Upload 2 files

bert_din_quant.py

@@ -0,0 +1,17 @@
+from optimum.onnxruntime import ORTModelForTokenClassification, ORTQuantizer
+from optimum.onnxruntime.configuration import AutoQuantizationConfig
+
+onnx_model = ORTModelForTokenClassification.from_pretrained(
+    "nickprock/bert-italian-finetuned-ner",
+     export=True
+)
+quantizer = ORTQuantizer.from_pretrained(onnx_model)
+dqconfig = AutoQuantizationConfig.avx512_vnni(
+    is_static=False, 
+    per_channel=False
+)
+
+model_quantized_path = quantizer.quantize(
+    save_dir="bert-italian-ner-onnx-quantized-avx512",
+    quantization_config=dqconfig,
+)

bert_din_quant_test.py

@@ -0,0 +1,13 @@
+import time
+from transformers import AutoTokenizer, pipeline
+from optimum.onnxruntime import ORTModelForTokenClassification
+
+tokenizer = AutoTokenizer.from_pretrained("./bert-italian-ner-onnx-quantized-avx512")
+model = ORTModelForTokenClassification.from_pretrained("./bert-italian-ner-onnx-quantized-avx512")
+nerpipeline = pipeline('ner', model=model, tokenizer=tokenizer)
+
+text = "La sede storica della Olivetti è ad Ivrea"
+start_time = time.time()
+output = nerpipeline(text)
+print(f"--- {time.time() - start_time} seconds ---")
+print(output)