Classes
ReflectionAction
Action to take based on evaluation.Values
| Value | Description |
|---|---|
ACCEPT | Accept the current response |
REVISE | Revise the response based on feedback |
RETRY | Retry with different approach |
CLARIFY | Need clarification from user |
EvaluationCriteria
Criteria for evaluating responses.Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
accuracy | float | - | Accuracy of the information (0-1) |
completeness | float | - | Completeness of the response (0-1) |
relevance | float | - | Relevance to the task (0-1) |
clarity | float | - | Clarity and readability (0-1) |
Functions
overall_score
Calculate overall evaluation score.
Returns:
float: The overall evaluation score
EvaluationResult
Result of response evaluation.Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
criteria | EvaluationCriteria | - | Criteria for evaluating responses |
overall_score | float | - | Overall evaluation score (0-1) |
feedback | str | - | Detailed feedback for improvement |
suggested_improvements | List[str] | - | Specific improvement suggestions |
action | ReflectionAction | - | Recommended action |
confidence | float | - | Confidence in evaluation (0-1) |
Functions
__post_init__
Calculate overall score after initialization.
ReflectionConfig
Configuration for reflection process.Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
max_iterations | int | 3 | Maximum reflection iterations |
acceptance_threshold | float | 0.8 | Minimum score to accept response |
evaluator_model | Optional[str] | None | Model for evaluation |
enable_self_critique | bool | True | Enable self-critique |
enable_improvement_suggestions | bool | True | Enable improvement suggestions |
ReflectionState
State tracking for reflection process.Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
iteration | int | 0 | Current iteration number |
evaluations | List[EvaluationResult] | [] | History of evaluations |
responses | List[str] | [] | History of responses |
final_response | Optional[str] | None | Final accepted response |
terminated_reason | Optional[str] | None | Reason for termination |
Functions
add_evaluation
Add an evaluation result to the state.
Parameters:
response(str): The response textevaluation(EvaluationResult): The evaluation result
get_latest_evaluation
Get the most recent evaluation.
Returns:
Optional[EvaluationResult]: The most recent evaluation result
should_continue
Check if reflection should continue.
Parameters:
config(ReflectionConfig): The reflection configuration
bool: True if reflection should continue
ReflectionPrompts
Prompt templates for reflection process.Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
EVALUATION_PROMPT | str | - | Template for evaluation prompts |
IMPROVEMENT_PROMPT | str | - | Template for improvement prompts |
SELF_CRITIQUE_PROMPT | str | - | Template for self-critique prompts |