Spaces:
Running
Running
SilverElixir commited on
Commit ·
e980e7c
1
Parent(s): ebbf46e
Add the thought-for block to non-streamed fallback answers in private chats
Browse files- lumen_routes.py +13 -2
- tests/test_bot_routes.py +27 -0
lumen_routes.py
CHANGED
|
@@ -16,7 +16,7 @@ from datetime import date, datetime
|
|
| 16 |
from typing import Any, Callable
|
| 17 |
|
| 18 |
import aiohttp
|
| 19 |
-
from aiogram.enums import ParseMode
|
| 20 |
from aiogram.types import Message
|
| 21 |
from google.genai import types
|
| 22 |
|
|
@@ -39,6 +39,7 @@ from lumen_router_config import (
|
|
| 39 |
_looks_like_heavy_query,
|
| 40 |
)
|
| 41 |
from lumen_security import _scrub_identity_leak, _detect_garbled_mix, _is_garbled_echo
|
|
|
|
| 42 |
|
| 43 |
# Транскрипт длинного войса режем сверху — иначе маршрут ниже упрётся в лимиты моделей.
|
| 44 |
_TRANSCRIPT_MAX_CHARS = 4000
|
|
@@ -1005,6 +1006,7 @@ async def _run_route(
|
|
| 1005 |
import bot
|
| 1006 |
if not route:
|
| 1007 |
raise RuntimeError("Empty route — no model to choose from.")
|
|
|
|
| 1008 |
# Порядок внутри провайдера — по измеренным задержкам; сами блоки и защита квоты Gemini не трогаем.
|
| 1009 |
route = _reorder_route_by_speed(route)
|
| 1010 |
if deadline is None:
|
|
@@ -1110,7 +1112,16 @@ async def _run_route(
|
|
| 1110 |
fits_one_message = len(_split_text_chunks(ans, bot.TG_MAX_LEN)) == 1
|
| 1111 |
reused = False
|
| 1112 |
if fits_one_message:
|
| 1113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1114 |
if html_text is not None:
|
| 1115 |
edited = await bot._tg_call(
|
| 1116 |
reusable_placeholder.edit_text, html_text,
|
|
|
|
| 16 |
from typing import Any, Callable
|
| 17 |
|
| 18 |
import aiohttp
|
| 19 |
+
from aiogram.enums import ChatType, ParseMode
|
| 20 |
from aiogram.types import Message
|
| 21 |
from google.genai import types
|
| 22 |
|
|
|
|
| 39 |
_looks_like_heavy_query,
|
| 40 |
)
|
| 41 |
from lumen_security import _scrub_identity_leak, _detect_garbled_mix, _is_garbled_echo
|
| 42 |
+
from lumen_status import thought_block
|
| 43 |
|
| 44 |
# Транскрипт длинного войса режем сверху — иначе маршрут ниже упрётся в лимиты моделей.
|
| 45 |
_TRANSCRIPT_MAX_CHARS = 4000
|
|
|
|
| 1006 |
import bot
|
| 1007 |
if not route:
|
| 1008 |
raise RuntimeError("Empty route — no model to choose from.")
|
| 1009 |
+
route_started = time.monotonic()
|
| 1010 |
# Порядок внутри провайдера — по измеренным задержкам; сами блоки и защита квоты Gemini не трогаем.
|
| 1011 |
route = _reorder_route_by_speed(route)
|
| 1012 |
if deadline is None:
|
|
|
|
| 1112 |
fits_one_message = len(_split_text_chunks(ans, bot.TG_MAX_LEN)) == 1
|
| 1113 |
reused = False
|
| 1114 |
if fits_one_message:
|
| 1115 |
+
if status is not None:
|
| 1116 |
+
html_text, plain_text = status.compose_answer(ans)
|
| 1117 |
+
elif getattr(getattr(message, "chat", None), "type", None) == ChatType.PRIVATE:
|
| 1118 |
+
# Обычный текст после раннего сбоя стрима: статуса нет,
|
| 1119 |
+
# но «Thought for Ns» над ответом всё равно показываем.
|
| 1120 |
+
html_text, plain_text = thought_block(
|
| 1121 |
+
bot._chat_lang(chat_id), int(time.monotonic() - route_started), ans,
|
| 1122 |
+
)
|
| 1123 |
+
else:
|
| 1124 |
+
html_text, plain_text = None, ans
|
| 1125 |
if html_text is not None:
|
| 1126 |
edited = await bot._tg_call(
|
| 1127 |
reusable_placeholder.edit_text, html_text,
|
tests/test_bot_routes.py
CHANGED
|
@@ -747,6 +747,33 @@ def test_run_route_reuses_stream_placeholder_when_fallback_succeeds():
|
|
| 747 |
bot.ask_openrouter_text = original_or_text
|
| 748 |
|
| 749 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 750 |
def test_run_route_deletes_orphaned_placeholder_when_whole_route_fails():
|
| 751 |
chat_id = 999308
|
| 752 |
placeholder = _FakeSentMessage()
|
|
|
|
| 747 |
bot.ask_openrouter_text = original_or_text
|
| 748 |
|
| 749 |
|
| 750 |
+
def test_run_route_reuse_adds_thought_block_in_private_chat(monkeypatch):
|
| 751 |
+
# Обычный текст после раннего сбоя стрима не должен терять «Thought for Ns»:
|
| 752 |
+
# статуса в этом пути нет, блок строим по времени маршрута.
|
| 753 |
+
chat_id = 999322
|
| 754 |
+
placeholder = _FakeSentMessage()
|
| 755 |
+
incoming = _FakeIncomingMessage(chat_id)
|
| 756 |
+
|
| 757 |
+
async def fake_stream_fail(cid, prompt, message, model_id, deadline=None, status=None):
|
| 758 |
+
return None, placeholder
|
| 759 |
+
|
| 760 |
+
async def fake_text(cid, prompt, model_chain, deadline=None):
|
| 761 |
+
return "Ответ от резервной модели"
|
| 762 |
+
|
| 763 |
+
monkeypatch.setattr(bot, "_try_openrouter_streaming", fake_stream_fail)
|
| 764 |
+
monkeypatch.setattr(bot, "ask_openrouter_text", fake_text)
|
| 765 |
+
route = [("openrouter", "meta-llama/llama-3.3-70b-instruct:free"),
|
| 766 |
+
("openrouter", "openai/gpt-oss-20b:free")]
|
| 767 |
+
ans, sent = asyncio.run(bot._run_route(chat_id, "привет", route, message=incoming, allow_stream=True))
|
| 768 |
+
assert ans == "Ответ от резервной модели"
|
| 769 |
+
assert sent is True
|
| 770 |
+
final_text = placeholder.edits[-1][0]
|
| 771 |
+
assert final_text.startswith("<blockquote expandable>")
|
| 772 |
+
assert "Thought for" in final_text
|
| 773 |
+
assert "Ответ от резервной модели" in final_text
|
| 774 |
+
bot.chat_state.pop(chat_id, None)
|
| 775 |
+
|
| 776 |
+
|
| 777 |
def test_run_route_deletes_orphaned_placeholder_when_whole_route_fails():
|
| 778 |
chat_id = 999308
|
| 779 |
placeholder = _FakeSentMessage()
|