from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# https://www.youtube.com/watch?v=irjYqV6EebU

model_name = "facebook/blenderbot-1B-distill"

# https://huggingface.co/models?sort=trending&search=facebook%2Fblenderbot
# facebook/blenderbot-3B
# facebook/blenderbot-1B-distill
# facebook/blenderbot-400M-distill
# facebook/blenderbot-90M
# facebook/blenderbot_small-90M

def load():
  global model
  global tokenizer
  
  model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
  tokenizer = AutoTokenizer.from_pretrained(model_name)

def generate(input_text):
  # Tokenize the input text
  input_ids = tokenizer.encode(input_text, return_tensors="pt")
  
  # Generate output using the model
  output_ids = model.generate(input_ids, no_repeat_ngram_size=2, max_new_tokens=200, num_beams=2, eos_token_id=tokenizer.eos_token_id)
  
  return tokenizer.decode(output_ids[0], skip_special_tokens=True)