|
--- |
|
frameworks: |
|
- Pytorch |
|
license: llama3 |
|
tasks: |
|
- text-generation |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
--- |
|
# 魔搭Llama3 8b中文Agent智能体模型 |
|
|
|
本模型使用Llama3-8b-instruct基模型进行训练,适配中文通用场景,且支持ReACT格式的Agent调用。 |
|
|
|
## 模型使用 |
|
|
|
### 推理 |
|
|
|
```shell |
|
# 安装依赖 |
|
pip install ms-swift -U |
|
``` |
|
|
|
```python |
|
# 推理 |
|
swift infer --model_type llama3-8b-instruct --model_id_or_path swift/Llama3-Chinese-8B-Instruct-Agent-v1 |
|
``` |
|
|
|
```shell |
|
# 部署 |
|
swift deploy --model_type llama3-8b-instruct --model_id_or_path swift/Llama3-Chinese-8B-Instruct-Agent-v1 |
|
``` |
|
|
|
本模型可以联合ModelScopeAgent框架使用,请参考: |
|
|
|
https://github.com/modelscope/swift/blob/main/docs/source/LLM/Agent%E5%BE%AE%E8%B0%83%E6%9C%80%E4%BD%B3%E5%AE%9E%E8%B7%B5.md#%E6%90%AD%E9%85%8Dmodelscope-agent%E4%BD%BF%E7%94%A8 |
|
|
|
### 模型训练信息 |
|
|
|
为了适配中文及Agent场景,我们针对语料进行了一定混合配比,训练Llama3使用的语料如下: |
|
|
|
\- COIG-CQIA:https://modelscope.cn/datasets/AI-ModelScope/COIG-CQIA/summary 该数据集包含了中国传统知识、豆瓣、弱智吧、知乎等中文互联网信息 |
|
|
|
\- 魔搭通用Agent训练数据集: https://modelscope.cn/datasets/AI-ModelScope/ms-agent-for-agentfabric/summary |
|
|
|
\- alpaca-en: https://modelscope.cn/datasets/AI-ModelScope/alpaca-gpt4-data-en/summary |
|
|
|
\- ms-bench魔搭通用中文问答数据集: https://modelscope.cn/datasets/iic/ms_bench/summary |
|
|
|
| **超参数** | **值** | |
|
| --------------------------- | ------ | |
|
| lr | 5e-5 | |
|
| epoch | 2 | |
|
| lora_rank | 8 | |
|
| lora_alpha | 32 | |
|
| lora_target_modules | ALL | |
|
| batch_size | 2 | |
|
| gradient_accumulation_steps | 16 | |
|
|
|
## 模型训练命令 |
|
|
|
```shell |
|
NPROC_PER_NODE=8 \ |
|
swift sft \ |
|
--model_type llama3-8b-instruct \ |
|
--dataset ms-agent-for-agentfabric-default alpaca-en ms-bench ms-agent-for-agentfabric-addition coig-cqia-ruozhiba coig-cqia-zhihu coig-cqia-exam coig-cqia-chinese-traditional coig-cqia-logi-qa coig-cqia-segmentfault coig-cqia-wiki \ |
|
--batch_size 2 \ |
|
--max_length 2048 \ |
|
--use_loss_scale true \ |
|
--gradient_accumulation_steps 16 \ |
|
--learning_rate 5e-5 \ |
|
--use_flash_attn true \ |
|
--eval_steps 500 \ |
|
--save_steps 500 \ |
|
--train_dataset_sample -1 \ |
|
--dataset_test_ratio 0.1 \ |
|
--val_dataset_sample 10000 \ |
|
--num_train_epochs 2 \ |
|
--check_dataset_strategy none \ |
|
--gradient_checkpointing true \ |
|
--weight_decay 0.01 \ |
|
--warmup_ratio 0.03 \ |
|
--save_total_limit 2 \ |
|
--logging_steps 10 \ |
|
--sft_type lora \ |
|
--lora_target_modules ALL \ |
|
--lora_rank 8 \ |
|
--lora_alpha 32 |
|
``` |
|
|
|
## 模型评测信息 |
|
|
|
| 评测模型 | ARC | CEVAL | GSM8K | |
|
| ----------------------------------- | ------ | ------ | ------ | |
|
| Llama3-8b-instruct | 0.7645 | 0.5089 | 0.7475 | |
|
| Llama3-Chinese-8B-Instruct-Agent-v1 | 0.7577 | 0.4903 | 0.652 | |
|
|
|
GSM8K英文数学能力下降了8个点左右,经过消融实验我们发现去除alpaca-en语料会导致GSM8K下降至少十个点以上。 |
|
|