pedroapfilho's picture
Add custom tag input and fix model download (zip directory)
6ca081e unverified
"""
ACE-Step 1.5 Custom Edition - Main Application
A comprehensive music generation system with three main interfaces:
1. Standard ACE-Step GUI
2. Custom Timeline-based Workflow
3. LoRA Training Studio
"""
import gradio as gr
import pandas as pd
import torch
import numpy as np
from pathlib import Path
import json
from typing import Optional, List, Tuple
try:
import spaces
except ImportError:
# Local dev — make @spaces.GPU a no-op
class _Spaces:
def GPU(self, fn=None, **kwargs):
return fn if fn else lambda f: f
spaces = _Spaces()
from src.ace_step_engine import ACEStepEngine
from src.timeline_manager import TimelineManager
from src.lora_trainer import download_hf_dataset, upload_dataset_json_to_hf
from src.audio_processor import AudioProcessor
from src.utils import setup_logging, load_config
from acestep.training.dataset_builder import DatasetBuilder
from acestep.training.configs import LoRAConfig, TrainingConfig
from acestep.training.trainer import LoRATrainer as FabricLoRATrainer
# Setup
logger = setup_logging()
config = load_config()
# Lazy initialize components (will be initialized on first use)
ace_engine = None
timeline_manager = None
dataset_builder = None
audio_processor = None
# Module-level mutable dict for training stop signal
# (gr.State is not shared between concurrent Gradio calls)
_training_control = {"should_stop": False}
def get_ace_engine():
"""Lazy-load ACE-Step engine."""
global ace_engine
if ace_engine is None:
ace_engine = ACEStepEngine(config)
return ace_engine
def get_timeline_manager():
"""Lazy-load timeline manager."""
global timeline_manager
if timeline_manager is None:
timeline_manager = TimelineManager(config)
return timeline_manager
def get_dataset_builder():
"""Lazy-load dataset builder."""
global dataset_builder
if dataset_builder is None:
dataset_builder = DatasetBuilder()
return dataset_builder
def get_audio_processor():
"""Lazy-load audio processor."""
global audio_processor
if audio_processor is None:
audio_processor = AudioProcessor(config)
return audio_processor
# ==================== TAB 1: STANDARD ACE-STEP GUI ====================
@spaces.GPU
def standard_generate(
prompt: str,
lyrics: str,
duration: int,
temperature: float,
top_p: float,
seed: int,
style: str,
use_lora: bool,
lora_path: Optional[str] = None
) -> Tuple[str, str]:
"""Standard ACE-Step generation with all original features."""
try:
logger.info(f"Standard generation: {prompt[:50]}...")
# Get engine instance
engine = get_ace_engine()
# Generate audio
audio_path = engine.generate(
prompt=prompt,
lyrics=lyrics,
duration=duration,
temperature=temperature,
top_p=top_p,
seed=seed,
style=style,
lora_path=lora_path if use_lora else None
)
info = f"✅ Generated {duration}s audio successfully"
return audio_path, info
except Exception as e:
logger.error(f"Standard generation failed: {e}")
return None, f"❌ Error: {str(e)}"
@spaces.GPU
def standard_variation(audio_path: str, variation_strength: float) -> Tuple[str, str]:
"""Generate variation of existing audio."""
try:
result = get_ace_engine().generate_variation(audio_path, variation_strength)
return result, "✅ Variation generated"
except Exception as e:
return None, f"❌ Error: {str(e)}"
@spaces.GPU
def standard_repaint(
audio_path: str,
start_time: float,
end_time: float,
new_prompt: str
) -> Tuple[str, str]:
"""Repaint specific section of audio."""
try:
result = get_ace_engine().repaint(audio_path, start_time, end_time, new_prompt)
return result, f"✅ Repainted {start_time}s-{end_time}s"
except Exception as e:
return None, f"❌ Error: {str(e)}"
@spaces.GPU
def standard_lyric_edit(
audio_path: str,
new_lyrics: str
) -> Tuple[str, str]:
"""Edit lyrics while maintaining music."""
try:
result = get_ace_engine().edit_lyrics(audio_path, new_lyrics)
return result, "✅ Lyrics edited"
except Exception as e:
return None, f"❌ Error: {str(e)}"
# ==================== TAB 2: CUSTOM TIMELINE WORKFLOW ====================
@spaces.GPU
def timeline_generate(
prompt: str,
lyrics: str,
context_length: int,
style: str,
temperature: float,
seed: int,
session_state: dict
) -> Tuple[str, str, str, dict]:
"""
Generate 32-second clip with 2s lead-in, 28s main, 2s lead-out.
