add Incremental Pre-training Guide (#235)
This commit is contained in:
xzw
GitHub
commit
6371a5c703
@ -98,6 +98,7 @@
|
||||
</table>
|
||||
|
||||
### 🎇最近更新
|
||||
- 【2024.5.7】[增量预训练指南](xtuner_config/pt/README.md)
|
||||
- 【2024.4.20】[LLAMA3微调指南](xtuner_config/README_llama3_8b_instruct_qlora_alpaca_e3_M.md)及基于[LLaMA3_8b_instruct的艾薇](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM-LLaMA3_8b_instruct_aiwei)开源
|
||||
- 【2023.4.14】新增[快速开始](docs/quick_start.md)和保姆级教程[BabyEmoLLM](Baby_EmoLLM.ipynb)
|
||||
- 【2024.4.2】在 Huggingface 上传[老母亲心理咨询师](https://huggingface.co/brycewang2018/EmoLLM-mother/tree/main)
|
||||
|
||||
@ -101,6 +101,7 @@ The Model aims to fully understand and promote the mental health of individuals,
|
||||
</table>
|
||||
|
||||
### Recent Updates
|
||||
- [2024.5.7][Incremental Pre-training Guide](xtuner_config/pt/README.md)
|
||||
- [2024.4.20] [LLAMA3 fine-tuning guide](xtuner_config/README_llama3_8b_instruct_qlora_alpaca_e3_M.md) and based on [LLaMA3_8b_instruct's aiwei](https://openxlab.org.cn/models/detail/ajupyter/EmoLLM-LLaMA3_8b_instruct_aiwei) open source
|
||||
- [2023.4.14] Added [Quick Start](docs/quick_start_EN.md) and Nanny level tutorial [BabyEmoLLM](Baby_EmoLLM.ipynb)
|
||||
- [2024.4.2] Uploaded at Huggingface [Old Mother Counsellor](https://huggingface.co/brycewang2018/EmoLLM-mother/tree/main)
|
||||
|
||||
36
xtuner_config/pt/README.md
Normal file
36
xtuner_config/pt/README.md
Normal file
@ -0,0 +1,36 @@
|
||||
# 增量预训练教程
|
||||
|
||||
# 增量预训练简介
|
||||
增量预训练旨在提升模型在特定领域或任务的能力。
|
||||
|
||||
|
||||
# 预训练流程
|
||||
- Step1 处理数据
|
||||
- Step2 配置config（全量、Lora、Qlora）
|
||||
- Step3 启动训练（单卡、多卡、是否使用deepspeed）
|
||||
- Step4 模型合成
|
||||
- Step5 模型测试
|
||||
- Step6 模型上传
|
||||
|
||||
# EmoLLM增量预训练教程
|
||||
基于微调中的数据集[datasets](../../datasets)修改而来
|
||||
|
||||
- Step1 修改`ft2pt.py`中的文件路径
|
||||
这里以[output2.json](../../datasets/processed/output2.json)为例，运行脚本生成[pt.json](../../datasets/pt/pt.json)
|
||||
|
||||
- Step2 [config](./internlm2_chat_1_8b_qlora_e3_pt.py)
|
||||
注意：本config采用了**变长注意力 (Variable Length Attention)**
