Wenhao Chen

icon indicating an email [email protected]

icon company University of Waterloo icon location Waterloo, ON, CA icon location Ph.D. at UWaterloo. B.S. of Shanghai Jiao Tong University. Previous FTE at @hpcaitech. Previous Intern at @Microsoft Research Asia, SRG.

Results 11 comments of Wenhao Chen

[BUG]: bug in training rm with ddp strategy with single machine multi-GPUs!

Thanks for reporting. #4023 Contains this now.

[BUG]: Chat train_sft.py SupervisedDataset: TypeError: __init__() got an unexpected keyword argument 'max_length'

We have updated our code. Try out the latest version. Please reopen this issue if the problem persists.

[BUG]: shardformer: pipeline forward error with customized layer distribution

Thanks for reporting. Would you like to submit a PR to solve this issue :)

[BUG]: shardformer: pipeline forward error with customized layer distribution

Sorry for the delayed update, since I was assigned to another task for the last several months, and this issue is finally resolved.

[BUG]: ColossalChat performance reproduction FAILED

Thanks for reporting this issue. The following description explains how this fails and what are the meanings of these parameters. (problem 1 & 3) ```python for episode in range(num_episodes): for...

[BUG]: ColossalChat performance reproduction FAILED

> If that is true, is it means that I should always keep `experience_batch_size * num_collect_steps == train_batch_size * nproc_per_node * num_update_steps ` (with `sample_buffer == True`) when I training?...

Hope that users can be given the option to more precisely control crawling behavior

Thanks for opening this issue and sorry for the delayed reply. You are welcome to drop a PR if you would like to help enhance it. Besides, it would be...

Hope that users can be given the option to more precisely control crawling behavior

As for these two issues, existing APIs can partially resolve them. 1. By enabling `download_config.with_tag`, tags of artworks, including avatar description, will be collected. However, it seems the title and...

Hope that users can be given the option to more precisely control crawling behavior

> need `with_metadata` for dataAnalysis What kind of metadata do you need? 👀

[DOC] Update README and add explanation of proxy

Thanks for the PR. I am on vacation now and will be available in next week for review.