompt. If `None`, the default format of the task will be used, available
            formats are `openai`, `chatml`, `llama2`, `zephyr`, and `default`. Defaults to `None`,
            but `default` (concatenation of `system_prompt` and `formatted_prompt` with a line-break)
            will be used if no `prompt_formatting_fn` is provided.
        prompt_formatting_fn (Union[Callable[..., str], None], optional): a function to be
            applied to the prompt before generation. If `None`, no formatting will be applied.
            Defaults to `None`.

    Raises:
        AssertionError: if the provided `model` is not available in your OpenAI account.

    Examples:
        >>> from distilabel.tasks.text_generation import TextGenerationTask as Task
        >>> from distilabel.llm import OpenAILLM
        >>> task = Task()
        >>> llm = OpenAILLM(model="gpt-3.5-turbo", task=task)
    """
    super().__init__(
        task=task,
        num_threads=num_threads,
        prompt_format=prompt_format,
        prompt_formatting_fn=prompt_formatting_fn,
    )

    if not _OPENAI_AVAILABLE:
        raise ImportError(
            "`OpenAILLM` cannot be used as `openai` is not installed, please "
            " install it with `pip install openai`."
        )

    self.max_tokens = max_new_tokens
    self.frequency_penalty = frequency_penalty
    self.presence_penalty = presence_penalty
    self.temperature = temperature
    self.top_p = top_p

    self.client = client or OpenAI(api_key=openai_api_key, max_retries=6)

    assert (
        model in self.available_models
    ), f"Provided `model` is not available in your OpenAI account, available models are {self.available_models}"
    self.model = model