|
||||
需要安装flash_attn
|
||||
`MAX_JOBS=4 pip install flash-attn --no-build-isolation`
|
||||
|
||||
|
||||
- Step3 训练：
|
||||
```
|
||||
# On a single GPU
|
||||
xtuner train internlm2_chat_1_8b_qlora_e3_pt.py --deepspeed deepspeed_zero2
|
||||
# On multiple GPUs
|
||||
(DIST) NPROC_PER_NODE=${GPU_NUM} xtuner train internlm2_chat_1_8b_qlora_e3_pt.py --deepspeed deepspeed_zero2
|
||||
(SLURM) srun ${SRUN_ARGS} xtuner train internlm2_chat_1_8b_qlora_e3_pt.py --launcher slurm --deepspeed deepspeed_zero2
|
||||
```
|
||||
|
||||
- 其余流程请参考[微调教程](../../xtuner_config/README.md)
|
||||
48
xtuner_config/pt/ft2pt.py
Normal file
48
xtuner_config/pt/ft2pt.py
Normal file
@ -0,0 +1,48 @@
|
||||
# 将微调的数据格式转为预训练的格式
|
||||
import json
|
||||
|
||||
|
||||
def convert(data_path:str, target_path:str):
|
||||
# 假设原始JSON数据存储在名为'data.json'的文件中
|
||||
filename = data_path
|
||||
|
||||
# 读取文件内容
|
||||
with open(filename, 'rt', encoding='utf-8') as file:
|
||||
original_json = file.read()
|
||||
|
||||
# 将原始JSON字符串解析为Python对象
|
||||
data = json.loads(original_json)
|
||||
|
||||
# 遍历每个对话
|
||||
converted_data = []
|
||||
|
||||
# 遍历原始数据中的每个对话对象
|
||||
for conversation_group in data:
|
||||
# 遍历每个对话
|
||||
for dialog in conversation_group["conversation"]:
|
||||
# 创建一个新的对话对象，用于存储转换后的对话
|
||||
new_conversation_group = {
|
||||
"conversation": []
|
||||
}
|
||||
# 创建一个新的对话，其中输出被替换为"xxx"
|
||||
new_dialog = {
|
||||
"input": '',
|
||||
"output": f'问题:{dialog["input"]}\n答案:{dialog["output"]}',
|
||||
}
|
||||
# 将新的对话添加到新对话对象的列表中
|
||||
new_conversation_group["conversation"].append(new_dialog)
|
||||
|
||||
# 将新对话对象添加到转换后的数据列表中
|
||||
converted_data.append(new_conversation_group)
|
||||
|
||||
|
||||
# 将更新后的数据转换回JSON字符串，并格式化输出
|
||||
updated_json = json.dumps(converted_data, indent=4, ensure_ascii=False)
|
||||
|
||||
|
||||
# 将更新后的JSON数据写入到新的文件中
|
||||
with open(f'{target_path}', 'wt', encoding='utf-8') as file:
|
||||
file.write(updated_json)
|
||||
|
||||
if __name__ == '__main__':
|
||||
convert(data_path='./output2.json', target_path='pt.json')
