utils/model.py

from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch


class Model():
    def __init__(self, model="lmsys/vicuna-7b-v1.5") -> None:
        self.tokenizer = AutoTokenizer.from_pretrained(model)
        self.pipeline = transformers.pipeline(
            "text-generation",
            model=model,
            tokenizer=self.tokenizer,
            torch_dtype=torch.bfloat16,
            trust_remote_code=True,
            device_map="auto",
        )

    def gen(self, content, temp=0.1, max_length=500):
        sequences = self.pipeline(
            content,
            max_new_tokens=max_length,
            do_sample=True,
            temperature=temp,
            num_return_sequences=1,
            eos_token_id=self.tokenizer.eos_token_id,
            return_full_text=False
        )
        
        return sequences[-1]['generated_text'] #'\n'.join([seq['generated_text'] for seq in sequences])