Zi-Yuan Hu

icon indicating a website link https://henryhzy.github.io/

icon company LaVi Lab led by Prof. Liwei Wang @ CSE, CUHK icon location Hong Kong icon location Interested in multimodal learning (vision-and-language) and parameter-efficient learning.

Results 13 comments of Zi-Yuan Hu

[Q]Sync offline metrics on another machine

> Delete the files in the wandb/file works with me.

How to use Qwen3-VL generate() with num_return_sequences > 1?

> refer to [#1621 (comment)](https://github.com/QwenLM/Qwen3-VL/issues/1621#issuecomment-3420454200) Thanks for your quick reply. It solves my problem!

[BUG] <title>多图单轮对话全量SFT，loss 变为零

Changing full fine-tuning to lora fine-tuning works for me.