Buckets:

download
raw
3.83 kB
# tests/test_regression_cost_model.py
"""Tests for veronica.regression_cost_model -- EMA-based cost estimation."""
from __future__ import annotations
import pytest
from veronica.regression_cost_model import RegressionCostModel
from veronica.types import HistoryView, StepIntent
def _intent(model: str = "gpt-4") -> StepIntent:
return StepIntent(
step_id="s1",
request_id="r1",
chain_id="c1",
kind="llm",
model=model,
tool_name=None,
timeout_ms=30000,
metadata={},
)
def _history(
depth: int = 10,
cost_ema_by_model: dict | None = None,
) -> HistoryView:
return HistoryView(
chain_id="c1",
last_n=(),
rolling_cost_usd=0.0,
failure_streak=0,
depth=depth,
loop_score=0.0,
cost_per_step_ema_by_model=cost_ema_by_model or {},
)
class TestRegressionCostModel:
def test_uses_ema_when_available(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent("gpt-4"),
_history(depth=10, cost_ema_by_model={"gpt-4": 0.05}),
None,
)
assert result.estimated_usd == pytest.approx(0.05)
assert result.basis == "historical"
assert result.confidence == 0.75 # depth=10 -> 0.75
def test_graduated_confidence_low(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent(),
_history(depth=3, cost_ema_by_model={"gpt-4": 0.05}),
None,
)
assert result.confidence == 0.60
def test_graduated_confidence_mid(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent(),
_history(depth=10, cost_ema_by_model={"gpt-4": 0.05}),
None,
)
assert result.confidence == 0.75
def test_graduated_confidence_high(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent(),
_history(depth=25, cost_ema_by_model={"gpt-4": 0.05}),
None,
)
assert result.confidence == 0.85
def test_fallback_to_pricing_table(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent("gpt-4"),
_history(depth=0, cost_ema_by_model={}),
None,
)
assert result.basis == "pricing_table"
assert result.confidence == 0.2
assert result.estimated_usd > 0
def test_unknown_model_fallback(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent("totally-unknown-model"),
_history(cost_ema_by_model={}),
None,
)
assert result.estimated_usd > 0
assert result.basis == "pricing_table"
def test_zero_ema_uses_fallback(self) -> None:
model = RegressionCostModel()
result = model.estimate(
_intent(),
_history(cost_ema_by_model={"gpt-4": 0.0}),
None,
)
assert result.basis == "pricing_table"
def test_tool_intent(self) -> None:
model = RegressionCostModel()
intent = StepIntent(
step_id="s1",
request_id="r1",
chain_id="c1",
kind="tool",
model=None,
tool_name="web_search",
timeout_ms=30000,
metadata={},
)
result = model.estimate(intent, _history(), None)
assert result.estimated_usd == 0.0
assert result.basis == "pricing_table"
def test_protocol_compatible(self) -> None:
from veronica.protocols import CostModelProtocol
model = RegressionCostModel()
assert isinstance(model, CostModelProtocol)

Xet Storage Details

Size:
3.83 kB
·
Xet hash:
8902b883e9b956fa6208c0240a0cd81448aa2d56d3a47e69803b6bd9d35ad2da

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.