wehub-resource-sync
30 lines
891 B
Python
30 lines
891 B
Python
import argparse
|
|
import os
|
|
from collections import OrderedDict
|
|
from glob import glob
|
|
|
|
import torch
|
|
from safetensors import safe_open
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM, AutoModel
|
|
from transformers.models.llama import LlamaForCausalLM, LlamaConfig
|
|
from accelerate import init_empty_weights
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument("--input_dir", type=str)
|
|
parser.add_argument("--output_dir", type=str)
|
|
parser.add_argument("--tp_size", type=int)
|
|
args = parser.parse_args()
|
|
|
|
model = AutoModelForCausalLM.from_pretrained(args.input_dir)
|
|
model.resize_token_embeddings(new_num_tokens=None, pad_to_multiple_of=args.tp_size)
|
|
model.save_pretrained(args.output_dir)
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(args.input_dir)
|
|
tokenizer.save_pretrained(args.output_dir)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|