|
||||
202
xtuner_config/pt/internlm2_chat_1_8b_qlora_e3_pt.py
Normal file
202
xtuner_config/pt/internlm2_chat_1_8b_qlora_e3_pt.py
Normal file
@ -0,0 +1,202 @@
|
||||
# Copyright (c) OpenMMLab. All rights reserved.
|
||||
import torch
|
||||
from datasets import load_dataset
|
||||
from mmengine.dataset import DefaultSampler
|
||||
from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
|
||||
LoggerHook, ParamSchedulerHook)
|
||||
from mmengine.optim import AmpOptimWrapper, CosineAnnealingLR, LinearLR
|
||||
from peft import LoraConfig
|
||||
from torch.optim import AdamW
|
||||
from transformers import (AutoModelForCausalLM, AutoTokenizer,
|
||||
BitsAndBytesConfig)
|
||||
|
||||
from xtuner.dataset import process_hf_dataset
|
||||
from xtuner.dataset.collate_fns import default_collate_fn
|
||||
from xtuner.dataset.map_fns import alpaca_map_fn, template_map_fn_factory
|
||||
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
|
||||
VarlenAttnArgsToMessageHubHook)
|
||||
from xtuner.engine.runner import TrainLoop
|
||||
from xtuner.model import SupervisedFinetune
|
||||
from xtuner.parallel.sequence import SequenceParallelSampler
|
||||
from xtuner.utils import PROMPT_TEMPLATE, SYSTEM_TEMPLATE
|
||||
|
||||
#######################################################################
|
||||
# PART 1 Settings #
|
||||
#######################################################################
|
||||
# Model
|
||||
pretrained_model_name_or_path = '/root/share/new_models/Shanghai_AI_Laboratory/internlm2-chat-1_8b'
|
||||
use_varlen_attn = True # True
|
||||
|
||||
# Data
|
||||
data_path = '/root/wxz/work/pt/pt.json'
|
||||
prompt_template = PROMPT_TEMPLATE.internlm2_chat
|
||||
max_length = 2048
|
||||
pack_to_max_length = True
|
||||
|
||||
# parallel
|
||||
sequence_parallel_size = 1
|
||||
|
||||
# Scheduler & Optimizer
|
||||
batch_size = 1 # per_device
|
||||
accumulative_counts = 1
|
||||
accumulative_counts *= sequence_parallel_size
|
||||
dataloader_num_workers = 0
|
||||
max_epochs = 3
|
||||
optim_type = AdamW
|
||||
lr = 2e-4
|
||||
betas = (0.9, 0.999)
|
||||
weight_decay = 0
|
||||
max_norm = 1 # grad clip
|
||||
warmup_ratio = 0.03
|
||||
|
||||
# Save
|
||||
save_steps = 500
|
||||
save_total_limit = 2 # Maximum checkpoints to keep (-1 means unlimited)
|
||||
|
||||
# Evaluate the generation performance during the training
|
||||
evaluation_freq = 500
|
||||
SYSTEM = SYSTEM_TEMPLATE.alpaca
|
||||
evaluation_inputs = [
|
||||
'我们经常因为一些小事争吵，他总是忽略我的感受。我感到很孤独，'
|
||||
]
|
||||
|
||||
#######################################################################
|
||||
# PART 2 Model & Tokenizer #
|
||||
#######################################################################
|
||||
tokenizer = dict(
|
||||
type=AutoTokenizer.from_pretrained,
|
||||
pretrained_model_name_or_path=pretrained_model_name_or_path,
|
||||
trust_remote_code=True,
|
||||
padding_side='right')
|
||||
|
||||
model = dict(
|
||||
type=SupervisedFinetune,
|
||||
use_varlen_attn=use_varlen_attn,
|
||||
llm=dict(
|
||||
type=AutoModelForCausalLM.from_pretrained,
|
||||
pretrained_model_name_or_path=pretrained_model_name_or_path,
|
||||
trust_remote_code=True,
|
||||
torch_dtype=torch.float16,),
|
||||
lora=dict(
|
||||
type=LoraConfig,
|
||||
r=64,
|
||||
lora_alpha=16,
|
||||
lora_dropout=0.1,
|
||||
bias='none',
|
||||
task_type='CAUSAL_LM'))
|
||||
|
||||
#######################################################################
|
||||
# PART 3 Dataset & Dataloader #
|
||||
#######################################################################
|
||||
alpaca_en = dict(
|
||||
type=process_hf_dataset,
|
||||
dataset=dict(type=load_dataset, path='json', data_files=dict(train=data_path)),
|
||||
tokenizer=tokenizer,
|
||||
