stainless-steel-rat comments

Repositories
Issues
Comments

Results 4 comments of


                                            stainless-steel-rat

Your tensorflow code may have bugs, it is very hard to train.

@luohongcheng hello, I also tried to train the network, but got a blurry depth map too, Do you find some reasons causing it? thank you.

lstm_predictor.py ImportError: No module named mxnet_predict

遇到了同样的问题，请问找到解决方法了吗？

多节点使用zero3速度很慢

> [@hiyouga](https://github.com/hiyouga) 我尝试过使用zero++，设置了zero_hpz_partition_size为8,确实加速了训练，但是训练过程中loss一直固定为11，grad_norm一直为0，训练失败，请问是llamafactory还没适配zero++吗，我该怎样才能正确使用zero++？我在进行dpo训练的时候也遇到了这个问题。

Some instances from the SWE-Bench Verified are failing to be built. (matplotlibmatplotlib-25775, matplotlibmatplotlib-26208, and others)

I meet the same problem. Is there any solution for this?

stainless-steel-rat

Your tensorflow code may have bugs, it is very hard to train.

lstm_predictor.py ImportError: No module named mxnet_predict

多节点使用zero3速度很慢

Some instances from the SWE-Bench Verified are failing to be built. (matplotlib__matplotlib-25775, matplotlib__matplotlib-26208, and others)

Some instances from the SWE-Bench Verified are failing to be built. (matplotlibmatplotlib-25775, matplotlibmatplotlib-26208, and others)