星辰语义大模型 Telechat

@article{wang2024telechat,
      title={TeleChat Technical Report},
      author={Zihan Wang and Xinzhang Liu and Shixuan Liu and Yitong Yao and Yuyao Huang and Zhongjiang He and Xuelong Li and Yongxiang Li and Zhonghao Che and Zhaoxi Zhang and Yan Wang and Xin Wang and Luwen Pu and Huihan Xu and Ruiyu Fang and Yu Zhao and Jie Zhang and Xiaomeng Huang and Zhilong Lu and Jiaxin Peng and Wenjun Zheng and Shiquan Wang and Bingkai Yang and Xuewei he and Zhuoru Jiang and Qiyi Xie and Yanhan Zhang and Zhongqiu Li and Lingling Shi and Weiwei Fu and Yin Zhang and Zilu Huang and Sishi Xiong and Yuxiang Zhang and Chao Wang and Shuangyong Song},
      journal={arXiv preprint arXiv:2401.03804},
      year={2024}
}

python mindformers/research/telechat/convert_weight_torch_to_ms.py \
--torch_path TORCH_CKPT_DIR \
--mindspore_path {path}/MS_CKPT_NAME \
--model_name 'telechat_7b'

python mindformers/research/telechat/convert_weight_torch_to_ms.py \
--torch_path TORCH_CKPT_DIR \
--mindspore_path {path}/MS_CKPT_NAME \
--model_name 'telechat_52b' \
--mp 8

# 参数说明
torch_path: torch版本权重保存目录路径
mindspore_path: 权重保存文件名，可以指定自定义保存路径
model_name: 模型的名称
mp: 目标切分个数，比如指定为8时，得到的完整权重适用于mp=8的权重自动转换；

python telechat_preprocess.py \
--input_dataset_file multi_turn_example.jsonl \
--output_path path/to/dataset_dir \
--vocab_file_path path/to/tokenizer.json \
--max_length 2048 \
--pad_token '<pad>'

python telechat_preprocess.py \
--input_dataset_file multi_turn_example.jsonl \
--output_path path/to/dataset_dir \
--vocab_file_path path/to/tokenizer.model \
--max_length 1024

python telechat_preprocess.py \
--input_dataset_file multi_turn_example.jsonl \
--output_path path/to/dataset_dir \
--vocab_file_path path/to/tokenizer.model \
--max_length 4096 \
--start_token '<reserve3>' \
--user_token '<reserve1>' \
--bot_token '<reserve2>' \
--pad_token '<pad>'

bash scripts/msrun_launcher.sh "python run_mindformer.py \
--config ./research/telechat/finetune_telechat_7b.yaml \
--train_dataset path/to/dataset_dir \
--load_checkpoint path/to/ckpt_path \
--auto_trans_ckpt True \
--use_parallel True \
--register_path ./research/telechat"

# 1号机器
bash scripts/msrun_launcher.sh "python run_mindformer.py \
--config telechat/finetune_telechat_52b.yaml \
--train_dataset path/to/dataset_dir \
--load_checkpoint path/to/ckpt_path \
--auto_trans_ckpt True \
--use_parallel True \
--register_path ./research/telechat" 16 8 机器IP 8118 0 output/msrun_log False 300

# 2号机器
bash scripts/msrun_launcher.sh "python run_mindformer.py \
--config telechat/finetune_telechat_52b.yaml \
--train_dataset path/to/dataset_dir \
--load_checkpoint path/to/ckpt_path \
--auto_trans_ckpt True \
--use_parallel True \
--register_path ./research/telechat" 16 8 机器IP 8118 1 output/msrun_log False 300

# 参数说明
config: 配置文件路径
train_dataset: 数据集路径
load_checkpoint: 权重路径
auto_tans_ckpt: 权重自动转换开关
use_parallel: 并行模式开关
register_path: 外部模型注册路径

{"input": "生抽与老抽的区别？"}

processor:
  tokenizer:
    tokenizer_file: 'path/to/tokenizer.json'

python run_mindformer.py \
--config ./research/telechat/predict_telechat_7b.yaml \
--load_checkpoint path/to/ckpt_path \
--predict_data "<_user>生抽与老抽的区别？<_bot>" \
--register_path ./research/telechat

processor:
  tokenizer:
    vocab_file: 'path/to/tokenizer.model'

python run_mindformer.py \
--config ./research/telechat/predict_telechat_12b.yaml \
--load_checkpoint path/to/ckpt_path \
--predict_data "<_user>生抽与老抽的区别？<_bot>" \
--register_path ./research/telechat

processor:
  tokenizer:
    vocab_file: 'path/to/tokenizer.model'

bash scripts/msrun_launcher.sh "python run_mindformer.py \
--config ./research/telechat/predict_telechat_52b.yaml \
--load_checkpoint path/to/ckpt_path \
--predict_data '<reserve3><reserve1>生抽与老抽的区别？<reserve2>' \
--auto_trans_ckpt True \
--register_path ./research/telechat" 4

# 参数说明
input_file: 多轮对话文件路径
train_dataset: 数据集路径
load_checkpoint: 权重路径
auto_tans_ckpt: 权重自动转换开关
use_parallel: 并行模式开关

生抽与老抽的区别？
答：生抽和老抽是两种不同的酱油，它们的主要区别在于颜色、味道和用途。
1.颜色：生抽的颜色较浅，呈红褐色;老抽的颜色较深，呈棕褐色。
2.味道：生抽的味道较咸，鲜味较浓;老抽的味道较甜，香味较浓。
3.用途：生抽主要用于调味，可以增加菜肴的鲜味和咸味;老抽主要用于上色，可以使菜肴的颜色更加鲜艳。
总的来说，生抽和老抽在颜色、味道和用途上都有所不同，可以根据个人口味和烹饪需求选择使用。