Sangwoong Yoon comments

Results 6 comments of


                                            Sangwoong Yoon

Is StyleGAN2 code provided in this repo fully tested?

Thanks for the super cool repo. The training script for StyleGAN2 should be awesome. Also waiting for it!

[CLI]: wandb.finish() stuck when uploaded all data

Same issue here.

Which version of vllm should be installed

I am afraid the version of transformers does matter. When I run `run_sppo_gemma-2.sh`, I get the following error. ```ValueError: Tokenizer class GemmaTokenizer does not exist or is not currently imported.```...

Which version of vllm should be installed

@angelahzyuan thank you so much for a very prompt answer. Updating `transformers==4.42.3`, `tokensizers==0.19.1`, and `trl==0.9.4` solved the issue on my side (running Llama3), but not thoroughly tested yet for other...

Gradient accumulation gives worse results when using DeepSpeed ZeRO 2

@kashif Thank you very much for looking into this issue. At least in the above example, it was only zero 2. However, maybe different trainers gives different behaviors. Should I...

[Bug] DeepSpeed ZeRO-2 is unstable with GRPOTrainer using Qwen3-VL MoE

This may be related to this issue that I posted in accelerate: https://github.com/huggingface/accelerate/issues/3877