GLM-ASR-Nano-2512 cantonese performance is not good

[ming030890]

System Info / 系統信息

Thanks for the effort! However, I ran some evaluations, and GLM-ASR-Nano-2512v has a higher error rate than SenseVoice and WenetSpeech-Yue in terms of Cantonese performance.

Eval sets:
ming030890/cantonese_asr_eval_mdcc_long
commonvoice
ming030890/youtube_caption_yue

Who can help? / 谁可以帮助到您？

No response

Information / 问题信息

• The official example scripts / 官方的示例脚本
• My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

Eval sets:
ming030890/cantonese_asr_eval_mdcc_long
commonvoice
ming030890/youtube_caption_yue

Expected behavior / 期待表现

Lower error rate given that it is much bigger than sensevoice models.

[ming030890]

attached the full eval result

cv stands for commonvoice

[Starrylyn]

sorry, we'll continue to improve