model.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import warnings





torch_dtype = torch.bfloat16
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_name = "mosaicml/mpt-7b"


model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch_dtype,
    trust_remote_code=True,
    use_auth_token=None,
)
tokenizer = AutoTokenizer.from_pretrained(
    model_name,
    trust_remote_code=True,
    use_auth_token=None,
)
model.eval()
model.to(device=device, dtype=torch_dtype)
if tokenizer.pad_token_id is None:
    warnings.warn(
        "pad_token_id is not set for the tokenizer. Using eos_token_id as pad_token_id."
    )
    tokenizer.pad_token = tokenizer.eos_token
    tokenizer.padding_side = "left"

gkw = {
    "temperature": 0.5,
    "top_p": 0.92,
    "top_k": 0,
    "max_new_tokens": 512,
    "use_cache": True,
    "do_sample": True,
    "eos_token_id": tokenizer.eos_token_id,
    "pad_token_id": tokenizer.pad_token_id,
    "repetition_penalty": 1.1,  # 1.0 means no penalty, > 1.0 means penalty, 1.2 from CTRL paper
}


def mpt_7b(s):
    input_ids = tokenizer(s, return_tensors="pt").input_ids
    input_ids = input_ids.to(model.device)
    with torch.no_grad():
        output_ids = model.generate(input_ids, **gkw)
        # Slice the output_ids tensor to get only new tokens
        new_tokens = output_ids[0, len(input_ids[0]) :]
        output_text = tokenizer.decode(new_tokens, skip_special_tokens=True)
    return output_text