| import gradio as gr |
| import onnxruntime as rt |
| from transformers import AutoTokenizer |
| import torch, json |
|
|
| tokenizer = AutoTokenizer.from_pretrained("roberta-base") |
|
|
| with open("recipe_types_encoded19.json", "r") as fp: |
| encode_types = json.load(fp) |
|
|
| recipe = list(encode_types.keys()) |
|
|
| inf_session = rt.InferenceSession('recipe-classifier.onnx') |
| input_name = inf_session.get_inputs()[0].name |
| output_name = inf_session.get_outputs()[0].name |
|
|
| def classify_recipe(description): |
| input_ids = tokenizer(description)['input_ids'][:512] |
| logits = inf_session.run([output_name], {input_name: [input_ids]})[0] |
| logits = torch.FloatTensor(logits) |
| probs = torch.sigmoid(logits)[0] |
| result = {class_label: float(prob) for class_label, prob in zip(recipe,probs)} |
| return result |
|
|
| label = gr.outputs.Label(num_top_classes=5) |
| iface = gr.Interface(fn=classify_recipe, inputs="text", outputs=label) |
| iface.launch(inline=False) |