Blends with previous clips based on context_length.
"""
try:
# Initialize session state if None
if session_state is None:
session_state = {"timeline_id": None, "total_clips": 0}
logger.info(f"Timeline generation with {context_length}s context")
# Get managers
tm = get_timeline_manager()
engine = get_ace_engine()
ap = get_audio_processor()
# Get context from timeline
context_audio = tm.get_context(
session_state.get("timeline_id"),
context_length
)
# Generate 32s clip
clip = engine.generate_clip(
prompt=prompt,
lyrics=lyrics,
duration=32,
context_audio=context_audio,
style=style,
temperature=temperature,
seed=seed
)
# Blend with timeline (2s lead-in and lead-out)
blended_clip = ap.blend_clip(
clip,
tm.get_last_clip(session_state.get("timeline_id")),
lead_in=2.0,
lead_out=2.0
)
# Add to timeline
timeline_id = tm.add_clip(
session_state.get("timeline_id"),
blended_clip,
metadata={
"prompt": prompt,
"lyrics": lyrics,
"context_length": context_length
}
)
# Update session
session_state["timeline_id"] = timeline_id
session_state["total_clips"] = session_state.get("total_clips", 0) + 1
# Get full timeline audio
full_audio = tm.export_timeline(timeline_id)
# Get timeline visualization
timeline_viz = tm.visualize_timeline(timeline_id)
info = f"✅ Clip {session_state['total_clips']} added • Total: {tm.get_duration(timeline_id):.1f}s"
return blended_clip, full_audio, timeline_viz, session_state, info
except Exception as e:
logger.error(f"Timeline generation failed: {e}")
return None, None, None, session_state, f"❌ Error: {str(e)}"
def timeline_extend(
prompt: str,
lyrics: str,
context_length: int,
session_state: dict
) -> Tuple[str, str, str, dict]:
"""Extend current timeline with new generation."""
return timeline_generate(
prompt, lyrics, context_length, "auto", 0.7, -1, session_state
)
@spaces.GPU
def timeline_inpaint(
start_time: float,
end_time: float,
new_prompt: str,
session_state: dict
) -> Tuple[str, str, dict]:
"""Inpaint specific region in timeline."""
try:
# Initialize session state if None
if session_state is None:
session_state = {"timeline_id": None, "total_clips": 0}
tm = get_timeline_manager()
timeline_id = session_state.get("timeline_id")
result = tm.inpaint_region(
timeline_id,
start_time,
end_time,
new_prompt
)
full_audio = tm.export_timeline(timeline_id)
timeline_viz = tm.visualize_timeline(timeline_id)
info = f"✅ Inpainted {start_time:.1f}s-{end_time:.1f}s"
return full_audio, timeline_viz, session_state, info
except Exception as e:
return None, None, session_state, f"❌ Error: {str(e)}"
def timeline_reset(session_state: dict) -> Tuple[None, None, str, dict]:
"""Reset timeline to start fresh."""
# Initialize session state if None
if session_state is None:
session_state = {"timeline_id": None, "total_clips": 0}
elif session_state.get("timeline_id"):
get_timeline_manager().delete_timeline(session_state["timeline_id"])
session_state = {"timeline_id": None, "total_clips": 0}
return None, None, "Timeline cleared", session_state
# ==================== TAB 3: LORA TRAINING STUDIO ====================
DATAFRAME_HEADERS = ["#", "Filename", "Duration", "Lyrics", "Labeled", "BPM", "Key", "Caption"]
def _build_progress_summary():
"""Build a one-line progress summary from current dataset builder state."""
