Convert gpt4all-alpaca-oa-codealpaca-lora-13b to a HF checkpoint

convert-gpt4all-alpaca-oa-codealpaca-lora-13b.py

# Combined https://github.com/tloen/alpaca-lora/blob/main/export_hf_checkpoint.py and
# https://huggingface.co/jordiclive/gpt4all-alpaca-oa-codealpaca-lora-13b#example-inference-code-note-several-embeddings-need-to-be-loaded-along-with-the-lora-weights-assumes-on-gpu-and-torchfloat16

import os

import torch
import transformers
from peft import PeftModel
from transformers import LlamaForCausalLM, LlamaTokenizer, GenerationConfig
from typing import List, NamedTuple
#from huggingface_hub import hf_hub_download

device = "cpu"
tokenizer = LlamaTokenizer.from_pretrained("jordiclive/gpt4all-alpaca-oa-codealpaca-lora-13b")

base_model = LlamaForCausalLM.from_pretrained(
    "decapoda-research/llama-13b-hf",
    load_in_8bit=False,
    torch_dtype=torch.float16,
    device_map={"": "cpu"},
)
base_model.resize_token_embeddings(
    32005 # original had 32016 ??
)  # This model repo also contains several embeddings for special tokens that need to be loaded.


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,
    "jordiclive/gpt4all-alpaca-oa-codealpaca-lora-13b",
    device_map={"": "cpu"},
    torch_dtype=torch.float16,
)

filename = hf_hub_download("jordiclive/gpt4all-alpaca-oa-codealpaca-lora-13b", "extra_embeddings.pt")
embed_weights = torch.load(
    filename, map_location=torch.device("cpu")
)  # Load embeddings for special tokens
base_model.model.embed_tokens.weight[32000:32005, :] = embed_weights[0:5, :].to(
    base_model.model.embed_tokens.weight.dtype
)  # Add special token embeddings

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
# pip install git+https://github.com/huggingface/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, "./hf_ckpt", state_dict=deloreanized_sd, max_shard_size="1000MB"
)

requirements.txt

huggingface-hub==0.13.4
peft @ git+https://github.com/huggingface/peft@0bdb54f03f651dab7818056f876c6629ef58f568
torch==2.0.0
transformers==4.28.1
typing_extensions==4.5.0