kiln_ai.adapters.repair.repair_task
1import json 2 3from pydantic import BaseModel, Field 4 5from kiln_ai.adapters.prompt_builders import BasePromptBuilder, prompt_builder_from_id 6from kiln_ai.datamodel import Priority, Project, Task, TaskRequirement, TaskRun 7 8 9# We should add evaluator rating 10class RepairTaskInput(BaseModel): 11 original_prompt: str 12 original_input: str 13 original_output: str 14 evaluator_feedback: str = Field( 15 min_length=1, 16 description="Feedback from an evaluator on how to repair the task run.", 17 ) 18 19 20class RepairTaskRun(Task, parent_of={}): 21 def __init__(self, original_task: Task): 22 # Keep the typechecker happy 23 tmp_project = Project(name="Repair") 24 super().__init__( 25 name="Repair", 26 parent=tmp_project, 27 description="Repair a task run, given feedback from an evaluator about how the response can be improved.", 28 instruction="You are an assistant which helps improve output from another assistant (original assistant). You'll be provided a task that the original assistant executed (prompt), \ 29the input it was given, and the output it generated. An evaluator has determined that the output it generated did not satisfy the task and should be improved. The evaluator will provide \ 30feedback describing what should be improved. Your job is to understand the evaluator's feedback and improve the response.", 31 requirements=[ 32 TaskRequirement( 33 name="Follow Eval Feedback", 34 instruction="The evaluator's feedback is the most important thing to consider. If it conflicts with the original task instruction or prompt, prioritize the evaluator's feedback.", 35 priority=Priority.p0, 36 ) 37 ], 38 input_json_schema=json.dumps(RepairTaskInput.model_json_schema()), 39 output_json_schema=original_task.output_json_schema, 40 ) 41 42 @classmethod 43 def _original_prompt(cls, run: TaskRun, task: Task) -> str: 44 if run.output.source is None or run.output.source.properties is None: 45 raise ValueError("No source properties found") 46 47 # Get the prompt builder id. Need the second check because we used to store this in a prompt_builder_name field, so loading legacy runs will need this. 48 prompt_id = run.output.source.properties.get( 49 "prompt_id" 50 ) or run.output.source.properties.get("prompt_builder_name", None) 51 if prompt_id is not None and isinstance(prompt_id, str): 52 prompt_builder = prompt_builder_from_id(prompt_id, task) 53 if isinstance(prompt_builder, BasePromptBuilder): 54 return prompt_builder.build_prompt(include_json_instructions=False) 55 56 raise ValueError(f"Prompt builder '{prompt_id}' is not a valid prompt builder") 57 58 @classmethod 59 def build_repair_task_input( 60 cls, original_task: Task, task_run: TaskRun, evaluator_feedback: str 61 ) -> RepairTaskInput: 62 original_prompt = cls._original_prompt(task_run, original_task) 63 return RepairTaskInput( 64 original_prompt=original_prompt, 65 original_input=task_run.input, 66 original_output=task_run.output.output, 67 evaluator_feedback=evaluator_feedback, 68 )
11class RepairTaskInput(BaseModel): 12 original_prompt: str 13 original_input: str 14 original_output: str 15 evaluator_feedback: str = Field( 16 min_length=1, 17 description="Feedback from an evaluator on how to repair the task run.", 18 )
Usage docs: https://docs.pydantic.dev/2.10/concepts/models/
A base class for creating Pydantic models.
Attributes:
__class_vars__: The names of the class variables defined on the model.
__private_attributes__: Metadata about the private attributes of the model.
__signature__: The synthesized __init__
[Signature
][inspect.Signature] of the model.
__pydantic_complete__: Whether model building is completed, or if there are still undefined fields.
__pydantic_core_schema__: The core schema of the model.
__pydantic_custom_init__: Whether the model has a custom `__init__` function.
__pydantic_decorators__: Metadata containing the decorators defined on the model.
This replaces `Model.__validators__` and `Model.__root_validators__` from Pydantic V1.
__pydantic_generic_metadata__: Metadata for generic models; contains data used for a similar purpose to
__args__, __origin__, __parameters__ in typing-module generics. May eventually be replaced by these.
__pydantic_parent_namespace__: Parent namespace of the model, used for automatic rebuilding of models.
__pydantic_post_init__: The name of the post-init method for the model, if defined.
__pydantic_root_model__: Whether the model is a [`RootModel`][pydantic.root_model.RootModel].
__pydantic_serializer__: The `pydantic-core` `SchemaSerializer` used to dump instances of the model.
__pydantic_validator__: The `pydantic-core` `SchemaValidator` used to validate instances of the model.
__pydantic_fields__: A dictionary of field names and their corresponding [`FieldInfo`][pydantic.fields.FieldInfo] objects.
__pydantic_computed_fields__: A dictionary of computed field names and their corresponding [`ComputedFieldInfo`][pydantic.fields.ComputedFieldInfo] objects.
