Merge pull request #450 from lion-agi/dev_new_direct

added new select

lionagi/core/director/models/__init__.py

@@ -0,0 +1,13 @@
+from .action_model import ActionModel
+from .brainstorm_model import BrainstormModel
+from .plan_model import PlanModel
+from .reason_model import ReasonModel
+from .step_model import StepModel
+
+__all__ = [
+    "ReasonModel",
+    "StepModel",
+    "BrainstormModel",
+    "ActionModel",
+    "PlanModel",
+]

lionagi/core/director/models/action_model.py

@@ -0,0 +1,61 @@
+from typing import Any
+
+from lionfuncs import to_dict, validate_str
+from pydantic import BaseModel, Field, field_validator
+
+
+class ActionModel(BaseModel):
+
+    title: str = Field(
+        ...,
+        title="Title",
+        description="Provide a concise title summarizing the action.",
+    )
+    content: str = Field(
+        ...,
+        title="Content",
+        description="Provide a brief description of the action to be performed.",
+    )
+    function: str = Field(
+        ...,
+        title="Function",
+        description=(
+            "Specify the name of the function to execute. **Choose from the provided "
+            "`tool_schema`; do not invent function names.**"
+        ),
+        examples=["print", "add", "len"],
+    )
+    arguments: dict[str, Any] = Field(
+        {},
+        title="Arguments",
+        description=(
+            "Provide the arguments to pass to the function as a dictionary. **Use "
+            "argument names and types as specified in the `tool_schema`; do not "
+            "invent argument names.**"
+        ),
+        examples=[{"num1": 1, "num2": 2}, {"x": "hello", "y": "world"}],
+    )
+
+    @field_validator("title", mode="before")
+    def validate_title(cls, value: Any) -> str:
+        return validate_str(value, "title")
+
+    @field_validator("content", mode="before")
+    def validate_content(cls, value: Any) -> str:
+        return validate_str(value, "content")
+
+    @field_validator("function", mode="before")
+    def validate_function(cls, value: Any) -> str:
+        return validate_str(value, "function")
+
+    @field_validator("arguments", mode="before")
+    def validate_arguments(cls, value: Any) -> dict[str, Any]:
+        return to_dict(
+            value,
+            fuzzy_parse=True,
+            suppress=True,
+            recursive=True,
+        )
+
+
+__all__ = ["ActionModel"]

lionagi/core/director/models/brainstorm_model.py

@@ -0,0 +1,42 @@
+from typing import Any
+
+from lionfuncs import validate_str
+from pydantic import BaseModel, Field, field_validator
+
+from .reason_model import ReasonModel
+from .step_model import StepModel
+
+
+class BrainstormModel(BaseModel):
+
+    title: str = Field(
+        ...,
+        title="Title",
+        description="Provide a concise title summarizing the brainstorming session.",
+    )
+    content: str = Field(
+        ...,
+        title="Content",
+        description="Describe the context or focus of the brainstorming session.",
+    )
+    ideas: list[StepModel] = Field(
+        ...,
+        title="Ideas",
+        description="A list of ideas for the next step, generated during brainstorming.",
+    )
+    reason: ReasonModel = Field(
+        ...,
+        title="Reason",
+        description="Provide the high level reasoning behind the brainstorming session.",
+    )
+
+    @field_validator("title", mode="before")
+    def validate_title(cls, value: Any) -> str:
+        return validate_str(value, "title")
+
+    @field_validator("content", mode="before")
+    def validate_content(cls, value: Any) -> str:
+        return validate_str(value, "content")
+
+
+__all__ = ["BrainstormModel"]

