Upload 2 files

g2p_t5.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77f18426d1321bd08c29a7f0fb743b2d8e83b850178a7bab16a3b64d3f04a415
-size 1208296672
+oid sha256:e15c8d2249cc940232f58442a2e93fe9e27fbaaaebdfb356f5f7a3a0fb7ec9c5
+size 1208441138

infer.py

@@ -0,0 +1,30 @@
+import onnxruntime
+import torch
+
+from transformers import AutoTokenizer
+
+# setup GPU
+if torch.cuda.is_available():
+    device = [0]  # use 0th CUDA device
+    accelerator = 'gpu'
+else:
+    device = 1
+    accelerator = 'cpu'
+
+map_location = torch.device('cuda:{}'.format(device[0]) if accelerator == 'gpu' else 'cpu')
+
+tokenizer = AutoTokenizer.from_pretrained('google/byt5-small')
+
+sentence = "Kupil sem bicikel in mu zamenjal stol.".lower()
+
+ort_session = onnxruntime.InferenceSession("g2p_t5.onnx", providers=["CPUExecutionProvider"])
+input_ids = [sentence]
+input_encoding = tokenizer(
+    input_ids, padding='longest', max_length=512, truncation=True, return_tensors='pt',
+)
+input_ids, attention_mask = input_encoding.input_ids, input_encoding.attention_mask
+ort_inputs = {'input_ids': input_ids.numpy()}
+ort_outs = ort_session.run(None, ort_inputs)
+generated_ids = [ort_outs[0]]
+generated_texts = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
+print(generated_texts)