builder = get_dataset_builder()
total = builder.get_sample_count()
labeled = builder.get_labeled_count()
preprocessed = builder.get_preprocessed_count()
remaining = total - labeled
return f"Total: {total} | Labeled: {labeled} | Preprocessed: {preprocessed} | Remaining: {remaining}"
def _build_review_dataframe():
"""Build editable dataframe rows from current dataset builder state."""
builder = get_dataset_builder()
return builder.get_samples_dataframe_data()
def lora_download_hf(dataset_id, custom_tag, max_files, hf_offset, training_state):
"""Download HuggingFace dataset batch, restore labels from HF repo, and scan."""
try:
if not dataset_id or not dataset_id.strip():
return "Enter a dataset ID (e.g. username/dataset-name)", training_state, int(hf_offset or 0), _build_progress_summary()
offset_val = int(hf_offset or 0)
max_files_val = int(max_files)
local_dir, dl_status = download_hf_dataset(
dataset_id.strip(),
max_files=max_files_val,
offset=offset_val,
)
if not local_dir:
return f"Download failed: {dl_status}", training_state, offset_val, _build_progress_summary()
builder = get_dataset_builder()
# Set trigger word for LoRA training
tag = custom_tag.strip() if custom_tag else ""
if tag:
builder.set_custom_tag(tag)
# Restore labels/flags from dataset.json pulled from HF repo
dataset_json_path = str(Path(local_dir) / "dataset.json")
if Path(dataset_json_path).exists():
builder.load_dataset(dataset_json_path)
dl_status += " | Restored labels from HF repo"
# Scan directory — skips already-tracked files via existing_paths check
samples, scan_status = builder.scan_directory(local_dir)
training_state = training_state or {}
training_state["audio_dir"] = local_dir
training_state["dataset_id"] = dataset_id.strip()
training_state["dataset_path"] = dataset_json_path
next_offset = offset_val + max_files_val
return f"{dl_status} | {scan_status}", training_state, next_offset, _build_progress_summary()
except Exception as e:
logger.error(f"HF download failed: {e}")
return f"Error: {e}", training_state or {}, int(hf_offset or 0), _build_progress_summary()
def lora_save_dataset_to_json(training_state):
"""Explicitly save the current dataset to JSON."""
try:
builder = get_dataset_builder()
if builder.get_sample_count() == 0:
return "No samples to save"
training_state = training_state or {}
dataset_path = training_state.get("dataset_path")
if not dataset_path:
audio_dir = training_state.get("audio_dir", "lora_training")
dataset_path = str(Path(audio_dir) / "dataset.json")
training_state["dataset_path"] = dataset_path
return builder.save_dataset(dataset_path)
except Exception as e:
logger.error(f"Save dataset failed: {e}")
return f"Error: {e}"
@spaces.GPU(duration=300)
def lora_auto_label(label_batch_size, training_state, progress=gr.Progress()):
"""Auto-label unlabeled samples in batches using LLM analysis, then auto-save."""
try:
builder = get_dataset_builder()
if builder.get_sample_count() == 0:
return [], "No samples loaded. Upload files or download a dataset first.", training_state, _build_progress_summary()
engine = get_ace_engine()
if not engine.is_initialized():
return [], "ACE-Step engine not initialized. Models may still be loading.", training_state, _build_progress_summary()
def progress_callback(msg):
progress(0, desc=msg)
samples, status = builder.label_all_samples(
dit_handler=engine.dit_handler,
llm_handler=engine.llm_handler,
only_unlabeled=True,
max_count=int(label_batch_size),
progress_callback=progress_callback,
)
training_state = training_state or {}
dataset_path = training_state.get("dataset_path")
if not dataset_path:
audio_dir = training_state.get("audio_dir", "lora_training")
dataset_path = str(Path(audio_dir) / "dataset.json")
training_state["dataset_path"] = dataset_path
save_status = builder.save_dataset(dataset_path)
status += f"\n{save_status}"
# Sync to HF repo so labels persist across sessions
dataset_id = training_state.get("dataset_id")
if dataset_id:
hf_status = upload_dataset_json_to_hf(dataset_id, dataset_path)
status += f"\n{hf_status}"
return _build_review_dataframe(), status, training_state, _build_progress_summary()
except Exception as e:
logger.error(f"Auto-label failed: {e}")
return [], f"Error: {e}", training_state or {}, _build_progress_summary()
def lora_save_edits(df_data, training_state):
"""Save user edits from the review dataframe back to samples."""
