app.py

import os
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from sentence_transformers import SentenceTransformer
import torch

# Define the API schema for the request body
class EmbeddingRequest(BaseModel):
    inputs: list[str]

# Initialize FastAPI app
app = FastAPI()

# Check for GPU and load model accordingly
device = "cuda" if torch.cuda.is_available() else "cpu"
model_name = os.getenv("MODEL_NAME")
model = SentenceTransformer(model_name, device=device, trust_remote_code=True)

# Define the embedding endpoint
@app.post("/embed")
async def get_embeddings(request: EmbeddingRequest):
    try:
        # Get embeddings for the input text
        embeddings = model.encode(request.inputs, convert_to_numpy=True).tolist()
        return embeddings
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))