lionagi/core/director/models/plan_model.py

@@ -0,0 +1,51 @@
+from typing import Any, List
+
+from lionfuncs import validate_str
+from pydantic import BaseModel, Field, field_validator
+
+from .reason_model import ReasonModel
+from .step_model import StepModel
+
+
+class PlanModel(BaseModel):
+    """
+    Represents a plan consisting of multiple steps, with an overall reason.
+
+    Attributes:
+        title (str): A concise title summarizing the plan.
+        content (str): A detailed description of the plan.
+        reason (ReasonModel): The overall reasoning behind the plan.
+        steps (List[StepModel]): A list of steps to execute the plan.
+    """
+
+    title: str = Field(
+        ...,
+        title="Title",
+        description="Provide a concise title summarizing the plan.",
+    )
+    content: str = Field(
+        ...,
+        title="Content",
+        description="Provide a detailed description of the plan.",
+    )
+    reason: ReasonModel = Field(
+        ...,
+        title="Reason",
+        description="Provide the reasoning behind the entire plan.",
+    )
+    steps: list[StepModel] = Field(
+        ...,
+        title="Steps",
+        description="A list of steps to execute the plan.",
+    )
+
+    @field_validator("title", mode="before")
+    def validate_title(cls, value: Any) -> str:
+        return validate_str(value, "title")
+
+    @field_validator("content", mode="before")
+    def validate_content(cls, value: Any) -> str:
+        return validate_str(value, "content")
+
+
+__all__ = ["PlanModel"]

lionagi/core/director/models/reason_model.py

@@ -0,0 +1,63 @@
+import logging
+from typing import Any
+
+from lionfuncs import to_num, validate_str
+from pydantic import BaseModel, Field, field_validator
+
+
+class ReasonModel(BaseModel):
+    title: str = Field(
+        ...,
+        title="Title",
+        description="Provide a concise title summarizing the reason.",
+    )
+    content: str = Field(
+        ...,
+        title="Content",
+        description=(
+            "Provide a detailed explanation supporting the reason, including relevant "
+            "information or context."
+        ),
+    )
+    confidence_score: float | None = Field(
+        None,
+        description=(
+            "Provide an objective numeric confidence score between 0 and 1 (with 3 "
+            "decimal places) indicating how likely you successfully achieved the task "
+            "according to user expectation. Interpret the score as:\n"
+            "- **1**: Very confident in a good job.\n"
+            "- **0**: Not confident at all.\n"
+            "- **[0.8, 1]**: You can continue the path of reasoning if needed.\n"
+            "- **[0.5, 0.8)**: Recheck your reasoning and consider reverting to a "
+            "previous, more confident reasoning path.\n"
+            "- **[0, 0.5)**: Stop because the reasoning is starting to be off track."
+        ),
+        examples=[0.821, 0.257, 0.923, 0.439],
+        ge=0,
+        le=1,
+    )
+
+    @field_validator("title", mode="before")
+    def validate_title(cls, value: Any) -> str:
+        return validate_str(value, "title")
+
+    @field_validator("content", mode="before")
+    def validate_content(cls, value: Any) -> str:
+        return validate_str(value, "content")
+
+    @field_validator("confidence_score", mode="before")
+    def validate_confidence_score(cls, value: Any) -> float:
+        try:
+            return to_num(
+                value,
+                upper_bound=1,
+                lower_bound=0,
+                num_type=float,
+                precision=3,
+            )
+        except Exception as e:
+            logging.error(f"Failed to convert {value} to a number. Error: {e}")
+            return 0.0
+
+
+__all__ = ["ReasonModel"]

