import torch
from transformers import GPT2LMHeadModel, GPT2Tokenizer, StoppingCriteria, StoppingCriteriaList

tokenizer_dpo = GPT2Tokenizer.from_pretrained("RaushanTurganbay/GPT2_sft_and_dpo_tuned")
model_dpo = GPT2LMHeadModel.from_pretrained("RaushanTurganbay/GPT2_sft_and_dpo_tuned")

class StoppingCriteriaSub(StoppingCriteria):
    def __init__(self, stops=[], encounters=1):
        super().__init__()
        self.stops = [stop.to("cuda") for stop in stops]
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor):
        for stop in self.stops:
            if torch.all((stop == input_ids[0][-len(stop):])).item():
                return True
        return False


def stopping_criteria(tokenizer, stop_words):
    stop_words_ids = [tokenizer(stop_word, return_tensors='pt')['input_ids'].squeeze() for stop_word in stop_words]
    stopping_criteria = StoppingCriteriaList([StoppingCriteriaSub(stops=stop_words_ids)])
    return stopping_criteria


# Generate responses
stopping = stopping_criteria(tokenizer, ["\n\nHuman:"])
prompt = "\n\nHuman: {your_instruction}\n\nAssistant:"
inputs_dpo = tokenizer_dpo(prompt, return_tensors="pt")
outputs_dpo = model_dpo.generate(**inputs_dpo,  stopping_criteria=stopping, max_length=150)

print("Model Response:", tokenizer_dpo.batch_decode(outputs_dpo))