Skip to content

fix: resolve SessionInput state serialization to cloud evaluation service #6847

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
copybara-service wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: resolve SessionInput state serialization to cloud evaluation service #6847

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
115 changes: 113 additions & 2 deletions agentplatform/_genai/evals.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,56 @@
logger = logging.getLogger("agentplatform_genai.evals")


def _AgentRunConfig_from_vertex(
from_object: Union[dict[str, Any], object],
parent_object: Optional[dict[str, Any]] = None,
) -> dict[str, Any]:
to_object: dict[str, Any] = {}
if getv(from_object, ["sessionInput"]) is not None:
setv(
to_object,
["session_input"],
_SessionInput_from_vertex(getv(from_object, ["sessionInput"]), to_object),
)

if getv(from_object, ["agentEngine"]) is not None:
setv(to_object, ["agent_engine"], getv(from_object, ["agentEngine"]))

if getv(from_object, ["userSimulatorConfig"]) is not None:
setv(
to_object,
["user_simulator_config"],
getv(from_object, ["userSimulatorConfig"]),
)

return to_object


def _AgentRunConfig_to_vertex(
from_object: Union[dict[str, Any], object],
parent_object: Optional[dict[str, Any]] = None,
) -> dict[str, Any]:
to_object: dict[str, Any] = {}
if getv(from_object, ["session_input"]) is not None:
setv(
to_object,
["sessionInput"],
_SessionInput_to_vertex(getv(from_object, ["session_input"]), to_object),
)

if getv(from_object, ["agent_engine"]) is not None:
setv(to_object, ["agentEngine"], getv(from_object, ["agent_engine"]))

if getv(from_object, ["user_simulator_config"]) is not None:
setv(
to_object,
["userSimulatorConfig"],
getv(from_object, ["user_simulator_config"]),
)

return to_object


def _CreateEvaluationItemParameters_to_vertex(
from_object: Union[dict[str, Any], object],
parent_object: Optional[dict[str, Any]] = None,
Expand Down Expand Up @@ -137,6 +187,9 @@ def _CreateEvaluationRunParameters_to_vertex(
[item for item in getv(from_object, ["analysis_configs"])],
)

if getv(from_object, ["dummy_session_input"]) is not None:
_SessionInput_to_vertex(getv(from_object, ["dummy_session_input"]), to_object)

return to_object


Expand Down Expand Up @@ -464,7 +517,13 @@ def _EvaluationRunInferenceConfig_from_vertex(
setv(to_object, ["prompt_template"], getv(from_object, ["promptTemplate"]))

if getv(from_object, ["agentRunConfig"]) is not None:
setv(to_object, ["agent_run_config"], getv(from_object, ["agentRunConfig"]))
setv(
to_object,
["agent_run_config"],
_AgentRunConfig_from_vertex(
getv(from_object, ["agentRunConfig"]), to_object
),
)

if getv(from_object, ["agents"]) is not None:
setv(to_object, ["agent_configs"], getv(from_object, ["agents"]))
Expand All @@ -487,7 +546,13 @@ def _EvaluationRunInferenceConfig_to_vertex(
setv(to_object, ["promptTemplate"], getv(from_object, ["prompt_template"]))

if getv(from_object, ["agent_run_config"]) is not None:
setv(to_object, ["agentRunConfig"], getv(from_object, ["agent_run_config"]))
setv(
to_object,
["agentRunConfig"],
_AgentRunConfig_to_vertex(
getv(from_object, ["agent_run_config"]), to_object
),
)

if getv(from_object, ["agent_configs"]) is not None:
setv(to_object, ["agents"], getv(from_object, ["agent_configs"]))
Expand Down Expand Up @@ -902,6 +967,40 @@ def _RubricBasedMetricSpec_to_vertex(
return to_object


def _SessionInput_from_vertex(
from_object: Union[dict[str, Any], object],
parent_object: Optional[dict[str, Any]] = None,
) -> dict[str, Any]:
to_object: dict[str, Any] = {}
if getv(from_object, ["userId"]) is not None:
setv(to_object, ["user_id"], getv(from_object, ["userId"]))

if getv(from_object, ["sessionState"]) is not None:
setv(to_object, ["state"], getv(from_object, ["sessionState"]))

if getv(from_object, ["parameters", "app_name"]) is not None:
setv(to_object, ["app_name"], getv(from_object, ["parameters", "app_name"]))

return to_object


def _SessionInput_to_vertex(
from_object: Union[dict[str, Any], object],
parent_object: Optional[dict[str, Any]] = None,
) -> dict[str, Any]:
to_object: dict[str, Any] = {}
if getv(from_object, ["user_id"]) is not None:
setv(to_object, ["userId"], getv(from_object, ["user_id"]))

