Playing with models.

app.py

@@ -5,14 +5,14 @@ from transformers import AutoTokenizer
 from transformers import pipeline
 from optimum.onnxruntime import ORTModelForSequenceClassification
 
-# "xenova/mobilebert-uncased-mnli" "typeform/mobilebert-uncased-mnli" Fast but small
+# "xenova/mobilebert-uncased-mnli" "typeform/mobilebert-uncased-mnli" Fast but small--same as bundled in Statosphere
 # "xenova/deberta-v3-base-tasksource-nli" Not impressed
 # "Xenova/bart-large-mnli" A bit slow
 # "Xenova/distilbert-base-uncased-mnli" "typeform/distilbert-base-uncased-mnli" Bad answers
 # "Xenova/deBERTa-v3-base-mnli" "MoritzLaurer/DeBERTa-v3-base-mnli" Still a bit slow and not great answers
-model_name = "xenova/mobilebert-uncased-mnli"
-file_name = "onnx/model.onnx"
-tokenizer_name = "typeform/mobilebert-uncased-mnli"
+model_name = "xenova/nli-deberta-v3-small"
+file_name = "onnx/model_fp16.onnx"
+tokenizer_name = "cross-encoder/nli-deberta-v3-small"
 model = ORTModelForSequenceClassification.from_pretrained(model_name, file_name=file_name)
 tokenizer = AutoTokenizer.from_pretrained(tokenizer_name, model_max_length=512)