Only 13b has been released, how can this be 7b?

#1
by underlines - opened

According to lmsys, only 13b model weights have been released, with 7b coming soon. How has this 7b been created? Did you fine-tune llama 7b on the vicuna dataset?

wondering the same, lmsys haven't even released the 7b delta nor the database, so how was this model created? Is this a native fine tune or a Lora?

Native finetune with the same hyperparams on sharegpt dataset filtered from "ethics"

GPTQ 4bit please, anyone T_T

Sign up or log in to comment