Skip to content

Latest commit

 

History

History
275 lines (249 loc) · 8.6 KB

README.md

File metadata and controls

275 lines (249 loc) · 8.6 KB
Raw

Qwen2 模型在Pai-Megatron-Patch的最佳实践

Table of Contents

安装

请在阿里云人工智能平台PAI产品中填写专属镜像地址： dsw-registry.cn-wulanchabu.cr.aliyuncs.com/pai/pai-megatron-patch:24.07

运行下列代码克隆Pai-Megatron-Patch

git clone --recurse-submodules https://github.com/alibaba/Pai-Megatron-Patch.git
cd Pai-Megatron-Patch

目前Qwen2已支持使用FlashAttention-3加速计算，但只能在Hopper架构的GPU卡上进行运算。若需要在H卡上使用FA3，请在DSW的容器中按如下指令安装并保存镜像

pip install "git+https://github.com/Dao-AILab/flash-attention.git#egg=flashattn-hopper&subdirectory=hopper"
python_path=`python -c "import site; print(site.getsitepackages()[0])"`
mkdir -p $python_path/flashattn_hopper
wget -P $python_path/flashattn_hopper https://raw.githubusercontent.com/Dao-AILab/flash-attention/main/hopper/flash_attn_interface.py

预训练数据集和模型下载

cd /mnt
mkdir qwen-ckpts
cd qwen-ckpts
wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/qwen-ckpts/Qwen2-0.5B.tgz
tar -zxf Qwen2-0.5B.tgz

mkdir qwen-datasets
wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/qwen-datasets/wudao_qwenbpe_text_document.bin
wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/qwen-datasets/wudao_qwenbpe_text_document.idx

wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/qwen-datasets/alpaca_zh-qwen-train.json
wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/qwen-datasets/alpaca_zh-qwen-valid.json

Megatron-Core模型训练流程

Megatron-Core模型格式转换

运行hf2mcore_qwen2_convertor.sh脚本，需要传入的参数列表如下

MODEL_SIZE=$1                  # 模型参数：0.5B/1.8B
SOURCE_CKPT_PATH=$2            # 源路径
TARGET_CKPT_PATH=$3            # 目标路径
TP=$4                          # 模型并行度
PP=$5                          # 流水并行度
EP=$6                          # 专家并行度
PR=$7                          # 转换精度
USE_TE=$8                      # 是否使用Transformer Engine建模
mg2hf=$9                       # 是否执行mcore2hf转换
HG_CKPT_PATH=${10}             # HF的CKPT的路径

例如，使用下述脚本将checkpoint转换到MCore-Dense并检查输出

cd /workspace/Pai-Megatron-Patch/toolkits/model_checkpoints_convertor/qwen
bash hf2mcore_qwen2_convertor.sh \
0.5B \
/mnt/qwen-ckpts/Qwen2-0.5B \
/mnt/qwen-ckpts/Qwen2-0.5B-hf-to-mcore-te-tp1-pp1  \
1  \
1  \
1 \
fp32 \
true \
false

还使用下述脚本将checkpoint转换到MCore-MoE并检查输出

cd /workspace/Pai-Megatron-Patch/toolkits/model_checkpoints_convertor/qwen \
sh hf2mcore_qwen2_convertor.sh \
A14B \
/mnt/qwen-ckpts/Qwen2-57B-A14B \
/mnt/qwen-ckpts/Qwen2-57B-A14B-hf-to-mcore-te-tp4-pp1-ep4  \
4  \
1  \
4 \
fp32 \
true \
false

Megatron-Core预训练及指令微调

在Qwen2中，我们已将预训练和微调整合到run_mcore_qwen.sh脚本，对于不同的使用场景，二者各参数的意义有所不同。

预训练&微调命令统一描述

需要传入的参数列表如下：

ENV=$1                          # 运行环境配置开关: dsw单机训练训练，dlc表示多机训练环境
MODEL_SIZE=$2                   # 模型结构参数量级: 7B, 72B, A14B
BATCH_SIZE=$3                   # 一次迭代一个数据并行内的样本数
GLOBAL_BATCH_SIZE=$4            # 一次迭代多个数据并行的总样本数
LR=$5                           # 学习率
MIN_LR=$6                       # 最小学习率
SEQ_LEN=$7                      # 序列长度
PAD_LEN=$8                      # Padding长度
PR=${9}                         # 训练精度: fp16, bf16, fp8
TP=${10}                        # 模型并行度
PP=${11}                        # 流水并行度
CP=${12}                        # 上下文并行度
EP=${13}                        # 专家并行度
SP=${14}                        # 是否使用序列并行: true, false
DO=${15}                        # 是否使用Megatron版Zero-1降显存优化器: true, false
FL=${16}                        # 是否优先使用Flash Attention: true, false
SFT=${17}                       # 是否执行微调训练: true, false
AC=${18}                        # 激活检查点模式: sel, full, offload, false
OPTIMIZER_OFFLOAD=${19}         # 是否启用Offload optimizer: false, static, auto
SAVE_INTERVAL=${20}             # 保存ckpt的间隔
DATASET_PATH=${21}              # 训练数据集路径
VALID_DATASET_PATH=${22}        # 验证数据集路径
PRETRAIN_CHECKPOINT_PATH=${23}  # 预训练模型路径
TRAIN_TOKENS_OR_ITERS=${24}     # 训练TOKEN或者Iter数
WARMUP_TOKENS_OR_ITERS=${25}    # 预热TOKEN或者Iter数        
OUTPUT_BASEPATH=${26}           # 训练输出日志文件路径

