LLaMA-Factory-Mirror/scripts/llamafy_qwen.py

163 lines
7.0 KiB
Python
Raw Normal View History

2023-10-08 22:05:36 +08:00
# coding=utf-8
2024-06-15 17:54:33 +08:00
# Copyright 2024 the LlamaFactory team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
2023-10-08 22:05:36 +08:00
2024-01-20 20:15:56 +08:00
import json
2023-10-08 22:05:36 +08:00
import os
2024-01-20 20:15:56 +08:00
from collections import OrderedDict
2024-08-09 19:16:23 +08:00
from typing import Any, Dict
2024-01-20 20:15:56 +08:00
2023-10-08 22:05:36 +08:00
import fire
import torch
from safetensors import safe_open
2023-12-25 18:29:34 +08:00
from safetensors.torch import save_file
2024-01-20 20:15:56 +08:00
from tqdm import tqdm
2023-12-25 18:29:34 +08:00
from transformers.modeling_utils import (
SAFE_WEIGHTS_INDEX_NAME,
2024-01-20 20:15:56 +08:00
SAFE_WEIGHTS_NAME,
WEIGHTS_INDEX_NAME,
2023-12-25 18:29:34 +08:00
WEIGHTS_NAME,
2024-01-20 20:15:56 +08:00
shard_checkpoint,
2023-12-25 18:29:34 +08:00
)
2023-10-08 22:05:36 +08:00
from transformers.utils import check_min_version
2024-01-20 20:15:56 +08:00
2023-10-08 22:05:36 +08:00
try:
check_min_version("4.34.0")
2024-01-20 20:15:56 +08:00
except Exception:
2023-10-08 22:05:36 +08:00
raise ValueError("Please upgrade `transformers` to 4.34.0")
CONFIG_NAME = "config.json"
2024-01-20 20:15:56 +08:00
def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool) -> str:
2023-10-08 22:05:36 +08:00
qwen_state_dict: Dict[str, torch.Tensor] = OrderedDict()
2024-01-18 00:37:37 +08:00
for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
2023-10-08 22:05:36 +08:00
if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".safetensors"):
with safe_open(os.path.join(input_dir, filepath), framework="pt", device="cpu") as f:
for key in f.keys():
qwen_state_dict[key] = f.get_tensor(key)
llama2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
torch_dtype = None
2023-12-25 18:29:34 +08:00
for key, value in tqdm(qwen_state_dict.items(), desc="Convert format"):
2023-10-08 22:05:36 +08:00
if torch_dtype is None:
torch_dtype = value.dtype
if "wte" in key:
llama2_state_dict["model.embed_tokens.weight"] = value
elif "ln_f" in key:
llama2_state_dict["model.norm.weight"] = value
else:
key = key.replace("transformer.h", "model.layers")
if "attn.c_attn" in key:
proj_size = value.size(0) // 3
llama2_state_dict[key.replace("attn.c_attn", "self_attn.q_proj")] = value[:proj_size, ...]
2024-01-20 20:15:56 +08:00
llama2_state_dict[key.replace("attn.c_attn", "self_attn.k_proj")] = value[
proj_size : 2 * proj_size, ...
]
llama2_state_dict[key.replace("attn.c_attn", "self_attn.v_proj")] = value[2 * proj_size :, ...]