if getv(from_object, ["state"]) is not None:
setv(to_object, ["sessionState"], getv(from_object, ["state"]))

if getv(from_object, ["app_name"]) is not None:
setv(to_object, ["parameters", "app_name"], getv(from_object, ["app_name"]))

return to_object


def _UnifiedMetric_from_vertex(
from_object: Union[dict[str, Any], object],
parent_object: Optional[dict[str, Any]] = None,
Expand Down Expand Up @@ -1174,6 +1273,10 @@ def _create_evaluation_run(
] = None,
config: Optional[types.CreateEvaluationRunConfigOrDict] = None,
analysis_configs: Optional[list[types.AnalysisConfigOrDict]] = None,
dummy_session_input: Optional[types.evals.SessionInputOrDict] = None,
dummy_user_simulator_config: Optional[
types.evals.UserSimulatorConfigOrDict
] = None,
) -> types.EvaluationRun:
"""
Creates an EvaluationRun.
Expand All @@ -1188,6 +1291,8 @@ def _create_evaluation_run(
inference_configs=inference_configs,
config=config,
analysis_configs=analysis_configs,
dummy_session_input=dummy_session_input,
dummy_user_simulator_config=dummy_user_simulator_config,
)

request_url_dict: Optional[dict[str, str]]
Expand Down Expand Up @@ -3321,6 +3426,10 @@ async def _create_evaluation_run(
] = None,
config: Optional[types.CreateEvaluationRunConfigOrDict] = None,
analysis_configs: Optional[list[types.AnalysisConfigOrDict]] = None,
dummy_session_input: Optional[types.evals.SessionInputOrDict] = None,
dummy_user_simulator_config: Optional[
types.evals.UserSimulatorConfigOrDict
] = None,
) -> types.EvaluationRun:
"""
Creates an EvaluationRun.
Expand All @@ -3335,6 +3444,8 @@ async def _create_evaluation_run(
inference_configs=inference_configs,
config=config,
analysis_configs=analysis_configs,
dummy_session_input=dummy_session_input,
dummy_user_simulator_config=dummy_user_simulator_config,
)

request_url_dict: Optional[dict[str, str]]
Expand Down
16 changes: 14 additions & 2 deletions agentplatform/_genai/types/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2560,13 +2560,13 @@ class AgentRunConfig(_common.BaseModel):
class AgentRunConfigDict(TypedDict, total=False):
"""Configuration for an Agent Run."""

session_input: Optional[evals_types.SessionInput]
session_input: Optional[evals_types.SessionInputDict]
"""The session input to get agent running results."""

agent_engine: Optional[str]
"""The resource name of the Agent Engine."""

user_simulator_config: Optional[evals_types.UserSimulatorConfig]
user_simulator_config: Optional[evals_types.UserSimulatorConfigDict]
"""Used for multi-turn agent run.
Contains configuration for a user simulator that
uses an LLM to generate messages on behalf of the user."""
Expand Down Expand Up @@ -2765,6 +2765,12 @@ class _CreateEvaluationRunParameters(_common.BaseModel):
analysis_configs: Optional[list[AnalysisConfig]] = Field(
default=None, description=""""""
)
dummy_session_input: Optional[evals_types.SessionInput] = Field(
default=None, description=""""""
)
dummy_user_simulator_config: Optional[evals_types.UserSimulatorConfig] = Field(
default=None, description=""""""
)


class _CreateEvaluationRunParametersDict(TypedDict, total=False):
Expand Down Expand Up @@ -2794,6 +2800,12 @@ class _CreateEvaluationRunParametersDict(TypedDict, total=False):
analysis_configs: Optional[list[AnalysisConfigDict]]
""""""

dummy_session_input: Optional[evals_types.SessionInputDict]
""""""

dummy_user_simulator_config: Optional[evals_types.UserSimulatorConfigDict]
""""""


_CreateEvaluationRunParametersOrDict = Union[
_CreateEvaluationRunParameters, _CreateEvaluationRunParametersDict
Expand Down
158 changes: 79 additions & 79 deletions agentplatform/_genai/types/evals.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,85 @@ class Importance(_common.CaseInSensitiveEnum):
"""Low importance."""


class SessionInput(_common.BaseModel):
"""This field is experimental and may change in future versions.

Input to initialize a session and run an agent, used for agent evaluation.
"""

user_id: Optional[str] = Field(default=None, description="""The user id.""")
state: Optional[dict[str, str]] = Field(
default=None, description="""The state of the session."""
)
app_name: Optional[str] = Field(
default=None,
description="""The name of the app, used for local ADK agent run Runner and Session.""",
)


class SessionInputDict(TypedDict, total=False):
"""This field is experimental and may change in future versions.

