Trying to use ONNX model.

app.py

@@ -4,7 +4,9 @@ import torch
 from transformers import AutoTokenizer
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
-from onnx_transformers import pipeline
+from transformers import pipeline
+from huggingface_hub import cached_download
+from optimum.onnxruntime import ORTModelForQuestionAnswering
 
 class OnnxTokenClassificationPipeline(TokenClassificationPipeline):
 
@@ -20,8 +22,13 @@ app.add_middleware(
 )
 
 model_name = "xenova/mobilebert-uncased-mnli"
+model = ORTModelForQuestionAnswering.from_pretrained(model_name)
+tokenizer = AutoTokenizer.from_pretrained("typeform/mobilebert-uncased-mnli")
 
-classifier = pipeline(task="zero-shot-classification", model=model_name, onnx=True)
+# file = cached_download("https://huggingface.co/" + model_name + "")
+# sess = InferenceSession(file)
+
+classifier = pipeline(task="zero-shot-classification", model=model, tokenizer=tokenizer)
 
 def zero_shot_classification(data_string):
     print(data_string)

requirements.txt

@@ -1,8 +1,9 @@
 fastapi==0.88.0
+huggingface_hub
 json5==0.9.10
 numpy==1.23.4
 onnxruntime==1.18.1
-https://github.com/patil-suraj/onnx_transformers
+optimum==1.21.3
 torch==1.12.1
 torchvision==0.13.1
 transformers==4.44.0