Benwei Lu

> hello, did you find the right source? upd: wrote script. you can find it in my fork Thanks a lot ,your work is amazing

open `utils.py` and comment line 75,76,98,99, which corresponds to ```python3 if not resume: copy_cur_env(work_dir, exp_path+ '/' + exp_name + '/code', exception) ```

Same issues here

I think I might know why , I upload a paper with sevealr formulas and charts,I got an erroe,but when I upload a file just words ,then it would work...

> You can try reducing max_ token or use gpt-3.5 turbo > > ``` > completions = openai.ChatCompletion.create( > model="gpt-3.5-turbo", > temperature=0, > messages=[ > {"role": "user", "content": prompt} >...

same feeling~

same issue here, but how?@

> I am using Lora to finetune Qwen-VL model, there are about 10,000 VQA data samples I used for finetuning, but the loss of final model is still high, I...

> 有个问题想请教各位大佬，根据finetune.py进行lora微调，这算是指令微调还是有监督精调啊，最近一直在纠结这个问题。 属于SFT(Supervised Fine-Tuning）监督微调

lora和Qlora确实都是SFT,至于指令微调目前还没有查到信息,如果有消息记得共享哦谢谢