Memory requirement increased

[nitinmukesh]

Hello @NZqian

Thank you for releasing the new model for us to try.

1. I have taken code from HF space like last time and using it with Gradio UI, but now chunked=True have no effect whatsoever. Even with 95s memory requirement is 8 GB. It used to be <6 GB before.
   Note that DiffRhythm-base requires a minimum of 8G of VRAM #19

Any suggestions please.
I am using the code from here and removed spaces dependency
https://huggingface.co/spaces/ASLP-lab/DiffRhythm

2. The audio generated is always 95s even if I select 285s

[ASLP-lab]

1. We deployed both 95s and 285s models on huggingface, so the memory will be larger than 6G.
   https://huggingface.co/spaces/ASLP-lab/DiffRhythm/blob/0d355fb2211e3f4f04112d8ed30cc9211a79c974/app.py#L30

2. The problem with both 95s and 285s models only generating music for 95s, I suspect it's loading the wrong model, you can check this line of code below.
   https://huggingface.co/spaces/ASLP-lab/DiffRhythm/blob/0d355fb2211e3f4f04112d8ed30cc9211a79c974/diffrhythm/infer/infer_utils.py#L16

[sujitvasanth]

you can use my fix #53 (comment) if you have more than 6-7Gb VRAM