lionagi/core/director/models/step_model.py

@@ -0,0 +1,65 @@
+import logging
+from typing import Any
+
+from lionfuncs import validate_boolean, validate_str
+from pydantic import BaseModel, Field, field_validator
+
+from .action_model import ActionModel
+from .reason_model import ReasonModel
+
+
+class StepModel(BaseModel):
+    title: str = Field(
+        ...,
+        title="Title",
+        description="Provide a concise title summarizing the step.",
+    )
+    content: str = Field(
+        ...,
+        title="Content",
+        description="Describe the content of the step in detail.",
+    )
+    reason: ReasonModel = Field(
+        ...,
+        title="Reason",
+        description="Provide the reasoning behind this step, including supporting details.",
+    )
+    action_required: bool = Field(
+        False,
+        title="Action Required",
+        description=(
+            "Indicate whether this step requires an action. Set to **True** if an "
+            "action is required; otherwise, set to **False**."
+        ),
+    )
+    actions: list[ActionModel] = Field(
+        [],
+        title="Actions",
+        description=(
+            "List of actions to be performed if `action_required` is **True**. Leave "
+            "empty if no action is required. **When providing actions, you must "
+            "choose from the provided `tool_schema`. Do not invent function or "
+            "argument names.**"
+        ),
+    )
+
+    @field_validator("title", mode="before")
+    def validate_title(cls, value: Any) -> str:
+        return validate_str(value, "title")
+
+    @field_validator("content", mode="before")
+    def validate_content(cls, value: Any) -> str:
+        return validate_str(value, "content")
+
+    @field_validator("action_required", mode="before")
+    def validate_action_required(cls, value: Any) -> bool:
+        try:
+            return validate_boolean(value)
+        except Exception as e:
+            logging.error(
+                f"Failed to convert {value} to a boolean. Error: {e}"
+            )
+            return False
+
+
+__all__ = ["StepModel"]

lionagi/core/director/operations/select.py

@@ -0,0 +1,93 @@
+from __future__ import annotations
+
+from collections.abc import Callable
+from enum import Enum
+
+from lionfuncs import choose_most_similar
+from pydantic import BaseModel
+
+from lionagi.core.director.models import ReasonModel
+from lionagi.core.session.branch import Branch
+
+from .utils import is_enum
+
+PROMPT = "Please select up to {max_num_selections} items from the following list {choices}. Provide the selection(s), and no comments from you"
+
+
+class SelectionModel(BaseModel):
+    selected: list[str | Enum]
+
+
+class ReasonSelectionModel(BaseModel):
+    selected: list[str | Enum]
+    reason: ReasonModel
+
+
+async def select(
+    choices: list[str] | type[Enum],
+    max_num_selections: int = 1,
+    instruction=None,
+    context=None,
+    system=None,
+    sender=None,
+    recipient=None,
+    reason: bool = False,
+    return_enum: bool = False,
+    enum_parser: Callable = None,  # parse the model string response to appropriate type
+    branch: Branch = None,
+    return_pydantic_model=False,
+    **kwargs,  # additional chat arguments
+):
+    selections = []
+    if return_enum and not is_enum(choices):
+        raise ValueError("return_enum can only be True if choices is an Enum")
+
+    if is_enum(choices):
+        selections = [selection.value for selection in choices]
+    else:
+        selections = choices
+
+    prompt = PROMPT.format(
+        max_num_selections=max_num_selections, choices=selections
+    )
+
+    if instruction:
+        prompt = f"{instruction}\n\n{prompt} \n\n "
+
+    branch = branch or Branch()
+    response: SelectionModel | ReasonSelectionModel | str = await branch.chat(
+        instruction=prompt,
+        context=context,
+        system=system,
+        sender=sender,
+        recipient=recipient,
+        pydantic_model=SelectionModel if not reason else ReasonSelectionModel,
+        return_pydantic_model=True,
+        **kwargs,
+    )
+
+    selected = response
+    if isinstance(response, SelectionModel | ReasonSelectionModel):
+        selected = response.selected
+    selected = [selected] if not isinstance(selected, list) else selected
+    corrected_selections = [
+        choose_most_similar(selection, selections) for selection in selected
+    ]
+
+    if return_enum:
+        out = []
+        if not enum_parser:
+            enum_parser = lambda x: x
+        for selection in corrected_selections:
+            selection = enum_parser(selection)
+            for member in choices.__members__.values():
+                if member.value == selection:
+                    out.append(member)
+        corrected_selections = out
+
+    if return_pydantic_model:
+        if not isinstance(response, SelectionModel | ReasonSelectionModel):
+            return SelectionModel(selected=corrected_selections)
+        response.selected = corrected_selections
+        return response
+    return corrected_selections