Skip to content

GitLab

"src/targets/gpu/vscode:/vscode.git/clone" did not exist on "1354c869b41e3e60d6477613e91575dc3933e7d8"
Commit ea2d13c2 authored by zhaoying1's avatar zhaoying1
Browse files

added llama_tencentpretrain_pytorch

parents
Pipeline #548 failed with stages
in 0 seconds
Expand all Show whitespace changes
Inline Side-by-side
Showing with 410 additions and 0 deletions
Dockerfile 0 → 100644
View file @ ea2d13c2
FROM image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04-py37-latest
COPY requirements.txt requirements.txt
RUN source /opt/dtk-23.04/env.sh
RUN cp /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && echo 'Asia/Shanghai' >/etc/timezone
ENV LANG C.UTF-8
RUN pip install -r requirements.txt -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
README.md 0 → 100644
View file @ ea2d13c2
## 基于TencentPretrain框架的LLaMA微调训练
## 模型介绍
LLaMA，这是一个基础语言模型的集合，参数范围从7B到65B。在数万亿的tokens上训练出的模型，并表明可以专门使用公开可用的数据集来训练最先进的模型，而不依赖于专有的和不可访问的数据集。特别是，llama 13B在大多数基准测试中优于GPT-3 (175B)， LLaMA 65B与最好的模型Chinchilla-70B和PaLM-540B具有竞争力。
LLAMA网络基于 Transformer 架构。提出了各种改进，并用于不同的模型，例如 PaLM。以下是与原始架构的主要区别：
**预归一化**。为了提高训练稳定性，对每个transformer 子层的输入进行归一化，而不是对输出进行归一化。使用 RMSNorm 归一化函数。
**SwiGLU 激活函数 [PaLM]**。使用 SwiGLU 激活函数替换 ReLU 非线性以提高性能。使用 2 /3 4d 的维度而不是 PaLM 中的 4d。
**旋转嵌入**。移除了绝对位置嵌入，而是添加了旋转位置嵌入 (RoPE)，在网络的每一层。
LLaMA 2是LLaMA的新一代版本，具有商业友好的许可证。 LLaMA 2 有 3 种不同的尺寸：7B、13B 和 70B。Llama 2训练语料相比LLaMA多出40%，上下文长度是由之前的2048升级到4096，可以理解和生成更长的文本。Llama 2采用了 Llama 1 的大部分预训练设置和模型架构，使用标准Transformer 架构，使用 RMSNorm 应用预归一化、使用 SwiGLU 激活函数和旋转位置嵌入RoPE。具体细节可参考论文：
[Llama 2: Open Foundation and Fine-Tuned Chat Models](https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/)
[LLaMA: Open and Efficient Foundation Language Models](https://arxiv.org/pdf/2302.13971v1.pdf)
### 依赖环境
* Python >= 3.6
* [torch >= 1.1](https://cancon.hpccube.com:65024/4/main/pytorch/dtk23.04)
* six >= 1.12.0
* argparse
* packaging
* regex
* [DeepSpeed](https://cancon.hpccube.com:65024/4/main/deepspeed/dtk23.04)
推荐使用docker方式运行，提供[光源](https://www.sourcefind.cn/#/service-list)拉取的docker镜像：image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04-py37-latest
#### Docker配置方式
```commandline
docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04-py37-latest
docker run -dit --network=host --name=llama-tencentpretrain --privileged --device=/dev/kfd --device=/dev/dri --ipc=host --shm-size=16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -u root --ulimit stack=-1:-1 --ulimit memlock=-1:-1 image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04-py37-latest
docker exec -it llama-tencentpretrain /bin/bash
pip install -r requirements.txt -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com
```
也可从目录下的Dockerfile构建镜像：
```commandline
docker build -t llama:latest .