try:
builder = get_dataset_builder()
if df_data is None:
return "No data to save"
if isinstance(df_data, pd.DataFrame):
if df_data.empty:
return "No data to save"
rows = df_data.values.tolist()
elif isinstance(df_data, list):
if len(df_data) == 0:
return "No data to save"
rows = df_data
else:
return "No data to save"
updated = 0
for row in rows:
idx = int(row[0])
updates = {}
# Map editable columns back to sample fields
bpm_val = row[5]
if bpm_val and bpm_val != "-":
try:
updates["bpm"] = int(bpm_val)
except (ValueError, TypeError):
pass
key_val = row[6]
if key_val and key_val != "-":
updates["keyscale"] = str(key_val)
caption_val = row[7]
if caption_val and caption_val != "-":
updates["caption"] = str(caption_val)
if updates:
builder.update_sample(idx, **updates)
updated += 1
return f"Updated {updated} samples"
except Exception as e:
logger.error(f"Save edits failed: {e}")
return f"Error: {e}"
@spaces.GPU(duration=300)
def lora_preprocess(preprocess_batch_size, training_state, progress=gr.Progress()):
"""Preprocess labeled samples to training tensors in batches."""
try:
builder = get_dataset_builder()
if builder.get_labeled_count() == 0:
return "No labeled samples. Run auto-label first.", _build_progress_summary()
engine = get_ace_engine()
if not engine.is_initialized():
return "ACE-Step engine not initialized.", _build_progress_summary()
tensor_dir = str(Path("lora_training") / "tensors")
def progress_callback(msg):
progress(0, desc=msg)
output_paths, status = builder.preprocess_to_tensors(
dit_handler=engine.dit_handler,
output_dir=tensor_dir,
max_count=int(preprocess_batch_size),
progress_callback=progress_callback,
)
training_state = training_state or {}
training_state["tensor_dir"] = tensor_dir
# Auto-save so preprocessed flags persist across sessions
dataset_path = training_state.get("dataset_path")
if not dataset_path:
audio_dir = training_state.get("audio_dir", "lora_training")
dataset_path = str(Path(audio_dir) / "dataset.json")
training_state["dataset_path"] = dataset_path
save_status = builder.save_dataset(dataset_path)
status += f"\n{save_status}"
# Sync to HF repo so preprocessed flags persist across sessions
dataset_id = training_state.get("dataset_id")
if dataset_id:
hf_status = upload_dataset_json_to_hf(dataset_id, dataset_path)
status += f"\n{hf_status}"
return status, _build_progress_summary()
except Exception as e:
logger.error(f"Preprocess failed: {e}")
return f"Error: {e}", _build_progress_summary()
@spaces.GPU(duration=600)
def lora_train_real(
lr, batch_size, epochs, rank, alpha,
grad_accum, model_name, training_state,
progress=gr.Progress(),
):
"""Train LoRA using the real Fabric-based trainer."""
try:
training_state = training_state or {}
tensor_dir = training_state.get("tensor_dir", "")
if not tensor_dir or not Path(tensor_dir).exists():
return "", "No preprocessed tensors found. Run preprocessing first."
engine = get_ace_engine()
if not engine.is_initialized():
return "", "ACE-Step engine not initialized."