__pydantic_extra__: A dictionary containing extra values, if [`extra`][pydantic.config.ConfigDict.extra]
is set to `'allow'`.
__pydantic_fields_set__: The names of fields explicitly set during instantiation.
__pydantic_private__: Values of private attributes set on the model instance.
21class RepairTaskRun(Task, parent_of={}): 22 def __init__(self, original_task: Task): 23 # Keep the typechecker happy 24 tmp_project = Project(name="Repair") 25 super().__init__( 26 name="Repair", 27 parent=tmp_project, 28 description="Repair a task run, given feedback from an evaluator about how the response can be improved.", 29 instruction="You are an assistant which helps improve output from another assistant (original assistant). You'll be provided a task that the original assistant executed (prompt), \ 30the input it was given, and the output it generated. An evaluator has determined that the output it generated did not satisfy the task and should be improved. The evaluator will provide \ 31feedback describing what should be improved. Your job is to understand the evaluator's feedback and improve the response.", 32 requirements=[ 33 TaskRequirement( 34 name="Follow Eval Feedback", 35 instruction="The evaluator's feedback is the most important thing to consider. If it conflicts with the original task instruction or prompt, prioritize the evaluator's feedback.", 36 priority=Priority.p0, 37 ) 38 ], 39 input_json_schema=json.dumps(RepairTaskInput.model_json_schema()), 40 output_json_schema=original_task.output_json_schema, 41 ) 42 43 @classmethod 44 def _original_prompt(cls, run: TaskRun, task: Task) -> str: 45 if run.output.source is None or run.output.source.properties is None: 46 raise ValueError("No source properties found") 47 48 # Get the prompt builder id. Need the second check because we used to store this in a prompt_builder_name field, so loading legacy runs will need this. 49 prompt_id = run.output.source.properties.get( 50 "prompt_id" 51 ) or run.output.source.properties.get("prompt_builder_name", None) 52 if prompt_id is not None and isinstance(prompt_id, str): 53 prompt_builder = prompt_builder_from_id(prompt_id, task) 54 if isinstance(prompt_builder, BasePromptBuilder): 55 return prompt_builder.build_prompt(include_json_instructions=False) 56 57 raise ValueError(f"Prompt builder '{prompt_id}' is not a valid prompt builder") 58 59 @classmethod 60 def build_repair_task_input( 61 cls, original_task: Task, task_run: TaskRun, evaluator_feedback: str 62 ) -> RepairTaskInput: 63 original_prompt = cls._original_prompt(task_run, original_task) 64 return RepairTaskInput( 65 original_prompt=original_prompt, 66 original_input=task_run.input, 67 original_output=task_run.output.output, 68 evaluator_feedback=evaluator_feedback, 69 )
Represents a specific task to be performed, with associated requirements and validation rules.
Contains the task definition, requirements, input/output schemas, and maintains a collection of task runs.
22 def __init__(self, original_task: Task): 23 # Keep the typechecker happy 24 tmp_project = Project(name="Repair") 25 super().__init__( 26 name="Repair", 27 parent=tmp_project, 28 description="Repair a task run, given feedback from an evaluator about how the response can be improved.", 29 instruction="You are an assistant which helps improve output from another assistant (original assistant). You'll be provided a task that the original assistant executed (prompt), \ 30the input it was given, and the output it generated. An evaluator has determined that the output it generated did not satisfy the task and should be improved. The evaluator will provide \ 31feedback describing what should be improved. Your job is to understand the evaluator's feedback and improve the response.", 32 requirements=[ 33 TaskRequirement( 34 name="Follow Eval Feedback", 35 instruction="The evaluator's feedback is the most important thing to consider. If it conflicts with the original task instruction or prompt, prioritize the evaluator's feedback.", 36 priority=Priority.p0, 37 ) 38 ], 39 input_json_schema=json.dumps(RepairTaskInput.model_json_schema()), 40 output_json_schema=original_task.output_json_schema, 41 )
Create a new model by parsing and validating input data from keyword arguments.
Raises [ValidationError
][pydantic_core.ValidationError] if the input data cannot be
validated to form a valid model.
self
is explicitly positional-only to allow self
as a field name.
59 @classmethod 60 def build_repair_task_input( 61 cls, original_task: Task, task_run: TaskRun, evaluator_feedback: str 62 ) -> RepairTaskInput: 63 original_prompt = cls._original_prompt(task_run, original_task) 64 return RepairTaskInput( 65 original_prompt=original_prompt, 66 original_input=task_run.input, 67 original_output=task_run.output.output, 68 evaluator_feedback=evaluator_feedback, 69 )
Configuration for the model, should be a dictionary conforming to [ConfigDict
][pydantic.config.ConfigDict].
122 def wrapped_model_post_init(self: BaseModel, context: Any, /) -> None: 123 """We need to both initialize private attributes and call the user-defined model_post_init 124 method. 125 """ 126 init_private_attributes(self, context) 127 original_model_post_init(self, context)
We need to both initialize private attributes and call the user-defined model_post_init method.