tilearn-llm 0.7.1

TILEARN for LLM

Tilearn.llm使用说明

1. CUDA Kernel（以LLAMA为例）

支持显卡：Ampere, Ada, or Hopper GPUs (e.g., A100, A800, H100, H800)

Dependencies: pytorch >= 2.0.0

当前版本完全兼容huggingface接口，不需要额外的操作

LLAMA1/LLAMA2 A800 16GPU seq=1024相比deepspeed zero2训练加速约20%

cuda kernel使用方法-代码修改如下

### TILEARN.LLM
from tilearn.llm.transformers import LlamaForCausalLM

### 模型接口与标准huggingface一致
model = LlamaForCausalLM.from_pretrained(...)

或者使用AutoModelForCausalLM接口

### TILEARN.LLM
from tilearn.llm.transformers import AutoModelForCausalLM

### 模型接口与标准huggingface一致
model = AutoModelForCausalLM.from_pretrained(...)

特殊说明：

1、由于baichuan1 13B和baichuan2 13B会产生冲突，目前tilearn.llm.transformers.AutoModelForCausalLM默认开启了baichuan1 13B，如果需要使用baichuan2 13B，需要在启动训练脚本中设置环境变量：export TILEARN_LLM_BAICHUAN_13B=2

### TILEARN_LLM_BAICHUAN_13B open baichuan2 model
export TILEARN_LLM_BAICHUAN_13B=2

2、目前加速已经支持的模型列表：

# llama
from tilearn.llm.transformers.models.llama.modeling_llama import LlamaForCausalLM

# bloom
from tilearn.llm.transformers.models.bloom.modeling_bloom import BloomForCausalLM

# baichuan1
from tilearn.llm.transformers.models.baichuan.baichuan1_13B.modeling_baichuan import BaichuanForCausalLM
from tilearn.llm.transformers.models.baichuan.baichuan1_7B.modeling_baichuan import BaiChuanForCausalLM

# baichuan2
# 默认使用TILEARN.LLM且无需任何设置
# 单独使用xformer，需安装xformer且设置环境变量TIACC_TRAINING_CUDA_KERNEL=2
from tilearn.llm.transformers.models.baichuan.baichuan2_7B.modeling_baichuan import BaichuanForCausalLM
from tilearn.llm.transformers.models.baichuan.baichuan2_13B.modeling_baichuan import BaichuanForCausalLM

2. Static Zero

适用场景：在deepspeed zero1、zero2、zero3、offload、int8等不同优化状态间切换

启动脚本修改如下

### TILEARN STATIC ZERO
### Open: TIACC_TRAINING_CUDA_KERNEL='O2' 
### support 'O2' / 'O2.5' / 'O3' / 'O3.5' / 'O3_Q8'(doing)
### Close: TIACC_TRAINING_CUDA_KERNEL='None'
export TIACC_TRAINING_STATIC_ZERO='None' #'O2'

代码修改如下

from transformers import HfArgumentParser
from tilearn.llm.transformers import TrainingArguments
	
### 接口与标准huggingface一致
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))