@csuhan Can you please clarify the following? In the paper, according to the Table 8 the data ratios are 2(und):2(gen):1(text) for MLLM pre-training and SFT, but in the code (link) it's 2(und):4(gen):1(text). Which one is correct?
Context: I am trying to continue finetuning the 7B model on some task and the above information is critical. Can you please clarify
@csuhan Can you please clarify the following? In the paper, according to the Table 8 the data ratios are 2(und):2(gen):1(text) for MLLM pre-training and SFT, but in the code (link) it's 2(und):4(gen):1(text). Which one is correct?
Context: I am trying to continue finetuning the 7B model on some task and the above information is critical. Can you please clarify