Robin

icon indicating a website link https://jianzhnie.github.io/llmtech/ icon indicating an email [email protected]

icon company LLMTech icon location Shen zhen icon location Machine learning， Reinforcement Learning， Transformers

Results 36 comments of Robin

Spark Mlib support in NNI

Here some other automl-toolkit have supported spark such as Hyperopt and Optuna, maybe we can refer and discuss how to support MLLib in NNI?

[Question] 实现了百川模型的低资源量化训练和部署

llama-2-13b的模型用单卡跑lora就会报错

试一下用微软的 deepspeed 呢

zero3保存的模型无法加载

是使用的 lora 训练的吗

32g内存+3060ti6G显存可以finetune 7B的模型吗？

完全可以的

Question regarding creation of Guanaco

Hi, NielsRogge, I am also very confused when using the code. It seems like he did not process the dataset in a conversational manner.

About llama-2-70B fine-tuning

Yes, 3\*8*A100 (40G) is enough for fine-tuning llama-2-70B

训练成功了，但是没有合并的脚本，请问如何合并？

Here is the merge code: https://github.com/jianzhnie/Efficient-Tuning-LLMs/blob/main/chatllms/utils/apply_lora.py

训练成功了，但是没有合并的脚本，请问如何合并？

https://github.com/jianzhnie/Efficient-Tuning-LLMs/blob/main/chatllms/utils/apply_lora.py

训练成功了，但是没有合并的脚本，请问如何合并？

CUDA_VISIBLE_DEVICES=0 python chatllms/utils/apply_lora.py \ --base-model-path ~/checkpoints/baichuan7b/ \ --lora-model-path ./work_dir/vicuna_merge_vicuna-baichuan-7b-1gpu/checkpoint-15000 \ --target-model-path ./work_dir/vicuna_merge_vicuna-baichuan-7b/merged_model