fix: download unified repo for shared components when using independent model repos

feat: add XL (4B DiT) turbo as default model + acemusic banner

fix: lower ZeroGPU duration to 120s for free-tier users

fix: move @spaces.GPU decorators to local wrappers to fix ZeroGPU pickling error

fix is_flash_attention_available

switch to zero gpu

resolve intermittent CUDA assertion error in concurrent serving scenarios

fix max audio code id

reverse fix

fix max audio code id

update tech report link

add acemusic link

fix cfg kv block allocate

enforce upload audio duration limits

fix duration

set default_concurrency_limit & set bs=2 no interactive

bigger sample btn

load audio fallback

fix torchcodec ffmpeg

fix CUDA Graph mismatch in nanovllm

fix cover load src audio using ffmpeg

hide files list

support shift3 download

update project page url

refact ui (fix repaint/cover score bug)

refact ui

fix audio play position reset

fix Qwen3ForCausalLM has no attribute

fix Qwen3ForCausalLM has no attribute

fix: auto download model

fix: update model repo_id

Merge branch 'main' into huggingface_space

feat: huggingface_space support

add model zoo

feat: update api server data

feat: update api

add docs and serve mode

fix cover logic for api_server

support lora trianing & inter

fix language issue simple mode

fix: 修复 offload 模式下的设备不匹配错误

reduce vRAM usage for vae decode

max_model_len 8192 -> 4096

feat: change windows dependencies

fix api server no duration

feat: add v1/model

Merge branch 'main' of https://github.com/ace-step/ACE-Step-1.5

fix: cot_caption

Merge branch 'dev' into main

feat: shift