app.py

from transformers import GPT2Tokenizer, GPT2LMHeadModel, StoppingCriteria, StoppingCriteriaList
import torch, os
import gradio as gr

model_path = "iopwsy/MatGPT-synthesis"
tokenizer = GPT2Tokenizer.from_pretrained(model_path, pad_token = '<|endoftext|>')
model = GPT2LMHeadModel.from_pretrained(model_path)
model.config.pad_token_id = model.config.eos_token_id
model.eval()

class StopforGPT2(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        return input_ids[0,-1].detach().cpu().numpy() == 50256

def get_path(text):
    input_ = tokenizer.encode_plus(text,
                                   add_special_tokens = True,
                                   return_token_type_ids = False,
                                   return_attention_mask = True,
                                   return_special_tokens_mask = False,
                                   return_tensors = 'pt')
    with torch.no_grad():
        res = model.generate(**input_,
                             do_sample=False,
                             max_new_tokens = 300,
                             stopping_criteria=StoppingCriteriaList([StopforGPT2()]))
    return tokenizer.decode(res[0,len(input_['input_ids'][0]):],skip_special_tokens=True)
    
demo = gr.Interface(get_path,
                    gr.Textbox(label="Input",info="How to synthesis xxx?\n"),
                    gr.Textbox(label="Output"))
if __name__ == "__main__":
    demo.launch()