81ada1d9b9
## Google File Picker The Google Picker lets users select or upload Google Drive files. Users can grant permission to your apps to access their Drive data, providing a secure and authorized way to interact with their files. The `generate_google_file_picker_url` returns a URL to a Google File Picker for the user.
135 lines
3.7 KiB
Python
135 lines
3.7 KiB
Python
from arcade.sdk import ToolCatalog
|
|
from arcade.sdk.eval import (
|
|
EvalRubric,
|
|
EvalSuite,
|
|
ExpectedToolCall,
|
|
tool_eval,
|
|
)
|
|
from arcade.sdk.eval.critic import BinaryCritic
|
|
|
|
import arcade_google
|
|
from arcade_google.tools import (
|
|
create_contact,
|
|
search_contacts_by_email,
|
|
search_contacts_by_name,
|
|
)
|
|
|
|
# Evaluation rubric
|
|
rubric = EvalRubric(
|
|
fail_threshold=0.9,
|
|
warn_threshold=0.95,
|
|
)
|
|
|
|
catalog = ToolCatalog()
|
|
catalog.add_module(arcade_google)
|
|
|
|
|
|
@tool_eval()
|
|
def contacts_eval_suite() -> EvalSuite:
|
|
"""Create an evaluation suite for Google Contacts tools."""
|
|
suite = EvalSuite(
|
|
name="Google Contacts Tools Evaluation",
|
|
system_message="You are an AI assistant that can manage Google Contacts using the provided tools.",
|
|
catalog=catalog,
|
|
rubric=rubric,
|
|
)
|
|
|
|
suite.add_case(
|
|
name="Search contacts by name",
|
|
user_message="Find my contact Bob",
|
|
expected_tool_calls=[
|
|
ExpectedToolCall(
|
|
func=search_contacts_by_name,
|
|
args={
|
|
"name": "Bob",
|
|
},
|
|
)
|
|
],
|
|
)
|
|
|
|
suite.add_case(
|
|
name="Search contacts by email",
|
|
user_message="Find my contact alice@example.com",
|
|
expected_tool_calls=[
|
|
ExpectedToolCall(
|
|
func=search_contacts_by_email,
|
|
args={
|
|
"email": "alice@example.com",
|
|
},
|
|
)
|
|
],
|
|
)
|
|
|
|
suite.add_case(
|
|
name="Search contacts with query and limit",
|
|
user_message="Find 5 contacts whose names include 'Alice'",
|
|
expected_tool_calls=[
|
|
ExpectedToolCall(
|
|
func=search_contacts_by_name,
|
|
args={
|
|
"name": "Alice",
|
|
"limit": 5,
|
|
},
|
|
)
|
|
],
|
|
critics=[
|
|
BinaryCritic(critic_field="query", weight=0.5),
|
|
BinaryCritic(critic_field="limit", weight=0.5),
|
|
],
|
|
)
|
|
|
|
suite.add_case(
|
|
name="Create new contact with only given name",
|
|
user_message="Create a new contact for Alice",
|
|
expected_tool_calls=[
|
|
ExpectedToolCall(
|
|
func=create_contact,
|
|
args={
|
|
"given_name": "Alice",
|
|
},
|
|
)
|
|
],
|
|
critics=[
|
|
BinaryCritic(critic_field="given_name", weight=1.0),
|
|
],
|
|
)
|
|
|
|
suite.add_case(
|
|
name="Create new contact with only email (infer name from email)",
|
|
user_message="Create a new contact for alice@example.com",
|
|
expected_tool_calls=[
|
|
ExpectedToolCall(
|
|
func=create_contact,
|
|
args={
|
|
"given_name": "Alice",
|
|
"email": "alice@example.com",
|
|
},
|
|
)
|
|
],
|
|
critics=[
|
|
BinaryCritic(critic_field="email", weight=0.5),
|
|
BinaryCritic(critic_field="given_name", weight=0.5),
|
|
],
|
|
)
|
|
|
|
suite.add_case(
|
|
name="Create new contact with full name and email",
|
|
user_message="Create a contact for Bob Smith (bob.smith@example.com)",
|
|
expected_tool_calls=[
|
|
ExpectedToolCall(
|
|
func=create_contact,
|
|
args={
|
|
"given_name": "Bob",
|
|
"family_name": "Smith",
|
|
"email": "bob.smith@example.com",
|
|
},
|
|
)
|
|
],
|
|
critics=[
|
|
BinaryCritic(critic_field="given_name", weight=0.33),
|
|
BinaryCritic(critic_field="family_name", weight=0.33),
|
|
BinaryCritic(critic_field="email", weight=0.34),
|
|
],
|
|
)
|
|
|
|
return suite
|