Xianghui Peng
Xianghui Peng
### Reminder - [X] I have read the README and searched the existing issues. ### System Info 我希望单机4卡微调7B模型,我的每张单卡VRAM有些紧张,因此我希望多张卡共同存放单个模型,例如用4张卡装下2-3(不是4个)个模型实例。 在文档(https://llamafactory.readthedocs.io/zh-cn/latest/advanced/distributed.html)和examples中,我发现有多个支持模型切分的方案,比如 1. FORCE_TORCHRUN=1 CUDA_VISIBLE_DEVICES=0,1,2,3 NPROC_PER_NODE=4 NNODES=1 RANK=0 llamafactory-cli train examples/train_lora/xxx.yaml (deepspeed ZeRO-3)...
### **Issue Description** I followed the instructions in the repository to reproduce the TruthX evaluation on **Llama-2-7B-Chat** but observed significant discrepancies in **multiple-choice metrics** and **generation outputs** compared to the...