daochu.py

import os

import torch
import transformers
from peft import PeftModel
from transformers import LlamaForCausalLM,AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

BASE_MODEL = "/root/autodl-tmp/meta-llama/Llama-Guard-3-8B"
assert (
    BASE_MODEL
), "Please specify a value for BASE_MODEL environment variable, e.g. `export BASE_MODEL=huggyllama/llama-7b`"  # noqa: E501

tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)

base_model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    load_in_8bit=False,
    torch_dtype=torch.float16,
    device_map={"": "cpu"},
)

first_weight = base_model.model.layers[0].self_attn.q_proj.weight
first_weight_old = first_weight.clone()

lora_model = PeftModel.from_pretrained(
    base_model,
    "/root/PATH/to/save/PEFT/model",
    device_map={"": "cpu"},
    torch_dtype=torch.float16,
)

lora_weight = lora_model.base_model.model.model.layers[
    0
].self_attn.q_proj.weight

assert torch.allclose(first_weight_old, first_weight)

# merge weights - new merging method from peft
lora_model = lora_model.merge_and_unload()

lora_model.train(False)

# did we do anything?
assert not torch.allclose(first_weight_old, first_weight)

lora_model_sd = lora_model.state_dict()
deloreanized_sd = {
    k.replace("base_model.model.", ""): v
    for k, v in lora_model_sd.items()
    if "lora" not in k
}

LlamaForCausalLM.save_pretrained(
    base_model, "./8b", state_dict=deloreanized_sd, max_shard_size="5000MB"
)