backend/models/llm.py

"""MiniCPM4.1-8B LLM client — runs on Modal GPU."""

from typing import Iterator, List, Optional

from models.modal_client import get_llm


class MiniCPMLLM:
    def __init__(self):
        self._remote = get_llm()

    def stream_answer(
        self,
        query: str,
        context: str,
        chat_history: Optional[List[dict]] = None,
    ) -> Iterator[str]:
        for token in self._remote.stream_answer.remote_gen(
            query, context, chat_history or []
        ):
            yield token

    def generate_summary(self, document_text: str, summary_type: str = "financial") -> str:
        return self._remote.generate_summary.remote(document_text, summary_type)

    def evaluate_confidence(self, query: str, context: str, answer: str) -> float:
        return self._remote.evaluate_confidence.remote(query, context, answer)

    def extract_entities(self, document_text: str) -> dict:
        return self._remote.extract_entities.remote(document_text)