Watch
1
0
Fork
You've already forked arc
0
forked from hesabix/arc
arc/hesabixAPI/app/services/ai/ai_history_summarizer.py

153 lines
4.5 KiB
Python

"""
خلاصه‌سازی تاریخچه مکالمه برای کاهش context.
"""
from __future__ import annotations
import json
import logging
from typing import Any, Dict, List, Optional
from sqlalchemy.orm import Session
from app.services.ai.ai_language_prompt import (
build_history_summary_system_prompt,
detect_message_language,
)
from app.services.ai.prompt_service import get_prompt_by_key
logger = logging.getLogger(__name__)
MAX_SUMMARY_CHARS = 3500
def _message_plain_text(msg: Dict[str, Any]) -> str:
content = msg.get("content")
if isinstance(content, str) and content.strip():
return content.strip()
if msg.get("tool_calls"):
return "[فراخوانی ابزار]"
return ""
def _resolve_summary_language(
messages: List[Dict[str, Any]],
*,
language: Optional[str] = None,
) -> str:
if language in ("fa", "en"):
return language
for msg in reversed(messages):
if msg.get("role") != "user":
continue
detected = detect_message_language(_message_plain_text(msg))
if detected:
return detected
return "fa"
def build_rule_based_history_summary(messages: List[Dict[str, Any]]) -> str:
"""خلاصهٔ سریع بدون LLM — برای fallback."""
lines: List[str] = []
for msg in messages:
role = msg.get("role")
if role not in ("user", "assistant"):
continue
text = _message_plain_text(msg)
if not text or text.startswith("{"):
continue
label = "کاربر" if role == "user" else "دستیار"
snippet = text.replace("\n", " ").strip()[:280]
lines.append(f"- {label}: {snippet}")
if not lines:
return ""
body = "\n".join(lines[-40:])
if len(body) > MAX_SUMMARY_CHARS:
body = body[-MAX_SUMMARY_CHARS:]
return body
def summarize_history_with_llm_detailed(
provider: Any,
model: str,
messages: List[Dict[str, Any]],
*,
db: Optional[Session] = None,
max_output_tokens: int = 500,
language: Optional[str] = None,
) -> tuple[Optional[str], Optional[Dict[str, Any]]]:
"""خلاصهٔ فشرده با LLM؛ usage را هم برمی‌گرداند تا در صورتحساب لحاظ شود."""
draft = build_rule_based_history_summary(messages)
if not draft:
return None, None
lang = _resolve_summary_language(messages, language=language)
system_content = build_history_summary_system_prompt(lang)
if db is not None:
try:
seeded = get_prompt_by_key(db, "aux.history_summary")
if seeded and lang == "fa":
system_content = seeded
except Exception:
pass
user_prefix = (
"Conversation text:\n\n"
if lang == "en"
else "متن مکالمه:\n\n"
)
prompt_messages = [
{"role": "system", "content": system_content},
{
"role": "user",
"content": f"{user_prefix}{draft[:12000]}",
},
]
try:
result = provider.chat_completion(
messages=prompt_messages,
model=model,
max_tokens=max_output_tokens,
temperature=0.2,
tools=None,
)
content = (result.get("message") or {}).get("content") or ""
text = str(content).strip()
if not text:
return None, None
usage = result.get("usage") if isinstance(result, dict) else None
return text[:MAX_SUMMARY_CHARS], usage if isinstance(usage, dict) else None
except Exception as exc:
logger.warning("summarize_history_with_llm failed: %s", exc)
return None, None
def summarize_history_with_llm(
provider: Any,
model: str,
messages: List[Dict[str, Any]],
*,
db: Optional[Session] = None,
max_output_tokens: int = 500,
language: Optional[str] = None,
) -> Optional[str]:
"""خلاصهٔ فشرده با یک فراخوانی LLM."""
text, _usage = summarize_history_with_llm_detailed(
provider,
model,
messages,
db=db,
max_output_tokens=max_output_tokens,
language=language,
)
return text
def messages_from_db_rows(db_messages: List[Any]) -> List[Dict[str, Any]]:
"""تبدیل پیام‌های DB به dict ساده برای extract_facts."""
out: List[Dict[str, Any]] = []
for msg in db_messages:
role = msg.role if isinstance(msg.role, str) else getattr(msg.role, "value", str(msg.role))
out.append({"role": role, "content": (msg.content or "").strip()})
return out