rag-demo / main.py
moraxgiga's picture
Update main.py
1f3c160 verified
raw
history blame
775 Bytes
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel, root_validator
from transformers import AutoModel
from typing import List
import os, platform , time
model = AutoModel.from_pretrained('jinaai/jina-embeddings-v2-base-en', trust_remote_code=True)
app = FastAPI()
class Validation(BaseModel):
prompt: List[str]
#Endpoint
@app.post("/jina_embedding")
async def generate_embeddings(item: Validation):
start_time = time.time()
embeddings = model.encode(item.prompt).tolist()
end_time = time.time()
time_taken = end_time - start_time # Calculate the time taken
return {
"embeddings": embeddings,
"time_taken": f"{time_taken:.2f} seconds",
"Number_of_sentence_processed" : len(item.prompt)
}