README.md 6.19 KB
Newer Older
liangjing's avatar
v1  
liangjing committed
1
2
3
# 内容
- [内容](#内容)
- [环境配置](#环境配置)
wxj's avatar
wxj committed
4
5
6
7
8
9
10
11
12
13
14
- [预训练](#预训练)
  - [GPT](##GPT)
    - [下载词汇文件](###下载词汇文件)
    - [下载训练数据](###下载训练数据)
    - [数据预处理](###数据预处理)
    - [GPT预训练](###GPT预训练)
  - [Llama](##Llama)
    - [下载tokenizer文件](###下载tokenizer文件)
    - [下载训练数据](###下载训练数据)
    - [数据预处理](###数据预处理)
    - [Llama预训练](###Llama预训练)
liangjing's avatar
v1  
liangjing committed
15
16
- [参考](#参考)

wxj's avatar
wxj committed
17
18
19
20
21
# 更新日志

2024.12.16适配了torch prof

使用方法: 启动脚本中添加下列参数, 即可采集对应的prof信息
silencealiang's avatar
silencealiang committed
22
23
24
25
26
27
28
29
30

```python
# 采集torchprof
mpirun -np 8 --allow-run-as-root train_mixtral_8x7B_1nodes.sh localhost --profiling=torch

# 采集hipprof
mpirun -np 8 --allow-run-as-root train_mixtral_8x7B_1nodes.sh localhost --profiling=hip
```

wxj's avatar
wxj committed
31
```bash
silencealiang's avatar
silencealiang committed
32
33
# prof相关参数
TORCH_PROFIE_ARGS=(
wxj's avatar
wxj committed
34
35
36
37
38
39
40
    --profile # 开启profile
    --profile-step-start 4 # skip前3个iter, warm第4个iter
    --profile-step-end 5 # 采集第5个iter
    --use-pytorch-profiler # 使用torch prof
    --profile-ranks 0 3 # 采集全局rank 第0和3
    --profile-dir ./prof_data # prof文件的保存目录
)
silencealiang's avatar
silencealiang committed
41
42
43
44
45
46
47
48

HIP_PROFIE_ARGS=(
    --profile
    --profile-ranks 0 1 2 3 4 5 6 7
    --profile-step-start 4
    --profile-step-end 5
    --use-hip-profiler
)
wxj's avatar
wxj committed
49
50
51
```


liangjing's avatar
v1  
liangjing committed
52
53
# 环境配置
1. 安装基础依赖包
Neel Kant's avatar
Neel Kant committed
54
<pre>
liangjing's avatar
v1  
liangjing committed
55
pip install -r requirements.txt
Neel Kant's avatar
Neel Kant committed
56
</pre>
wxj's avatar
wxj committed
57
2. 安装HCU相关whl包
Neel Kant's avatar
Neel Kant committed
58

wxj's avatar
wxj committed
59
HCU相关包下载目录:[https://cancon.hpccube.com:65024/4/main](https://cancon.hpccube.com:65024/4/main)
Neel Kant's avatar
Neel Kant committed
60

wxj's avatar
wxj committed
61
pytorch whl包:pytorch ---> dtk-24.04.1
liangjing's avatar
v1  
liangjing committed
62
根据python版本,下载对应pytorch的whl包
Neel Kant's avatar
Neel Kant committed
63
64

<pre>
liangjing's avatar
v1  
liangjing committed
65
pip install torch* (下载的torch的whl包)
Neel Kant's avatar
Neel Kant committed
66
</pre>
wxj's avatar
wxj committed
67
torchvision whl包:vision ---> dtk-24.04.1
liangjing's avatar
v1  
liangjing committed
68
根据python版本,下载对应torchvision的whl包
Mohammad's avatar
Mohammad committed
69
70

<pre>
liangjing's avatar
v1  
liangjing committed
71
pip install torchvision* (下载的torchvision的whl包)
Mohammad's avatar
Mohammad committed
72
</pre>
wxj's avatar
wxj committed
73
apex whl包:apex ---> dtk-24.04.1
liangjing's avatar
v1  
liangjing committed
74
根据python版本,下载对应apex的whl包
Mohammad's avatar
Mohammad committed
75
76

<pre>
liangjing's avatar
v1  
liangjing committed
77
pip install apex* (下载的apex的whl包)
78
</pre>
wxj's avatar
wxj committed
79

liangjing's avatar
v1  
liangjing committed
80
若使用 pip install 下载安装过慢,可添加源:-i https://pypi.tuna.tsinghua.edu.cn/simple/
Mohammad's avatar
Mohammad committed
81

wxj's avatar
wxj committed
82
83
84
# 预训练
## GPT
### 下载词汇文件
85

Mohammad's avatar
Mohammad committed
86
<pre>
liangjing's avatar
v1  
liangjing committed
87
88
wget https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-vocab.json
wget https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-merges.txt
Mohammad's avatar
Mohammad committed
89
</pre>
90

wxj's avatar
wxj committed
91
### 下载训练数据
liangjing's avatar
v1  
liangjing committed
92
使用1GB 79K jsonl数据集
Mohammad's avatar
Mohammad committed
93
<pre>
liangjing's avatar
v1  
liangjing committed
94
95
wget https://huggingface.co/bigscience/misc-test-data/resolve/main/stas/oscar-1GB.jsonl.xz
xz -d oscar-1GB.jsonl.xz
Mohammad's avatar
Mohammad committed
96
</pre>
wxj's avatar
wxj committed
97
解压后为单个`oscar-1GB.jsonl`文件
Mohammad's avatar
Mohammad committed
98

