Spaces:
Sleeping
Sleeping
import os | |
from typing import Union | |
from fastapi import FastAPI, WebSocket | |
from epub2txt import epub2txt | |
from fastllm_pytools import llm | |
model = llm.model(os.getenv("checkpoint_path")) | |
prompt = os.getenv("prompt") | |
app = FastAPI() | |
async def read_root(*, | |
websocket: WebSocket, | |
f_name: str): | |
await websocket.accept() | |
ch_list = epub2txt(f_name, outputlist=True) | |
chapter_titles = epub2txt.content_titles | |
title = epub2txt.title | |
idx = 0 | |
sm_list = [] | |
for text in ch_list[2:]: | |
idx += 1 | |
docs = [] | |
for i in range(0, len(text)//2000+1, 2000): | |
t = text[i:i+2048] | |
if len(t) > 0: | |
docs.append(model.response(prompt+t)) | |
await websocket.send_text(f"chsum: {docs[-1]}") | |
hist = docs[0] | |
for doc in docs[1:]: | |
hist = model.response(prompt+"\n"+hist+"\n"+doc) | |
await websocket.send_text(f"draft_sum: {hist}") | |
sm_list.append(hist) | |
mdobj_str = f"# {title}\n\n{hist}\n\n\n" | |
for ct, sm in zip(chapter_titles[2:], sm_list): | |
mdobj_str += f"## {ct}\n\n{sm}\n\n\n" | |
await websocket.send_text(f"output: {mdobj_str}") | |
# uvicorn api:app --reload |