docker run -dit --network=host --name=llama-tencentpretrain --privileged --device=/dev/kfd --device=/dev/dri --ipc=host --shm-size=16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -u root --ulimit stack=-1:-1 --ulimit memlock=-1:-1 llama:latest
```
### 模型权重下载
1. 方式一：下载huggingface格式模型。以 7B 模型为例，首先下载预训练[LLaMA权重](https://huggingface.co/decapoda-research/llama-7b-hf)，转换到TencentPretrain格式：
```commandline
python3 scripts/convert_llama_from_huggingface_to_tencentpretrain.py --input_model_path $LLaMA_HF_PATH \
--output_model_path models/llama-7b.bin --type 7B
```
2. 方式二：也可以直接下载[TencentPretrain对应格式模型](https://huggingface.co/Linly-AI/)进行微调训练，不需要转换格式。
### 全参数增量预训练
##### 数据预处理
1. 构建预训练数据集
txt预训练语料：多个txt需要合并到一个 .txt 文件并按行随机打乱，语料格式如下：
```commandline
doc1
doc2
doc3
```
jsonl 预训练语料：为了支持代码等包含换行符的数据，预训练数据也可以整理成jsonl格式，格式如下：
```commandline
{"text": "doc1"}
{"text": "doc2"}
{"text": "doc3"}
```
2. 按如下方式进行预处理
```commandline
python3 preprocess.py --corpus_path $CORPUS_PATH --spm_model_path $LLaMA_PATH/tokenizer.model \
--dataset_path $OUTPUT_DATASET_PATH --data_processor lm --seq_length 1024
```
可选参数： --json_format_corpus：使用jsonl格式数据；
--full_sentences：对长度不足的样本使用其他样本进行填充（没有 pad token）；
##### 增量预训练
1. 单机
```commandline
deepspeed pretrain.py --deepspeed --deepspeed_config models/deepspeed_zero3_config.json --enable_zero3 \
--pretrained_model_path models/llama-7b.bin \
--dataset_path $OUTPUT_DATASET_PATH --spm_model_path $LLaMA_PATH/tokenizer.model \
--config_path models/llama/7b_config.json \
--output_model_path models/llama_zh_7b \
--world_size 8 --data_processor lm --deepspeed_checkpoint_activations \
--total_steps 300000 --save_checkpoint_steps 5000 --batch_size 24
```
2. 集群
```commandline
cd slurm_scripts
bash run-pt.sh
```
### 全参数指令微调
##### 数据预处理
1. 构建指令数据集：指令数据为 json 格式，包含instruction、input、output三个字段（可以为空），每行一条样本。
示例：
```commandline
{"instruction": "在以下文本中提取所有的日期。", "input": "6月21日是夏至，这是一年中白天最长的一天。", "output": "6月21日"}
{"instruction": "", "input": "请生成一个新闻标题，描述一场正在发生的大型自然灾害。\\n\n", "output": "\"强烈飓风肆虐，数百万人疏散！\""}
```
2. 按如下方式进行预处理
```commandline
python3 preprocess.py --corpus_path $INSTRUCTION_PATH --spm_model_path $LLaMA_PATH/tokenizer.model \
--dataset_path $OUTPUT_DATASET_PATH --data_processor alpaca --seq_length 1024
```
##### 微调训练
1. 单机
```commandline
deepspeed pretrain.py --deepspeed --deepspeed_config models/deepspeed_zero3_config.json --enable_zero3 \
--pretrained_model_path models/llama_zh_7b.bin \
--dataset_path $OUTPUT_DATASET_PATH --spm_model_path $LLaMA_PATH/tokenizer.model \
--config_path models/llama/7b_config.json \
--output_model_path models/chatflow_7b \
--world_size 8 --data_processor alpaca --prefix_lm_loss --deepspeed_checkpoint_activations \
--total_steps 20000 --save_checkpoint_steps 2000 --batch_size 24
```
2. 集群
```commandline
cd slurm_scripts
bash run-ift.sh
```
### 模型分块
训练初始化时，每张卡会加载一个模型的拷贝，因此内存需求为模型大小*GPU数量。内存不足时可以通过以下方式将模型分块，然后使用分块加载。
```commandline
python3 scripts/convert_model_into_blocks.py \
--input_model_path path/to/chinese_llama_13b.bin \
--output_model_path path/to/chinese_llama_13b \
--block_size 10
```
其中，--input_model_path 输入模型路径； --output_model_path 输出模型目录； --block_size 分块大小；在训练加载模型时，将 pretrained_model_path 改为以上输出的目录即可。
### 模型推理
TencentPretrain格式模型推理请参考[LLAMA_pytorch](https://developer.hpccube.com/codes/modelzoo/llama_pytorch)
