main.py 2.9 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697
  1. """FastAPI 应用入口:挂载路由、配置 CORS、启动 uvicorn
  2. 启动方式:
  3. python run_backend.py
  4. 或
  5. uvicorn backend.api.main:app --host 0.0.0.0 --port 8000 --reload
  6. """
  7. from __future__ import annotations
  8. import logging
  9. from fastapi import FastAPI
  10. from fastapi.middleware.cors import CORSMiddleware
  11. from backend.api.routes import router
  12. # 配置日志
  13. logging.basicConfig(
  14. level=logging.INFO,
  15. format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
  16. )
  17. logger = logging.getLogger(__name__)
  18. app = FastAPI(
  19. title="Paper Innovation Agent API",
  20. description="论文创新 Agent 异步接口:提交分析、SSE 进度推送、报告查询",
  21. version="1.0.0",
  22. )
  23. # 配置 CORS:允许前端 localhost:5173 访问
  24. app.add_middleware(
  25. CORSMiddleware,
  26. allow_origins=[
  27. "http://localhost:5173",
  28. "http://127.0.0.1:5173",
  29. ],
  30. allow_credentials=True,
  31. allow_methods=["*"],
  32. allow_headers=["*"],
  33. )
  34. # 挂载路由
  35. app.include_router(router)
  36. @app.on_event("startup")
  37. async def on_startup() -> None:
  38. """应用启动日志 + 预加载重型模块
  39. 预加载 torch/transformers/chromadb 等重型依赖,避免首次处理论文时
  40. 才触发 import(实测 import torch 单独就要 100+ 秒)。
  41. """
  42. import time as _t
  43. _t0 = _t.perf_counter()
  44. logger.info("Paper Innovation Agent API 已启动,开始预加载重型模块...")
  45. # 顺序预加载:torch → transformers → chromadb → embedding(含模型加载)
  46. try:
  47. _t1 = _t.perf_counter()
  48. import torch # noqa: F401
  49. logger.info("[预加载] import torch: %.2fs", _t.perf_counter() - _t1)
  50. _t2 = _t.perf_counter()
  51. from transformers import AutoModel, AutoTokenizer # noqa: F401
  52. logger.info("[预加载] import transformers: %.2fs", _t.perf_counter() - _t2)
  53. _t3 = _t.perf_counter()
  54. import chromadb # noqa: F401
  55. logger.info("[预加载] import chromadb: %.2fs", _t.perf_counter() - _t3)
  56. # 预加载 Embedder 单例(首次会触发模型加载 + CUDA 上下文初始化)
  57. _t4 = _t.perf_counter()
  58. from backend.rag.embedding import get_default_embedder
  59. _embedder = get_default_embedder() # 注册为模块级单例
  60. # 触发模型加载 + 首次 forward(完成 CUDA 上下文初始化)
  61. _embedder.embed_texts(["预加载测试文本"])
  62. logger.info(
  63. "[预加载] Embedder 模型加载 + CUDA warmup: %.2fs (device=%s)",
  64. _t.perf_counter() - _t4,
  65. getattr(_embedder, "_device", "unknown"),
  66. )
  67. except Exception as exc:
  68. logger.warning("[预加载] 失败(不影响启动): %s", exc)
  69. logger.info("[预加载] 全部完成: %.2fs", _t.perf_counter() - _t0)
  70. if __name__ == "__main__":
  71. import uvicorn
  72. uvicorn.run(
  73. "backend.api.main:app",
  74. host="0.0.0.0",
  75. port=8000,
  76. reload=False,
  77. )