Baichuan2-Explained

Baichuan2代码的逐行解析版本，适合小白

• fine-tune/

  • fine-tune.py/
    • class ModelArguments : const model_name_or_path
    • class DataArguments : const data_path
    • class TrainingArguments ： const cache_dir、const optim、const model_max_length、 const use_lora、
    • class SupervisedDataset
  • func train

• web_demo.py/

  • ads_generation.md（分布式运行范例）

• ./Baichuan2-13B-Chat/modelling_baichuan.py

  • const logger
  • func _get_interleave
  • func _get_interleave_power_of_2
  • func _fill_with_neg_inf
  • func _buffered_future_mask
  • func _gen_alibi_mask
  • class RMSNorm
    • func init
    • func forward
  • class MLP
    • func init
    • func init
    • func _shape
    • func forward
  • class BaichuanLayer
    • func init
    • func forward
  • class BaichuanPreTrainedModel

• README.md

CSDN彩色博客版：

• ./Baichuan2-13B-Chat/modelling_baichuan.py
  • Baichuan2源码解析之：Baichuan2-13B-Chat/modelling_baichuan.py
• src/
  • CSDN彩色源码解析fine-tune/fine-tune.py (一)
    • common.py
    • peft_trainer.py
  • CSDN彩色源码解析web_demo.py
• README.md

ChatGLM Efficient Tuning源码解析train_sft.py https://zengxiaojian.blog.csdn.net/article/details/131458667

引用 - 源项目

@Misc{Baichuan2,
  title = {Baichuan2},
  author = {Baichuan2},
  howpublished = {\url{https://github.com/baichuan-inc/Baichuan2}},
  year = {2023}
}