### 训练实验结果
- 利用公开指令数据集[alpaca_gpt4_data_zh.json](https://huggingface.co/datasets/shibing624/alpaca-zh)，基于汉化ChineseLLaMA的7B、13B基础模型，我们进行指令微调训练实验，以下为训练Loss：
<div align="center">
<figure class="half">
<img width = '300' height ='250' src="./data/media/ift_7B_bs2_32node_128cards.jpg">
<img width = '300' height ='250' src="./data/media/ift_13B_bs2_32node_128cards.jpg">
</figure>
</div>
- 利用公开指令数据集[alpaca_gpt4_data_zh.json](https://huggingface.co/datasets/shibing624/alpaca-zh)，基于meta开源的[meta-llama/Llama-2-7b-chat-hf](https://pan.xunlei.com/s/VN_kQa1_HBvV-X9QVI6jV2kOA1?pwd=xmra) ，我们进行中文指令微调训练实验，以下为训练Loss：
<div align="center">
<img src="./data/media/ift_llama2_7B_bs2_32node_128cards.jpg" width="300" height="250">
</div>
## 源码仓库及问题反馈
- https://developer.hpccube.com/codes/modelzoo/llama1-2
## 参考
* https://github.com/CVI-SZU/Linly
* https://github.com/Tencent/TencentPretrain/
* https://github.com/ProjectD-AI/llama_inference
\ No newline at end of file
data/alpaca_gpt4_data_zh.json 0 → 100644
View file @ ea2d13c2
This diff is collapsed.
data/dataset.pt 0 → 100644
View file @ ea2d13c2
File added
model.properties 0 → 100644
View file @ ea2d13c2
# 模型名称
modelName=LLaMA(TencentPretrain)_Pytorch
# 模型描述
modelDescription=基于Pytorch框架的LLaMA微调训练
# 应用场景(多个标签以英文逗号分割)
appScenario=训练,train,nlp,智能聊天助手
# 框架类型(多个标签以英文逗号分割)
frameType=Pytorch
models/albert/base_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 128,
"feedforward_size": 3072,
"hidden_size": 768,
"hidden_act": "relu",
"heads_num": 12,
"layers_num": 12,
"max_seq_length": 512,
"dropout": 0.0,
"data_processor": "albert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"factorized_embedding_parameterization": true,
"parameter_sharing": true,
"target": ["mlm", "sp"]
}
\ No newline at end of file
models/albert/large_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 128,
"feedforward_size": 4096,
"hidden_size": 1024,
"hidden_act": "relu",
"heads_num": 16,
"layers_num": 24,
"max_seq_length": 512,
"dropout": 0.0,
"data_processor": "albert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"factorized_embedding_parameterization": true,
"parameter_sharing": true,
"target": ["mlm", "sp"]
}
\ No newline at end of file
models/albert/xlarge_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 128,
"feedforward_size": 8192,
"hidden_size": 2048,
"hidden_act": "relu",
"heads_num": 16,
"layers_num": 24,
"max_seq_length": 512,
"dropout": 0.0,
"data_processor": "albert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"factorized_embedding_parameterization": true,
"parameter_sharing": true,
"target": ["mlm", "sp"]
}
\ No newline at end of file
models/albert/xxlarge_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 128,
"feedforward_size": 16384,
"hidden_size": 4096,
"hidden_act": "relu",
"heads_num": 16,
"layers_num": 12,
"max_seq_length": 512,
"dropout": 0.0,
"data_processor": "albert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"factorized_embedding_parameterization": true,
"parameter_sharing": true,
"target": ["mlm", "sp"]
}
\ No newline at end of file
models/bart/base_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 768,
"feedforward_size": 3072,
"hidden_size": 768,
"hidden_act": "gelu",
"heads_num": 12,
"layers_num": 6,
"decoder_layers_num": 6,
"max_seq_length": 1024,
"dropout": 0.1,
