Skip to content

生成模型 tokenizer训练,模型初始化,模型预训练,指令微调。llama,creek

Notifications You must be signed in to change notification settings

fisheeped/creek

Repository files navigation

🤗https://huggingface.co/maheer/creek

从零创建生成大模型

下面代码中的参数,资源占用:预训练A800*8,每张卡显存占用70G+。 可以调小batch_size,max_length,最少单卡12G显存应该能完成下面步骤。

train tokenizer

python model_init/tokenization/train_eval_tokenizer.py

init model

python model_init/model_init.py

pretrain

bash pretrain.sh

finetune

bash sft.sh

About

生成模型 tokenizer训练,模型初始化,模型预训练,指令微调。llama,creek

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published