| import gradio as gr | |
| from transformers import MT5Tokenizer, MT5ForConditionalGeneration | |
| def summary(text): | |
| mt5_model = MT5ForConditionalGeneration.from_pretrained("google/mt5-large") | |
| mt5_tokenizer = MT5Tokenizer.from_pretrained("google/mt5-large") | |
| input_tokenized = mt5_tokenizer.encode(text, return_tensors='pt') | |
| summary_ids = mt5_model.generate(input_tokenized, | |
| length_penalty = 1, | |
| min_length = 0, | |
| max_length = 200, | |
| num_beams = 1, | |
| no_repeat_ngram_size = 2, | |
| early_stopping = True) | |
| output = mt5_tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False) | |
| return output[0] | |
| iface = gr.Interface(fn=summary, inputs="text", outputs="text") | |
| iface.launch() |