ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[misc] update license year & fix llama pro (#6814)

Browse Source 
* fix llamapro script

* change year

Former-commit-id: d9ae594178796994d400a5f207d6499712816f89
This commit is contained in:
hoshi-hiyouga
2025-02-05 01:53:33 +08:00
committed by GitHub
parent 0817c24c04
commit c2022431aa
114 changed files with 244 additions and 196 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
scripts/convert_ckpt/llamafy_baichuan2.py
View File

@@ -1,4 +1,4 @@
# Copyright 2024 the LlamaFactory team.
# Copyright 2025 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.
@@ -19,15 +19,10 @@ from typing import Any, Dict
import fire
import torch
from huggingface_hub import split_torch_state_dict_into_shards
from safetensors.torch import save_file
from tqdm import tqdm
from transformers.modeling_utils import (
SAFE_WEIGHTS_INDEX_NAME,
SAFE_WEIGHTS_NAME,
WEIGHTS_INDEX_NAME,
WEIGHTS_NAME,
shard_checkpoint,
)
from transformers.modeling_utils import SAFE_WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_INDEX_NAME, WEIGHTS_NAME
CONFIG_NAME = "config.json"
@@ -40,34 +35,42 @@ def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetenso
shard_weight = torch.load(os.path.join(input_dir, filepath), map_location="cpu")
baichuan2_state_dict.update(shard_weight)
llama2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
llama_state_dict: Dict[str, torch.Tensor] = OrderedDict()
for key, value in tqdm(baichuan2_state_dict.items(), desc="Convert format"):
if "W_pack" in key:
proj_size = value.size(0) // 3
llama2_state_dict[key.replace("W_pack", "q_proj")] = value[:proj_size, :]
llama2_state_dict[key.replace("W_pack", "k_proj")] = value[proj_size : 2 * proj_size, :]
llama2_state_dict[key.replace("W_pack", "v_proj")] = value[2 * proj_size :, :]
llama_state_dict[key.replace("W_pack", "q_proj")] = value[:proj_size, :]
llama_state_dict[key.replace("W_pack", "k_proj")] = value[proj_size : 2 * proj_size, :]
llama_state_dict[key.replace("W_pack", "v_proj")] = value[2 * proj_size :, :]
elif "lm_head" in key:
llama2_state_dict[key] = torch.nn.functional.normalize(value)
llama_state_dict[key] = torch.nn.functional.normalize(value)
else:
llama2_state_dict[key] = value
llama_state_dict[key] = value
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"):
filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
state_dict_split = split_torch_state_dict_into_shards(
llama_state_dict, filename_pattern=filename_pattern, max_shard_size=shard_size
)
for shard_file, tensors in tqdm(state_dict_split.filename_to_tensors.items(), desc="Save weights"):
shard = {tensor: llama_state_dict[tensor].contiguous() for tensor in tensors}
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))
if index is None:
print(f"Model weights saved in {os.path.join(output_dir, WEIGHTS_NAME)}")
if not state_dict_split.is_sharded:
print(f"Model weights saved in {os.path.join(output_dir, weights_name)}.")
else:
index = {
"metadata": state_dict_split.metadata,
"weight_map": state_dict_split.tensor_to_filename,
}
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:
json.dump(index, f, indent=2, sort_keys=True)
print(f"Model weights saved in {output_dir}")
print(f"Model weights saved in {output_dir}.")
def save_config(input_dir: str, output_dir: str):
@@ -81,6 +84,7 @@ def save_config(input_dir: str, output_dir: str):
with open(os.path.join(output_dir, CONFIG_NAME), "w", encoding="utf-8") as f:
json.dump(llama2_config_dict, f, indent=2)
print(f"Model config saved in {os.path.join(output_dir, CONFIG_NAME)}")