预训练示例

使用以下命令启动对qwen2的继续预训练。 备注：当AC=offload或full时，可设置MP_AC_LAYERS环境变量来控制Checkpointing或Offload的TransformerLayer层数（默认值：1）。

cd /workspace/Pai-Megatron-Patch/examples/qwen2
sh run_mcore_qwen.sh  \
dsw  \
0.5B   \
1    \
8 \
1e-5   \
1e-6   \
128  \
128  \
bf16  \
1   \
1  \
1 \
1 \
true \
true   \
true \
false \
false   \
false \
100000  \
/mnt/qwen-datasets/wudao_qwenbpe_text_document   \
/mnt/qwen-datasets/wudao_qwenbpe_text_document   \
/mnt/qwen-ckpts/Qwen2-0.5B-hf-to-mcore-te-tp1-pp1  \
10000  \
100   \
/workspace/output_mcore_qwen2_pretrain

指令微调示例

制作idxmap用于微调的数据集可以参考链接。 当准备好微调数据集后，将SFT开关设置为true即可进行指令微调。

cd /workspace/Pai-Megatron-Patch/examples/qwen2
sh run_mcore_qwen.sh  \
dsw  \
0.5B   \
1    \
8 \
1e-5   \
1e-6   \
128  \
128  \
bf16  \
1   \
1  \
1 \
1 \
true \
true   \
true \
true \
false   \
false \
100000  \
/mnt/qwen-datasets/path_to_your_dataset   \
/mnt/qwen-datasets/path_to_your_dataset   \
/path/to/pretraining/checkpoint  \
10000  \
100   \
/workspace/output_mcore_qwen2_finetune

通过设置MP_DATASET_TYPE环境变量，本脚本还可使用json格式的数据集进行指令微调

export MP_DATASET_TYPE="raw"
cd /workspace/Pai-Megatron-Patch/examples/qwen2
sh run_mcore_qwen.sh  \
dsw  \
0.5B   \
1    \
8 \
1e-5   \
1e-6   \
128  \
128  \
bf16  \
1   \
1  \
1 \
1 \
true \
true   \
true \
true \
false   \
false \
100000  \
/mnt/qwen-datasets/alpaca_zh-qwen-train.json    \
/mnt/qwen-datasets/alpaca_zh-qwen-valid.json   \
/mnt/qwen-ckpts/Qwen2-0.5B-hf-to-mcore-te-tp1-pp1  \
10000  \
100   \
/workspace/output_mcore_qwen2_finetune

下游任务评估

评估格式转换

您需要将训练/微调后保存的Megatron-Core转换为HuggingFace格式来进行推理评估。

cd /workspace/Pai-Megatron-Patch/toolkits/model_checkpoints_convertor/qwen
bash hf2mcore_qwen2_convertor.sh \
0.5B \
/mnt/qwen-ckpts/Qwen2-0.5B-hf-to-mcore-te-tp1-pp1  \
/mnt/qwen-ckpts/Qwen2-0.5B-mcore-te-to-hf    \
1  \
1  \
1 \
fp32 \
true \
true \
/mnt/qwen-ckpts/Qwen2-0.5B

运行评估工具

下载评估数据

# In container
cd /workspace

wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/evaluation-datasets/evaluate.tgz 
wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/evaluation-datasets/cmmlu.tgz 
wget https://atp-modelzoo-wlcb-pai.oss-cn-wulanchabu.aliyuncs.com/release/models/pai-megatron-patch/evaluation-datasets/ceval.tgz 

tar -xvzf cmmlu.tgz 
tar -xvzf ceval.tgz 
tar -xvzf evaluate.tgz

运行以下指令对转换后的模型进行评估。

cd /workspace/Pai-Megatron-Patch/LM-Evaluation-Harness-240310
accelerate launch --main_process_port 29051 -m lm_eval \
--model hf \
--model_args pretrained=/mnt/qwen-ckpts/Qwen2-0.5B-mcore-te-to-hf,trust_remote_code=True \
--tasks cmmlu,ceval-valid  \
--batch_size 16