import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import warnings
import glob
from peft import PeftModel
warnings.filterwarnings("ignore")


base_model_id = "mistralai/Mistral-7B-v0.1"
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16
)

base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
    use_auth_token=True
)

tokenizer = AutoTokenizer.from_pretrained(base_model_id, trust_remote_code=True, padding_side='left')  # <-- CHANGE MADE HERE
tokenizer.pad_token = tokenizer.eos_token



ft_model = PeftModel.from_pretrained(base_model, "PEFT-CHECPOINT-PATH")
prefix = "translate English Text to Hindi Text: "
eval_prompt = prefix+"Why people are crazy?.</end>"
# eval_prompt = "Translate in Hindi: I am good "
model_input = tokenizer(eval_prompt, return_tensors="pt").to("cuda")

ft_model.eval()

with torch.no_grad():
    print(tokenizer.decode(ft_model.generate(**model_input, max_new_tokens=70, pad_token_id=2, repetition_penalty=1.2)[0], skip_special_tokens=True))