ssbuild

Results 82 comments of ssbuild

lora 需要加载半精度权重, 你的权重下载错了,从官网下载权重试试。

推理可以, 训练的话半精度从 https://huggingface.co/THUDM/chatglm-6b 下载, 重新配置制作数据 训练应该就可以。

> > 推理可以, 训练的话半精度从 https://huggingface.co/THUDM/chatglm-6b 下载, 重新配置制作数据 训练应该就可以。 > > 大佬是bin模型也得用6b的吗,那我就没法跑了,容量危机了 ![(U}6}N~MZ5@LPK) I42TO D](https://user-images.githubusercontent.com/40910637/236613845-d66d1c44-e0cc-4e2b-8c89-056985433dac.jpg) int4 可以玩 ptv2 , 修改config/config_ptv2.json quantization_bit=4 , 并在train_info_args 改成 config/config_ptv2.json

> > _ptv2 > > 难绷 ![O~ECZ`E99{K@NCBA) UBL2U](https://user-images.githubusercontent.com/40910637/236614455-50ccd142-300b-477f-8271-bf959924a39b.gif) [ ![O~ECZ`E99{K@NCBA) UBL2U](https://user-images.githubusercontent.com/40910637/236614455-50ccd142-300b-477f-8271-bf959924a39b.gif) ](https://user-images.githubusercontent.com/40910637/236614455-50ccd142-300b-477f-8271-bf959924a39b.gif) [ ](https://user-images.githubusercontent.com/40910637/236614455-50ccd142-300b-477f-8271-bf959924a39b.gif) > > ``` > ╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮ > │ /root/autodl-tmp/chatglm_finetuning/train.py:133 in │...

@Ikaros-521 infer_lora_finetuning.py 是 lora ptv2 推理 使用infer_finetuning.py

感谢反馈,已更正。

try: pip install -i https://pypi.org/simple deep_training

tfrecords support only linux or windows , if macos , may instal failed.

@Amber-Men update you training code, install by requirement