Trying ONNX model on CPU.

app.py

@@ -2,12 +2,12 @@ import spaces
 import torch
 import gradio
 import json
-import onnxruntime
 import time
 from datetime import datetime
 from transformers import pipeline
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
+from optimum.onnxruntime import ORTModelForSequenceClassification
 
 # CORS Config - This isn't actually working; instead, I am taking a gross approach to origin whitelisting within the service.
 app = FastAPI()
@@ -27,8 +27,12 @@ print(f"CUDA device: {torch.cuda.get_device_name(torch.cuda.current_device())}")
 
 model_name = "MoritzLaurer/deberta-v3-base-zeroshot-v2.0"
 tokenizer_name = "MoritzLaurer/deberta-v3-base-zeroshot-v2.0"
+file_name = "onnx/model.onnx"
 
-classifier_cpu = pipeline(task="zero-shot-classification", model=model_name, tokenizer=tokenizer_name)
+model_cpu = ORTModelForSequenceClassification.from_pretrained(model_id=model_name, file_name=file_name)
+tokenizer_cpu = AutoTokenizer.from_pretrained(model_name)
+
+classifier_cpu = pipeline(task="zero-shot-classification", model=model_cpu, tokenizer=tokenizer_cpu)
 classifier_gpu = pipeline(task="zero-shot-classification", model=model_name, tokenizer=tokenizer_name, device="cuda:0")
 
 def classify(data_string, request: gradio.Request):

requirements.txt

@@ -4,5 +4,5 @@ huggingface_hub==0.26.0
 json5==0.9.25
 numpy
 uvicorn
-optimum[onnxruntime-gpu]==1.24.0
+optimum[exporters,onnxruntime,onnxruntime-gpu]==1.24.0
 transformers==4.36