import os import gradio as gr from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer from threading import Thread import torch tok = AutoTokenizer.from_pretrained("distilgpt2") model = AutoModelForCausalLM.from_pretrained("distilgpt2") device = torch.device("cuda" if torch.cuda.is_available() else "cpu") n_gpu = 0 if torch.cuda.is_available()==False else torch.cuda.device_count() model.to(device) def generate(text = "", max_new_tokens = 128): streamer = TextIteratorStreamer(tok, timeout=10.) if len(text) == 0: text = " " inputs = tok([text], return_tensors="pt").to(device) generation_kwargs = dict(inputs, streamer=streamer, repetition_penalty=2.0, do_sample=True, top_k=40, top_p=0.97, max_new_tokens=max_new_tokens, pad_token_id = model.config.eos_token_id, early_stopping=True, no_repeat_ngram_size=4) thread = Thread(target=model.generate, kwargs=generation_kwargs) thread.start() generated_text = "" for new_text in streamer: yield generated_text + new_text generated_text += new_text if tok.eos_token in generated_text: generated_text = generated_text[: generated_text.find(tok.eos_token) if tok.eos_token else None] streamer.end() yield generated_text return return generated_text demo = gr.Interface( title="TextIteratorStreamer + Gradio demo", fn=generate, inputs=[gr.inputs.Textbox(lines=5, label="Input Text"), gr.inputs.Slider(default=128,minimum=5, maximum=256, step=1, label="Maximum number of new tokens")], outputs=gr.outputs.Textbox(label="Generated Text"), allow_flagging="never" ) demo.queue() demo.launch()