max_length=max_length,
|
||||
dataset_map_fn=None,
|
||||
template_map_fn=None,
|
||||
remove_unused_columns=True,
|
||||
shuffle_before_pack=True,
|
||||
pack_to_max_length=pack_to_max_length,
|
||||
use_varlen_attn=use_varlen_attn)
|
||||
|
||||
sampler = SequenceParallelSampler \
|
||||
if sequence_parallel_size > 1 else DefaultSampler
|
||||
train_dataloader = dict(
|
||||
batch_size=batch_size,
|
||||
num_workers=dataloader_num_workers,
|
||||
dataset=alpaca_en,
|
||||
sampler=dict(type=sampler, shuffle=True),
|
||||
collate_fn=dict(type=default_collate_fn, use_varlen_attn=use_varlen_attn))
|
||||
|
||||
#######################################################################
|
||||
# PART 4 Scheduler & Optimizer #
|
||||
#######################################################################
|
||||
# optimizer
|
||||
optim_wrapper = dict(
|
||||
type=AmpOptimWrapper,
|
||||
optimizer=dict(
|
||||
type=optim_type, lr=lr, betas=betas, weight_decay=weight_decay),
|
||||
clip_grad=dict(max_norm=max_norm, error_if_nonfinite=False),
|
||||
accumulative_counts=accumulative_counts,
|
||||
loss_scale='dynamic',
|
||||
dtype='float16')
|
||||
|
||||
# learning policy
|
||||
# More information: https://github.com/open-mmlab/mmengine/blob/main/docs/en/tutorials/param_scheduler.md # noqa: E501
|
||||
param_scheduler = [
|
||||
dict(
|
||||
type=LinearLR,
|
||||
start_factor=1e-5,
|
||||
by_epoch=True,
|
||||
begin=0,
|
||||
end=warmup_ratio * max_epochs,
|
||||
convert_to_iter_based=True),
|
||||
dict(
|
||||
type=CosineAnnealingLR,
|
||||
eta_min=0.0,
|
||||
by_epoch=True,
|
||||
begin=warmup_ratio * max_epochs,
|
||||
end=max_epochs,
|
||||
convert_to_iter_based=True)
|
||||
]
|
||||
|
||||
# train, val, test setting
|
||||
train_cfg = dict(type=TrainLoop, max_epochs=max_epochs)
|
||||
|
||||
#######################################################################
|
||||
# PART 5 Runtime #
|
||||
#######################################################################
|
||||
# Log the dialogue periodically during the training process, optional
|
||||
custom_hooks = [
|
||||
dict(type=DatasetInfoHook, tokenizer=tokenizer),
|
||||
]
|
||||
|
||||
if use_varlen_attn:
|
||||
custom_hooks += [dict(type=VarlenAttnArgsToMessageHubHook)]
|
||||
|
||||
# configure default hooks
|
||||
default_hooks = dict(
|
||||
# record the time of every iteration.
|
||||
timer=dict(type=IterTimerHook),
|
||||
# print log every 10 iterations.
|
||||
logger=dict(type=LoggerHook, log_metric_by_epoch=False, interval=10),
|
||||
# enable the parameter scheduler.
|
||||
param_scheduler=dict(type=ParamSchedulerHook),
|
||||
# save checkpoint per `save_steps`.
|
||||
checkpoint=dict(
|
||||
type=CheckpointHook,
|
||||
by_epoch=False,
|
||||
interval=save_steps,
|
||||
max_keep_ckpts=save_total_limit),
|
||||
# set sampler seed in distributed evrionment.
|
||||
sampler_seed=dict(type=DistSamplerSeedHook),
|
||||
)
|
||||
|
||||
# configure environment
|
||||
env_cfg = dict(
|
||||
# whether to enable cudnn benchmark
|
||||
cudnn_benchmark=False,
|
||||
# set multi process parameters
|
||||
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
|
||||
# set distributed parameters
|
||||
dist_cfg=dict(backend='nccl'),
|
||||
)
|
||||
|
||||
# set visualizer
|
||||
visualizer = None
|
||||
|
||||
# set log level
|
||||
log_level = 'INFO'
|
||||
|
||||
# load from which checkpoint
|
||||
load_from = None
|
||||
|
||||
# whether to resume training from the loaded checkpoint
|
||||
resume = False
|
||||
|
||||
# Defaults to use random seed and disable `deterministic`
|
||||
randomness = dict(seed=None, deterministic=False)
|
||||
|
||||
# set log processor
|
||||
log_processor = dict(by_epoch=False)
|
||||
Loading…
Reference in New Issue
Block a user