"data_processor": "bart",
"embedding": ["word", "pos"],
"tgt_embedding": ["word", "pos"],
"share_embedding": true,
"encoder": "transformer",
"mask": "fully_visible",
"decoder": "transformer",
"target": ["lm"],
"tie_weights": true,
"has_lmtarget_bias": true
}
\ No newline at end of file
models/bart/large_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 1024,
"feedforward_size": 4096,
"hidden_size": 1024,
"hidden_act": "gelu",
"heads_num": 16,
"layers_num": 12,
"decoder_layers_num": 12,
"max_seq_length": 1024,
"dropout": 0.1,
"data_processor": "bart",
"embedding": ["word", "pos"],
"tgt_embedding": ["word", "pos"],
"share_embedding": true,
"encoder": "transformer",
"mask": "fully_visible",
"decoder": "transformer",
"target": ["lm"],
"tie_weights": true,
"has_lmtarget_bias": true
}
\ No newline at end of file
models/beit/base_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 768,
"feedforward_size": 3072,
"hidden_size": 768,
"hidden_act": "gelu",
"heads_num": 12,
"layers_num": 12,
"dropout": 0.1,
"data_processor": "beit",
"embedding": ["masked_patch", "pos"],
"encoder": "transformer",
"mask": "fully_visible",
"target": ["mlm"],
"image_height": 256,
"image_width": 256,
"patch_size": 16,
"image_preprocess": ["crop"],
"tokenizer": "vqgan",
"image_tokenizer": {
"is_gumbel": false,
"is_transformer": false,
"image_vocab_size": 16384,
"frame_size": 16
}
}
\ No newline at end of file
models/bert/base_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 768,
"feedforward_size": 3072,
"hidden_size": 768,
"hidden_act": "gelu",
"heads_num": 12,
"layers_num": 12,
"max_seq_length": 512,
"dropout": 0.1,
"data_processor": "bert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"target": ["mlm", "sp"],
"tie_weights": true
}
\ No newline at end of file
models/bert/large_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 1024,
"feedforward_size": 4096,
"hidden_size": 1024,
"hidden_act": "gelu",
"heads_num": 16,
"layers_num": 24,
"max_seq_length": 512,
"dropout": 0.1,
"data_processor": "bert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"target": ["mlm", "sp"],
"tie_weights": true
}
\ No newline at end of file
models/bert/medium_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 512,
"feedforward_size": 2048,
"hidden_size": 512,
"hidden_act": "gelu",
"heads_num": 8,
"layers_num": 8,
"max_seq_length": 512,
"dropout": 0.1,
"data_processor": "bert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"target": ["mlm", "sp"],
"tie_weights": true
}
\ No newline at end of file
models/bert/mini_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 256,
"feedforward_size": 1024,
"hidden_size": 256,
"hidden_act": "gelu",
"heads_num": 4,
"layers_num": 4,
"max_seq_length": 512,
"dropout": 0.1,
"data_processor": "bert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"target": ["mlm", "sp"],
"tie_weights": true
}
\ No newline at end of file
models/bert/small_config.json 0 → 100644
View file @ ea2d13c2
{
"emb_size": 512,
"feedforward_size": 2048,
"hidden_size": 512,
"hidden_act": "gelu",
"heads_num": 8,
"layers_num": 4,
"max_seq_length": 512,
"dropout": 0.1,
"data_processor": "bert",
"embedding": ["word", "pos", "seg"],
"encoder": "transformer",
"mask": "fully_visible",
"target": ["mlm", "sp"],
"tie_weights": true
}
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment