bn_llm_wrapper

import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig, GenerationConfig

model_path = os.environ.get("HF_REPO_ID")
access_token = os.environ.get("HF_TOKEN")


tokenizer = AutoTokenizer.from_pretrained(model_path, token=access_token)

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    # load_in_8bit=use_8_bit,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=getattr(torch, "bfloat16"),
    bnb_4bit_use_double_quant=True,
)

model = AutoModelForCausalLM.from_pretrained(model_path, token=access_token,
                                             quantization_config=bnb_config,
                                             torch_dtype=torch.float16,
                                             # attn_implementation="flash_attention_2",
                                             device_map='auto')

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

def generate(
    question,
    context=None,
    temperature=0.7,
    top_p=0.7,
    top_k=40,
    num_beams=4,
    max_new_tokens=256,):
        prompt = f"### CONTEXT:\n{context}\n\n### QUESTION:\n{question}\n\n### ANSWER:"
        inputs = tokenizer(prompt, return_tensors="pt")
        input_ids = inputs["input_ids"].to(device)
        generation_config = GenerationConfig(
            temperature=temperature,
            top_p=top_p,
            top_k=top_k,
            num_beams=num_beams,
        )
        # with torch.autocast("cuda"):
        with torch.no_grad():
            generation_output = model.generate(
                input_ids=input_ids,
                generation_config=generation_config,
                return_dict_in_generate=True,
                output_scores=True,
                max_new_tokens=max_new_tokens,
            )
        seq = generation_output.sequences[0]
        output = tokenizer.decode(seq)
        return output