mirror of
https://github.com/Colton-z/AstraBox.git
synced 2026-09-28 06:02:56 +08:00
Colton QiandClaude
5ed0137859
Strengthen sandbox isolation and authentication, make all five engines work through the bundled installer, and preserve conversations across sandbox and service restarts. Add team login and single-container deployment, with upgrade instructions for replacing existing 0.1.0 sandboxes. Co-Authored-By: Claude <noreply@anthropic.com>
54 lines
1.9 KiB
Python
54 lines
1.9 KiB
Python
"""The label output cap holds reasoning on routes that cannot turn it off.
|
|
|
|
Titles, title decisions and process summaries send ``reasoning_effort:
|
|
"none"``, but an Anthropic- or OpenAI-compatible reasoning model may think
|
|
anyway, and every thinking token counts against the cap. On DeepSeek's
|
|
thinking route a 256-token cap lost 20 of 80 titles and 1024 lost 1; 2048 lost
|
|
none. The reason lives beside ``TITLE_MODEL_MAX_TOKENS`` in
|
|
``astrabox.common.utils.settings``, the one definition: the settings default,
|
|
the title service and the env registry row all read it, and
|
|
``docs/configuration.md`` is generated from the registry.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
from types import SimpleNamespace
|
|
|
|
import httpx
|
|
|
|
from astrabox.common.utils.settings import TITLE_MODEL_MAX_TOKENS
|
|
from astrabox.core.service.orchestrator.session_title_service import (
|
|
ConversationTitleModel,
|
|
TitleModelRequestConfig,
|
|
)
|
|
|
|
|
|
def test_the_label_cap_covers_measured_title_reasoning() -> None:
|
|
assert TITLE_MODEL_MAX_TOKENS == 2048
|
|
|
|
|
|
async def test_the_cap_reaches_the_request() -> None:
|
|
bodies: list[dict] = []
|
|
|
|
def respond(request: httpx.Request) -> httpx.Response:
|
|
bodies.append(json.loads(request.content))
|
|
content = json.dumps({"title": "Label cap"})
|
|
return httpx.Response(200, json={"choices": [{"message": {"content": content}}]})
|
|
|
|
model = ConversationTitleModel(
|
|
settings=SimpleNamespace(title_model_request_timeout_seconds=60),
|
|
http_client_factory=lambda **kwargs: httpx.AsyncClient(
|
|
transport=httpx.MockTransport(respond), **kwargs
|
|
),
|
|
)
|
|
await model.generate_title(
|
|
user_text="Name this",
|
|
assistant_text="Named",
|
|
request_config=TitleModelRequestConfig(
|
|
base_url="https://gateway.test", api_key="test-only", model_name="label-model"
|
|
),
|
|
)
|
|
|
|
assert bodies[0]["max_tokens"] == TITLE_MODEL_MAX_TOKENS
|