16 lines
597 B
Python
16 lines
597 B
Python
"""多模态大模型 OCR 引擎：直接看图转写题目为 Markdown。"""
|
|
from app.services.llm.client import LLMClient
|
|
from app.services.llm.prompts import VLM_OCR_SYSTEM, VLM_OCR_USER
|
|
from app.services.ocr.base import OcrResult
|
|
|
|
|
|
class VlmOcrEngine:
|
|
def __init__(self, llm: LLMClient):
|
|
self._llm = llm
|
|
|
|
async def recognize(self, image_bytes: bytes, mime: str) -> OcrResult:
|
|
result = await self._llm.chat_vision(
|
|
VLM_OCR_SYSTEM, VLM_OCR_USER, image_bytes, mime=mime
|
|
)
|
|
return OcrResult(markdown=result.content, confidence=1.0, engine="vlm")
|