qianfan.common.prompt package

Submodules

qianfan.common.prompt.prompt module

class qianfan.common.prompt.prompt.Prompt(template: Optional[str] = None, name: Optional[str] = None, id: Optional[str] = None, identifier: Literal['{}', '{{}}', '[]', '[[]]', '()', '(())'] = '{}', variables: Optional[List[str]] = None, labels: List[PromptLabel] = [], type: PromptType = PromptType.User, scene_type: PromptSceneType = PromptSceneType.Text2Text, framework_type: PromptFrameworkType = PromptFrameworkType.NotUse, negative_template: Optional[str] = None, negative_variables: Optional[List[str]] = None, creator_name: Optional[str] = None)

Bases: HubSerializable

class PromptEvaluateResult(prompt: Prompt, scene: List[Dict[str, Any]], summary: str)

Bases: object

Evaluation result of a prompt

prompt: Prompt

scene: List[Dict[str, Any]]

summary: str

classmethod base_prompt(prompt: str, background: str = '', additional_data: str = '', output_schema: str = '') → str

Generates a base type prompt for language models.

Parameters:

prompt (str):

The main text prompt that defines the task or query for the language model.

background (str, optional):

Additional context or background information to provide more context to the model.

additional_data (str, optional):

Extra data that can be included to enhance the specificity or details of the prompt.

output_schema (str, optional):

The desired schema for formatting the output of the language model.

Please refer to the following link for more details about CRISPE prompt.

API Doc: https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Zlo55g7t3

creator_name: Optional[str] = None

classmethod crispe_prompt(statement: str, capacity: str = '', insight: str = '', personality: str = '', experiment: str = '') → str

Generates a CRISPE-type prompt for fine-tuning models.

Parameters:

statement (str):

The main task that the model should focus on.

capacity (str, optional):

Capacity information specifying what role you want the model to play.

insight (str, optional):

Insights or guidance to provide additional context for the fine-tuning task.

personality (str, optional):

The output style or personality of the model.

experiment (str, optional):

The limit of the output range of the model.

Please refer to the following link for more details about CRISPE prompt.

API Doc: https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Hlo56qd21

delete() → None

Delete the prompt from Qianfan.

classmethod evaluate(prompt_list: List[Prompt], scenes: List[Dict[str, Any]], model: Completion, standard: PromptScoreStandard = PromptScoreStandard.Semantic) → List[PromptEvaluateResult]

Evaluate a list of prompts against specified scenes using the given model.

Parameters:

prompt_list (List[“Prompt”]):

A list of prompt templates to be evaluated.

scenes (List[Dict[str, Any]]):

List of scenes represented as dictionaries containing relevant information. The dict should contain the following keys:

• args: A dict containing the variables to be replaced in the prompt.

• expected: The expected output of the prompt.

client (Completion):

An instance of the Completion client.

standard (PromptScoreStandard, optional):

The scoring standard to be used for evaluating prompts.

Returns:

List[PromptEvaluateResult]:

A list of evaluation results, each containing the original prompt, the result of each scene, and a summary string.

Example: result_list = PromptEvaluateResult.evaluate(

prompt_list=[prompt1, prompt2, prompt3], scenes=[{

“args”: {“name”: “Alice”}, “expected”: “Hello, Alice!”

}], client=Completion(model=”ERNIE-Bot-4”), standard=PromptScoreStandard.Semantic

)

classmethod fewshot_prompt(prompt: str = '', examples: List[Tuple[str, str]] = []) → str

Generates a few-shot prompt for model input.

Parameters:

prompt (str):

The main prompt that sets the context for what task should the model focus on.

examples (List[Tuple[str, str]]):

A list of example tuples, where each tuple contains a model input string and its corresponding expected output. These examples help the model understand the desired behavior.

Please refer to the following link for more details about fewshot prompt.

API Doc: https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Nlo57dbf4

framework_type: PromptFrameworkType

classmethod from_file(path: str, identifier: Literal['{}', '{{}}', '[]', '[[]]', '()', '(())'] = '{}') → Prompt

Create a Prompt object from file. The file should only contain the prompt template.

Parameters:

path (str):

The path of the prompt file.

identifier (Literal[“{}”, “{{}}”, “[]”, “[[]]”, “()”, “(())”]):

The identifier of the prompt.

id: Optional[str] = None

identifier: Literal['{}', '{{}}', '[]', '[[]]', '()', '(())']

labels: List[PromptLabel]

name: Optional[str] = None

negative_template: Optional[str] = None

negative_variables: Optional[List[str]] = None

optimize(optimize_quality: bool = True, simplify_prompt: bool = False, iteration_round: Literal[1, 2] = 1, enable_cot: bool = False, app_id: Optional[int] = None, service_name: Optional[str] = None, **kwargs: Any) → Prompt

Optimize a prompt for better performance and effectiveness.

Parameters:

optimize_quality (bool):

Flag indicating whether to optimize for prompt quality.

simplify_prompt (bool):

Flag indicating whether to simplify the prompt structure.

iteration_round (Literal[1, 2]):

The number of optimization iterations to perform (1 or 2).

enable_cot (bool):

Flag indicating whether to enable chain of thought.

app_id (Optional[int]):

Optional application ID for context-specific optimizations.

service_name (Optional[str]):

Optional service used for optimizing.

**kwargs (Any):

Additional keyword arguments for future extensibility.

Returns:

Prompt:

The optimized Prompt object.

Please refer to the following link for more details about prompt optimization.

API Doc: https://cloud.baidu.com/doc/WENXINWORKSHOP/s/olr8svd33

render(**kwargs: str) → Tuple[str, Optional[str]]

Render the prompt with given variables.

Parameters:

kwargs (Any):

The value of the variables to be used for variable replacement in the template.

save_to_file(path: str) → None

Save the prompt template to file.

Parameters:

path (str):

The path of the prompt file.

scene_type: PromptSceneType

set_negative_template(template: str) → None

Set the prompt’s negative template. The variables in the template will be extracted.

Parameters:

template (str):

The new negative template.

set_template(template: str) → None

Set the prompt’s template. The variables in the template will be extracted.

Parameters:

template (str):

The new template.

template: str

type: PromptType

variables: List[str]

class qianfan.common.prompt.prompt.PromptLabel(id: int, name: str, color: str)

Bases: HubSerializable

Class of prompt label

color: str

the color displayed in the console

id: int

id of the label

name: str

name of the label