Weikang Qiu comments

Results 34 comments of


                                            Weikang Qiu

Discounted Reward Calulcation (Generalized Advantage Estimation)

This inaccuracy (maybe I should call it a bug) troubles me a lot! Thanks @nikhilbarhate99 Also, I guess there is another issue. if the loop exit with `t >= max_ep_len`,...

Is DeepMosaics abandonware at this point?

Anyone who wants to Make It Great Again?

Is it possible to reuse a step's output for other two steps (avoid reruning the shared step)

I just wonder if it is possible.... Just reply me with a "no" if it is not possible...

[Feature request] Macos 分屏打开新浏览器窗口

全屏有办法吗

tango info got Deprecated warning

any updates?

Error when using pytorch dataloader with h5py when `num_workers` > 0

If I run the code without tango, i.e. use python command directly, I won't get any error

Question on padding_side "left" vs "right"

Would `position_ids` be a problem if the LLM uses RoPE, where the results only depend on the relative position?

run_finetune takes too much time.

I installed from source, but it still need 50+h

run_finetune takes too much time.

I found in the paper the number of candidate documents is set to 8. > For each example, we retrieve and marginalize over 8 candidate documents, including the null document...

run_finetune takes too much time.

Thanks. Does 5k runs slowly too on your computer (as far as I know the original implementation is TensorFlow)? If it does. Do you mind me optimizing the code and...