Fuhang
@Far0n Looking forwards to lgb support
> @iFe1er No, we cannot fine-tune the GPTQ models. > > If you don't have enough memory resources, you should consider using QLoRA, which also applies 4-bit quantization for LLM....
same issue. i will no longer use wandb.
1. same question here 2. how to convert a finetuned model to a INT4 version manually? would be very appreciated if anyone can reply @yuhangzang
同问， 1. chat_meta的前置prompt能否支持在finetune中加入？像Xtuner的微调是支持system prompt输入的。 2. 另外，XComposer2VL-Chat的finetune支持多轮对话吗？如果希望使用多轮对话数据，conversations中应该如何输入，历史消息需要在user的值中重复吗？比如下面两种方式： ``` { "id": "0", "image": ['path/to/image_0.jpg'] "conversations": [ { "from": "user", "value": "问题1" }, { "from": "assistant", "value": "回答1" }, { "from": "user", "value":...
> Please explicitly set use_reentrant=False in the torch checkpointing function and that should solve the issue. https://pytorch.org/docs/stable/checkpoint.html @jomayeri That doesn't work in my situation. I am working with Transformers Trainer,...
I am confused, why they guys can run galore with deepspeed but i encounter ``` if finetuning_args.use_galore and training_args.deepspeed is not None: raise ValueError("GaLore is incompatible with DeepSpeed.") ``` error...
根据原论文定的，很容易验证SMU在mu= 1000000.0 和SMU-1在mu=4.352665993287951e-9 基本是一样的
Thank you for the note, looking forward for the official version code release! @koushik313
Thanks for the remind, have already change the params.