Spaces:
Paused
Paused
Commit History
-
Update acestep/handler.py
Verified signature7a4c91f -
Update acestep/handler.py
Verified signaturef0d1e8c
-
fix: download unified repo for shared components when using independent model repos
XL models are in separate repos without VAE/text encoder. Ensure the unified repo (ACE-Step/Ace-Step1.5) is downloaded first to provide shared components (vae, Qwen3-Embedding-0.6B, silence_latent).7403460 -
feat: add XL (4B DiT) turbo as default model + acemusic banner
- Default model 1: acestep-v15-xl-turbo (was acestep-v15-turbo) - Default model 2: acestep-v15-turbo (was acestep-v15-turbo-shift3) - Add XL model entries to MODEL_REPO_MAPPING for auto-download - Enhanced acemusic.ai banner with gradient background - Update README frontmatter with XL model9d632db
-
fix is_flash_attention_available
b7a9cc9 -
switch to zero gpu
c60d64b
-
fix max audio code id
033008e -
reverse fix
048f54f -
fix max audio code id
f2036a7
-
enforce upload audio duration limits
5f3faee
-
load audio fallback
288b94f -
fix torchcodec ffmpeg
96b5f27 -
fix cover load src audio using ffmpeg
ac82e04 -
support shift3 download
fb234f6
-
refact ui
0b990cd
-
fix: auto download model
b04b635 -
fix: update model repo_id
ef09b17 -
feat: huggingface_space support
59ce525
-
support lora trianing & inter
3df46a2 -
fix language issue simple mode
d54a0fc
-
fix: 修复 offload 模式下的设备不匹配错误
主要修复: - 添加 _recursive_to_device 方法,更彻底地将模型参数移动到目标设备 - 添加 _ensure_silence_latent_on_device 辅助方法确保 silence_latent 在正确设备上 - 修改初始化逻辑,让 silence_latent 始终保持在 GPU 上 - 移除 model offload 时移动 silence_latent 的代码,避免后续使用时设备不匹配 - 在 _prepare_batch、infer_refer_latent、service_generate 中添加设备检查 - tiled_decode 的 offload_wav_to_cpu 默认值改为 True 问题原因: 当 offload_to_cpu 启用时,model context 退出后 silence_latent 被移到 CPU, 但后续代码中其他张量仍在 CUDA 上,导致 torch.cat 等操作报设备不匹配错误6b03a4e -
reduce vRAM usage for vae decode
b857422
-
support input timesteps
bc7e55b
-
add infer_method
f4d9d31
-
cover/repaint test ok
bdc442a
-
feat :sparkles: : add lyrics alignment scores
5ab4485
-
add lyc support
6922ca4
-
add shift support
bb87271
-
fix bugs and test profile
03f73c6
-
add inference code and doc
11860f1
-
add inference code and doc
24f370e -
refact ui and add i18n
1da0418
-
cot caption & language LM
1241c80
-
fix: bot return metas
b3f1425
-
fix llm gen for pt
4670365 -
fix audio stength control
fcbd6fb -
Merge branch 'main' of github.com:ace-step/ACE-Step-1.5 into main
e7130fe -
refact handler
376c43e -
move auto device select to handler& optimized swa impl for client side infer
fc3ecd6 -
refact handler
5ac3586 -
refact handler
bf5e1fd -
Merge branch 'main' of https://github.com/ace-step/ACE-Step-1.5
e6b6b6a -
refine device auto selection in test
4a2d51f
-
Merge branch 'main' of github.com:ace-step/ACE-Step-1.5 into main
8ff7c0c -
fix vllm
a3b47b7 -
refactor quantization method selection
70c780d -
dynamic quantization for dit model & data prepare for further static quantized vae
1daf6b4 -
fix typo
cd892b3 -
add 5hz llm test support & fix 5hz llm transformers inference
1bdda7b -
all handler log printing using loguru
447806b