2023-10-08 22:05:36 +08:00
elif "attn.c_proj" in key:
llama2_state_dict[key.replace("attn.c_proj", "self_attn.o_proj")] = value
2024-01-20 20:15:56 +08:00
llama2_state_dict[key.replace("attn.c_proj.weight", "self_attn.o_proj.bias")] = torch.zeros_like(
value[:, 0]
).squeeze()
2023-10-08 22:05:36 +08:00
elif "ln_1" in key:
llama2_state_dict[key.replace("ln_1", "input_layernorm")] = value
elif "ln_2" in key:
llama2_state_dict[key.replace("ln_2", "post_attention_layernorm")] = value
elif "mlp.w1" in key:
llama2_state_dict[key.replace("mlp.w1", "mlp.up_proj")] = value
elif "mlp.w2" in key:
llama2_state_dict[key.replace("mlp.w2", "mlp.gate_proj")] = value
elif "mlp.c_proj" in key:
llama2_state_dict[key.replace("mlp.c_proj", "mlp.down_proj")] = value
elif "lm_head" in key:
llama2_state_dict[key] = value
else:
raise KeyError("Unable to process key {}".format(key))
2023-12-25 18:29:34 +08:00
weights_name = SAFE_WEIGHTS_NAME if save_safetensors else WEIGHTS_NAME
shards, index = shard_checkpoint(llama2_state_dict, max_shard_size=shard_size, weights_name=weights_name)
for shard_file, shard in tqdm(shards.items(), desc="Save weights"):
if save_safetensors:
save_file(shard, os.path.join(output_dir, shard_file), metadata={"format": "pt"})
else:
torch.save(shard, os.path.join(output_dir, shard_file))
2023-10-08 22:05:36 +08:00
if index is None:
2023-12-25 18:29:34 +08:00
print("Model weights saved in {}".format(os.path.join(output_dir, weights_name)))
2023-10-08 22:05:36 +08:00
else:
2023-12-25 18:29:34 +08:00
index_name = SAFE_WEIGHTS_INDEX_NAME if save_safetensors else WEIGHTS_INDEX_NAME
with open(os.path.join(output_dir, index_name), "w", encoding="utf-8") as f:
2023-10-08 22:05:36 +08:00
json.dump(index, f, indent=2, sort_keys=True)
print("Model weights saved in {}".format(output_dir))
return str(torch_dtype).replace("torch.", "")
2024-01-20 20:15:56 +08:00
def save_config(input_dir: str, output_dir: str, torch_dtype: str):
2023-10-08 22:05:36 +08:00
with open(os.path.join(input_dir, CONFIG_NAME), "r", encoding="utf-8") as f:
qwen_config_dict: Dict[str, Any] = json.load(f)
llama2_config_dict: Dict[str, Any] = OrderedDict()
llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
llama2_config_dict["hidden_act"] = "silu"
llama2_config_dict["hidden_size"] = qwen_config_dict["hidden_size"]
llama2_config_dict["initializer_range"] = qwen_config_dict["initializer_range"]
llama2_config_dict["intermediate_size"] = qwen_config_dict["intermediate_size"] // 2
llama2_config_dict["max_position_embeddings"] = qwen_config_dict["max_position_embeddings"]
llama2_config_dict["model_type"] = "llama"
llama2_config_dict["num_attention_heads"] = qwen_config_dict["num_attention_heads"]
llama2_config_dict["num_hidden_layers"] = qwen_config_dict["num_hidden_layers"]
llama2_config_dict["num_key_value_heads"] = qwen_config_dict["hidden_size"] // qwen_config_dict["kv_channels"]
llama2_config_dict["pretraining_tp"] = 1
llama2_config_dict["rms_norm_eps"] = qwen_config_dict["layer_norm_epsilon"]
llama2_config_dict["rope_scaling"] = None
llama2_config_dict["tie_word_embeddings"] = qwen_config_dict["tie_word_embeddings"]
llama2_config_dict["torch_dtype"] = torch_dtype
llama2_config_dict["transformers_version"] = "4.34.0"
llama2_config_dict["use_cache"] = True
llama2_config_dict["vocab_size"] = qwen_config_dict["vocab_size"]
llama2_config_dict["attention_bias"] = True
with open(os.path.join(output_dir, CONFIG_NAME), "w", encoding="utf-8") as f:
json.dump(llama2_config_dict, f, indent=2)
print("Model config saved in {}".format(os.path.join(output_dir, CONFIG_NAME)))
2024-02-15 02:27:36 +08:00
def llamafy_qwen(
2024-08-09 19:16:23 +08:00
input_dir: str,
output_dir: str,
shard_size: str = "2GB",
save_safetensors: bool = False,
2024-02-15 02:27:36 +08:00
):
2024-06-15 17:54:33 +08:00
r"""
Converts the Qwen models in the same format as LLaMA2.
Usage: python llamafy_qwen.py --input_dir input --output_dir output
Converted model: https://huggingface.co/hiyouga/Qwen-14B-Chat-LLaMAfied
"""
2023-10-08 22:05:36 +08:00
try:
os.makedirs(output_dir, exist_ok=False)
except Exception as e:
raise print("Output dir already exists", e)
2023-12-25 18:29:34 +08:00
torch_dtype = save_weight(input_dir, output_dir, shard_size, save_safetensors)
save_config(input_dir, output_dir, torch_dtype)
2023-10-08 22:05:36 +08:00
if __name__ == "__main__":
fire.Fire(llamafy_qwen)