wxj's avatar
wxj committed
99
### 数据预处理
Mohammad's avatar
Mohammad committed
100

wxj's avatar
wxj committed
101
```shell
liangjing's avatar
v1  
liangjing committed
102
103
python tools/preprocess_data.py \
    --input oscar-1GB.jsonl \ 
wxj's avatar
wxj committed
104
105
    --output-prefix ./dataset/oscar-1GB-gpt \
    --vocab-file gpt2-vocab.json \
liangjing's avatar
v1  
liangjing committed
106
107
108
109
    --tokenizer-type GPT2BPETokenizer \
    --merge-file gpt2-merges.txt \
    --append-eod \
    --workers 8
Mohammad's avatar
Mohammad committed
110

wxj's avatar
wxj committed
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
# 参数说明
# --input				输入数据集路径,即oscar-1GB.jsonl.xz解压后的文件路径
# --output-prefix		输出数据路径(需要输出目录已创建),处理后会自动加上_text_document后缀
# --vocab-file				下载的gpt2-vocab.json词表文件路径
# --tokenizer-type 	tokenizer类型
# --merge-file		下载的gpt2-merges.txt文件路径		
# --append-eod		添加结束标志符		
# --workers			进程数
```


### GPT预训练
脚本: `GPT_pretraining.sh`

修改数据集与词汇文件路径
```shell
VOCAB_FILE=gpt2-vocab.json
MERGE_FILE=gpt2-merges.txt
DATA_PATH="./dataset/oscar-1GB-gpt_text_document"
```
- 单机多卡训练
  ```shell
  # 修改脚本中的分布式启动参数
  # 单机可以使用localhost指定通信地址为本地
  # -np 8指定8进程\(8卡\)并行
  # --allow-run-as-root以root权限启动
  mpirun --allow-run-as-root -np 8 GPT_pretraining.sh localhost >& GPT_pretraining.log
liangjing's avatar
v1  
liangjing committed
138
  ```
wxj's avatar
wxj committed
139
140
141
142
143
144
145
146
147
148
149
  注: 这里的`localhost`参数会传到脚本中的`--dist-url`

`GPT_pretraining.log`中查看训练日志

- 多机多卡训练
  
  多节点docker设置:
  1. 容器内执行/usr/sbin/sshd -p 12345,启动一个端口
  2. 容器间可通过该端口ssh登录,ssh ip -p 12345
  3. 如果需要免密,docker run容器时,docker -v /root/.ssh 挂载.ssh目录
  4. 容器间mpirun执行: `mpirun -np .. --hostfile hosts -mca plm_rsh_args "-p 12345" ./xx.sh master_ip`
Raul Puri's avatar
Raul Puri committed
150

wxj's avatar
wxj committed
151
152

  **案例**: 设有节点192.168.1.1和192.168.1.2两个节点, 每个节点上8张卡, 192.168.1.1作为master节点
153

wxj's avatar
wxj committed
154
155
156
157
  hosts文件:
  ```txt
  192.168.1.1 slots=8 
  192.168.1.2 slots=8
liangjing's avatar
v1  
liangjing committed
158
  ```
wxj's avatar
wxj committed
159
160
161
162

  在master节点执行命令

  ```shell
wxj's avatar
wxj committed
163
  mpirun --allow-run-as-root -np 16 --hostfile hosts -mca plm_rsh_no_tree_spawn 1 -mca plm_rsh_args "-p 12345" --bind-to none ./GPT_pretraining.sh 192.168.1.1 >& GPT_pretraining.log
liangjing's avatar
v1  
liangjing committed
164
  ```
wxj's avatar
wxj committed
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
`GPT_pretraining.log`中查看训练日志

## Llama
### 下载tokenizer文件

链接: https://www.modelscope.cn/models/shakechen/Llama-2-7b-hf/files
下载其中的tokenizer*文件

### 下载训练数据
使用1GB 79K jsonl数据集
<pre>
wget https://huggingface.co/bigscience/misc-test-data/resolve/main/stas/oscar-1GB.jsonl.xz
xz -d oscar-1GB.jsonl.xz
</pre>
解压后为单个`oscar-1GB.jsonl`文件

### 数据预处理

```shell
python tools/preprocess_data.py \
  --input oscar-1GB.jsonl \
  --output-prefix /datasets/oscar-1GB-llama\
  --tokenizer-type Llama2Tokenizer \
  --tokenizer-model /path/to/llama2_7b_hf/tokenizer.model \
  --workers 16 \
  --append-eod
```

### Llama预训练
脚本: `Llama_pretraining.sh`

修改数据集与tokenizer路径
```shell
DATA_PATH="/datasets/oscar-1GB-llama_text_document"
--tokenizer-model /path/to/llama2_7b_hf/tokenizer.model
```
- 单机多卡训练
  ```shell
  # 具体参数说明参考上文GPT
  mpirun --allow-run-as-root -np 8 Llama_pretraining.sh localhost >& Llama_pretraining.log
  ```
`Llama_pretraining.log`中查看训练日志

- 多机多卡训练
  
wxj's avatar
wxj committed
210
  **案例**: 设有节点192.168.1.1和192.168.1.2两个节点, 每个节点上8张卡, 192.168.1.1作为master节点
wxj's avatar
wxj committed
211
212
213
214
215
216

  hosts配置如上文GTP所示

  在master节点执行命令

  ```shell
wxj's avatar
wxj committed
217
  mpirun --allow-run-as-root -np 16 --hostfile hosts -mca plm_rsh_no_tree_spawn 1 -mca plm_rsh_args "-p 12345" --bind-to none ./Llama_pretraining.sh 192.168.1.1 >& Llama_pretraining.log
wxj's avatar
wxj committed
218
219
220
  ```

`Llama_pretraining.log`中查看训练日志
221

liangjing's avatar
v1  
liangjing committed
222
# 参考
223

silencealiang's avatar
silencealiang committed
224
- [README_ORIGIN](README_ORIGIN.md)