Input to initialize a session and run an agent, used for agent evaluation.
"""

user_id: Optional[str]
"""The user id."""

state: Optional[dict[str, str]]
"""The state of the session."""

app_name: Optional[str]
"""The name of the app, used for local ADK agent run Runner and Session."""


SessionInputOrDict = Union[SessionInput, SessionInputDict]


class UserSimulatorConfig(_common.BaseModel):
"""Configuration for a user simulator.

Uses an LLM to generate multi-turn messages that simulate a user.
"""

model_name: Optional[str] = Field(
default=None,
description="""The model name to get next user message for multi-turn agent run.""",
)
model_configuration: Optional[genai_types.GenerateContentConfig] = Field(
default=None, description="""The configuration for the model."""
)
max_turn: Optional[int] = Field(
default=None,
description="""Maximum number of invocations allowed by the multi-turn agent
running. This property allows us to stop a run-off conversation
where the agent and the user simulator get into a never ending loop.
The initial fixed prompt is also counted as an invocation.""",
)


class UserSimulatorConfigDict(TypedDict, total=False):
"""Configuration for a user simulator.

Uses an LLM to generate multi-turn messages that simulate a user.
"""

model_name: Optional[str]
"""The model name to get next user message for multi-turn agent run."""

model_configuration: Optional[genai_types.GenerateContentConfigDict]
"""The configuration for the model."""

max_turn: Optional[int]
"""Maximum number of invocations allowed by the multi-turn agent
running. This property allows us to stop a run-off conversation
where the agent and the user simulator get into a never ending loop.
The initial fixed prompt is also counted as an invocation."""


UserSimulatorConfigOrDict = Union[UserSimulatorConfig, UserSimulatorConfigDict]


class AgentConfig(_common.BaseModel):
"""Represents configuration for an Agent."""

Expand Down Expand Up @@ -456,41 +535,6 @@ class AgentInfoDict(TypedDict, total=False):
AgentInfoOrDict = Union[AgentInfo, AgentInfoDict]


class SessionInput(_common.BaseModel):
"""This field is experimental and may change in future versions.

Input to initialize a session and run an agent, used for agent evaluation.
"""

user_id: Optional[str] = Field(default=None, description="""The user id.""")
state: Optional[dict[str, str]] = Field(
default=None, description="""The state of the session."""
)
app_name: Optional[str] = Field(
default=None,
description="""The name of the app, used for local ADK agent run Runner and Session.""",
)


class SessionInputDict(TypedDict, total=False):
"""This field is experimental and may change in future versions.

Input to initialize a session and run an agent, used for agent evaluation.
"""

user_id: Optional[str]
"""The user id."""

state: Optional[dict[str, str]]
"""The state of the session."""

app_name: Optional[str]
"""The name of the app, used for local ADK agent run Runner and Session."""


SessionInputOrDict = Union[SessionInput, SessionInputDict]


class UserScenario(_common.BaseModel):
"""User scenario to help simulate multi-turn agent run results."""

Expand Down Expand Up @@ -586,50 +630,6 @@ class UserScenarioGenerationConfigDict(TypedDict, total=False):
]


class UserSimulatorConfig(_common.BaseModel):
"""Configuration for a user simulator.

Uses an LLM to generate multi-turn messages that simulate a user.
"""

model_name: Optional[str] = Field(
default=None,
description="""The model name to get next user message for multi-turn agent run.""",
)
model_configuration: Optional[genai_types.GenerateContentConfig] = Field(
default=None, description="""The configuration for the model."""
)
max_turn: Optional[int] = Field(
default=None,
description="""Maximum number of invocations allowed by the multi-turn agent
running. This property allows us to stop a run-off conversation
where the agent and the user simulator get into a never ending loop.
The initial fixed prompt is also counted as an invocation.""",
)


class UserSimulatorConfigDict(TypedDict, total=False):
"""Configuration for a user simulator.

Uses an LLM to generate multi-turn messages that simulate a user.
"""

model_name: Optional[str]
"""The model name to get next user message for multi-turn agent run."""

model_configuration: Optional[genai_types.GenerateContentConfigDict]
"""The configuration for the model."""

max_turn: Optional[int]
"""Maximum number of invocations allowed by the multi-turn agent
running. This property allows us to stop a run-off conversation
where the agent and the user simulator get into a never ending loop.
The initial fixed prompt is also counted as an invocation."""


UserSimulatorConfigOrDict = Union[UserSimulatorConfig, UserSimulatorConfigDict]


class Event(_common.BaseModel):
"""Represents an event in a conversation between agents and users.

Expand Down
Loading