handler.py

from typing import Dict, List, Any
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel, PeftConfig
import torch
import time

class EndpointHandler:
    def __init__(self, path="5iveDesignStudio/autotrain-TenderGPT-Festive-v2-0"):
        # load the model
        config = PeftConfig.from_pretrained(path)
        
        bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_use_double_quant=True,
            bnb_4bit_quant_type="nf4",
            bnb_4bit_compute_dtype=torch.float16,
        )      
        
        self.model = AutoModelForCausalLM.from_pretrained(
            config.base_model_name_or_path,
            return_dict=True,
            load_in_4bit=True,
            device_map={"":0},
            trust_remote_code=True,
            quantization_config=bnb_config,
        )
        
        self.tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
        self.tokenizer.pad_token = self.tokenizer.eos_token
        
        self.model = PeftModel.from_pretrained(self.model, path)

        self.device = "cuda" if torch.cuda.is_available() else "cpu"

    #def __call__(self, data: Any) -> Dict[str, Any]:
    def __call__(self, data: Dict[str, Any]) -> Dict[str, str]:
        """
        Args:
            inputs :obj:`list`:. The object should be like {"context": "some word", "question": "some word"} containing:
                - "context": 
                - "question": 
        Return:
            A :obj:`list`:. The object returned should be like {"answer": "some word", time: "..."} containing:
                - "answer": answer the question based on the context
                - "time": the time run predict
        """

        # process input
        inputs = data.pop("inputs", data)
        parameters = data.pop("parameters", None)

        prompt = f"""Below is an instruction that describes a task. If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.
>>TITLE<<: Tender Response.
>>CONTEXT<<: You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe in a conversational tone. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
>>QUESTION<<: {inputs}
>>ANSWER<<:
""".strip()

        # preprocess
        batch = self.tokenizer(
            prompt,
            padding=True,
            truncation=True,
            return_tensors="pt"
        ).to(self.device)

        # pass inputs with all kwargs in data
        #if parameters is not None:
        #    outputs = self.model.generate(**inputs, **parameters)
        #else:
        #    outputs = self.model.generate(**inputs)

        # postprocess the prediction
        #prediction = self.tokenizer.decode(outputs[0], skip_special_tokens=True)

        generation_config = self.model.generation_config
        generation_config.top_p = 0.75
        generation_config.temperature = 0.7
        generation_config.max_new_tokens = 140
        generation_config.num_return_sequences = 1
        generation_config.pad_token_id = self.tokenizer.eos_token_id
        generation_config.eos_token_id = self.tokenizer.eos_token_id

        start = time.time()
        with torch.cuda.amp.autocast():
            output_tokens = self.model.generate(
                input_ids = batch.input_ids,
                generation_config=generation_config,
            )
        end = time.time()

        generated_text = self.tokenizer.decode(
            output_tokens[0]
        )

        prediction = {'answer': generated_text.split('>>END<<')[0].split('>>ANSWER<<:')[1].strip(), 'time': f"{(end-start):.2f} s"} 

        return prediction