Layoutlmv2 Prediction/Inference from fine tuned .pth model

[nikithakriz]

Hai,

I have finetuned layoutlmv2 model using custom dataset which was annotated in FUNSD format. Now i can save my finetuned model in local(.pth) using torch.save. Now i need to get predictions using finetuned (.pth) state dict. Am beginner to this anyone know how to get predictions. .pth has state dict which is weights and biasis.

I tried to get predictions using:
**state_dict = torch.load("/content/model_17.pth")

Create the model and tokenizer using the state dictionary

model = LayoutLMv2ForTokenClassification.from_pretrained("/config.json",state_dict=state_dict, ignore_mismatched_sizes=True)**

but getting weird prediction output. Anybody knew solution for this?
Thanks in advance!