Add key/mode controls to eval generation and A/B

Enable LM planner CoT generation in LoRA UI eval

Add lyric-conditioning findings and dataset guidance

Persist adapter path on disk for ZeroGPU callback recovery

Persist adapter path and auto-reload LoRA for ZeroGPU eval

Auto-reload base model before adapter/eval on ZeroGPU

Fix adapter upload path fallback on HF Spaces

Fix HF Space README metadata emoji encoding

Merge HF Space history for deployment sync

Add LoRA upload flow in Space UI and update README

Consolidate AF3/Qwen pipelines, endpoint templates, and setup docs

github push2

github push

Add UI button to package and download trained adapter ZIP

Fix training prepare_condition is_covers type for ACE-Step builds

Allow uploading JSON sidecars with audio and auto-match by stem

Auto-label: resumable batches for ZeroGPU stability

ZeroGPU: persist LM init args and reload LM in auto-label callback

ZeroGPU: reload model inside auto-label callback when needed

Make model status reflect successful init on Spaces Zero callbacks

Add caption-only auto-label mode (ignore lyrics)

Auto-label skip only when core fields are fully populated

Wrap remaining GPU callbacks for Spaces Zero compatibility

Fix Spaces Zero LM init by wrapping auto-label LM init in spaces.GPU

Fix Spaces Zero CUDA init by wrapping model init in spaces.GPU

Fix HF Spaces Zero startup by importing spaces before torch

Fix auto-label audio fallback NameError by importing numpy

Auto-label: skip only when caption and lyrics already exist

Show per-track encode failures in training result

Fix list-to-tensor audio coercion in encode paths

Add soundfile fallback for audio decode in training and auto-label paths

Fix LoRA target module mapping for ACE-Step v1.5

Use valid emoji in Space metadata

Fix Space metadata for HF validation

Add ACE-Step 1.5 LoRA Space app

Add application file

initial commit

Patch Tensor.sort/argsort bool CUDA compatibility

Patch torch.sort bool CUDA compatibility for endpoint runtime

Add missing ACE-Step runtime deps for dynamic model import

Add transformers compatibility shim for ACE-Step config import

test 2?

Pin Ace step 1.5 dependencies

major changes

model inferencing

init custom endpoint scaffold