Drop gr.Video/make_waveform (removed in gradio 5); gr.Audio shows waveform natively

Build MODELS dict directly at module scope (no preload function wrapper)

Preload all 10 MusicGen variants on CUDA at module load (~36GB fits on 48GB ZeroGPU)

Render waveforms synchronously (daemonic process can't spawn ProcessPoolExecutor)

Add demucs runtime deps (openunmix, lameenc, dora-search, pyyaml)

Pin torch==2.8.0 + torchaudio==2.8.0 (ZeroGPU supported version)

ZeroGPU + gradio 5 migration: bump python/gradio, drop torch pins, demucs --no-deps, merge mic+upload, @spaces.GPU on predict_full

[Admin maintenance] Support new ZeroGPU hardware

disable external embeddings (#64)

Stereo demo update (#60)

Fix invalid colorFrom causing build failure when duplicating space (#59)

Update requirements.txt (#55)

Pin gradio_client to prevent incompatibility issues (#47)

Merge branch 'main' into our_hf2

linting and adding file cleaner

Merge remote-tracking branch 'hf2/main' into our_hf2

adding back header

Update README.md

Model can be loaded from local directory (#69)

add option to record from mic (#87)

Update app.py

again

trying to fix test on CI

readme

adding checks

Use greedy sampling path when temp is 0.0 to avoid division by zero (#53)

update change log

merge

rm old app

no progress on batch

bad merge

Merge branch 'longgen' into our_hf2

update space to clone

final changes

adding support for cpu

plop

kind of working

nit

nit

nit

nit

initial implem

support both torch and xformers + merge apps

fix

trying fix

trying to debug

Merge branch 'longgen' into our_hf2

new version

plop

real fix