# ### Keywords to Title Generator # - https://huggingface.co/EnglishVoice/t5-base-keywords-to-headline?text=diabetic+diet+plan # - Apache 2.0 import torch from transformers import T5ForConditionalGeneration,T5Tokenizer import gradio as gr device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = T5ForConditionalGeneration.from_pretrained("EnglishVoice/t5-base-keywords-to-headline") tokenizer = T5Tokenizer.from_pretrained("EnglishVoice/t5-base-keywords-to-headline", clean_up_tokenization_spaces=True, legacy=False) model = model.to(device) def title_gen(keywords, diversity, temp): if keywords!= "": text = "headline: " + keywords encoding = tokenizer.encode_plus(text, return_tensors = "pt") input_ids = encoding["input_ids"].to(device) attention_masks = encoding["attention_mask"].to(device) if diversity: num_beams = 20, num_beam_groups = 20, diversity_penalty=0.8, early_stopping = True, else: penalty_alpha = 0.8, beam_outputs = model.generate( input_ids = input_ids, attention_mask = attention_masks, max_new_tokens = 30, do_sample = True, num_return_sequences = 5, temperature = temp, top_k = 15, no_repeat_ngram_size = 3, #top_p = 0.60, ) titles = "
Title Suggestions:
" for i in range(len(beam_outputs)): result = tokenizer.decode(beam_outputs[i], skip_special_tokens=True) titles += f"{result}
" #Create string with titles andAI Model:
Default parameter details:
temperature = 1.2
, no_repeat_ngram_size=3
, top_k = 15
, penalty_alpha = 0.8
, max_new_tokens = 30
Diversity beam search params:
num_beams=20
, diversity_penalty=0.8
, num_beam_groups=20