An open-source document parsing model with SOTA accuracy and speed, using an RSI loop during post-training.
import asyncio
from typing import Dict, Any, Optional
def _build_task_payload(task_id: str) -> Dict[str, Any]:
return {
"record_id": None,
"task_id": task_id,
"status": "queuing",
"file_name": None,
"result": {},
"error": None,
"metadata": {},
}
class MemoryTaskStore:
def __init__(self):
self._store: Dict[str, Dict[str, Any]] = {}
self._lock = asyncio.Lock()
async def create(self, task_id: str, file_name: Optional[str] = None) -> None:
async with self._lock:
self._store[task_id] = _build_task_payload(task_id)
self._store[task_id]["file_name"] = file_name
async def set_processing(self, task_id: str) -> None:
async with self._lock:
if task_id in self._store:
self._store[task_id]["status"] = "processing"
async def set_success(self, task_id: str, result: Dict[str, Any]) -> None:
async with self._lock:
if task_id in self._store:
self._store[task_id].update(
status="success",
file_name=result.get("file_name", self._store[task_id].get("file_name")),
result=result.get("result", {}),
metadata=result.get("metadata", {}),
error=None,
)
async def set_failed(self, task_id: str, error: str) -> None:
async with self._lock:
if task_id in self._store:
self._store[task_id].update(status="failed", error=error)
async def get(self, task_id: str) -> Optional[Dict[str, Any]]:
async with self._lock:
return self._store.get(task_id)