lora_cfg = LoRAConfig(r=int(rank), alpha=int(alpha))
output_dir = str(Path("lora_training") / "models" / (model_name or "lora_model"))
train_cfg = TrainingConfig(
learning_rate=float(lr),
batch_size=int(batch_size),
max_epochs=int(epochs),
gradient_accumulation_steps=int(grad_accum),
output_dir=output_dir,
)
trainer = FabricLoRATrainer(
dit_handler=engine.dit_handler,
lora_config=lora_cfg,
training_config=train_cfg,
)
_training_control["should_stop"] = False
last_msg = ""
for step, loss, message in trainer.train_from_preprocessed(
tensor_dir=tensor_dir,
training_state=_training_control,
):
last_msg = f"Step {step} | Loss: {loss:.4f} | {message}"
progress(0, desc=last_msg)
if _training_control.get("should_stop"):
trainer.stop()
last_msg = f"Training stopped at step {step} (loss: {loss:.4f})"
break
final_path = str(Path(output_dir) / "final")
return final_path, last_msg
except Exception as e:
logger.error(f"Training failed: {e}")
return "", f"Error: {e}"
def lora_stop_training():
"""Signal the training loop to stop."""
_training_control["should_stop"] = True
return "Stop signal sent. Training will stop after current step."
def lora_download_model(model_path):
"""Zip the LoRA model directory and return the zip for Gradio file download."""
import shutil
if not model_path or not Path(model_path).exists():
return None
path = Path(model_path)
if path.is_dir():
zip_path = path.parent / path.name
shutil.make_archive(str(zip_path), "zip", root_dir=str(path.parent), base_dir=path.name)
return str(zip_path) + ".zip"
return model_path
# ==================== GRADIO UI ====================
def create_ui():
"""Create the three-tab Gradio interface."""
with gr.Blocks(title="ACE-Step 1.5 Custom Edition", theme=gr.themes.Soft()) as app:
gr.Markdown("""
# 🎵 ACE-Step 1.5 Custom Edition
**Three powerful interfaces for music generation and training**
Models will download automatically on first use (~7GB from HuggingFace)
""")
with gr.Tabs():
# ============ TAB 1: STANDARD ACE-STEP ============
with gr.Tab("🎼 Standard ACE-Step"):
gr.Markdown("### Full-featured standard ACE-Step 1.5 interface")
with gr.Row():
with gr.Column():
std_prompt = gr.Textbox(
label="Prompt",
placeholder="Describe the music style, mood, instruments...",
lines=3
)
std_lyrics = gr.Textbox(
label="Lyrics (optional)",
placeholder="Enter lyrics here...",
lines=5
)
with gr.Row():
std_duration = gr.Slider(
minimum=10, maximum=240, value=30, step=10,
label="Duration (seconds)"
)
std_style = gr.Dropdown(
choices=["auto", "pop", "rock", "jazz", "classical", "electronic", "hip-hop"],
value="auto",
label="Style"
)
with gr.Row():
std_temperature = gr.Slider(
minimum=0.1, maximum=1.5, value=0.7, step=0.1,
label="Temperature"
)
std_top_p = gr.Slider(
minimum=0.1, maximum=1.0, value=0.9, step=0.05,
label="Top P"
)
std_seed = gr.Number(label="Seed (-1 for random)", value=-1)
with gr.Row():
std_use_lora = gr.Checkbox(label="Use LoRA", value=False)
std_lora_path = gr.Textbox(
label="LoRA Path",
placeholder="Path to LoRA model (if using)"
)
std_generate_btn = gr.Button("🎵 Generate", variant="primary", size="lg")
with gr.Column():
gr.Markdown("### Audio Input (Optional)")
gr.Markdown("*Upload audio file or record to use as style guidance*")
std_audio_input = gr.Audio(
label="Style Reference Audio",
type="filepath"
)
gr.Markdown("### Generated Output")
std_audio_out = gr.Audio(label="Generated Audio")
std_info = gr.Textbox(label="Status", lines=2)
gr.Markdown("### Advanced Controls")
with gr.Accordion("🔄 Generate Variation", open=False):
std_var_strength = gr.Slider(0.1, 1.0, 0.5, label="Variation Strength")
std_var_btn = gr.Button("Generate Variation")
with gr.Accordion("🎨 Repaint Section", open=False):
std_repaint_start = gr.Number(label="Start Time (s)", value=0)
std_repaint_end = gr.Number(label="End Time (s)", value=10)
std_repaint_prompt = gr.Textbox(label="New Prompt", lines=2)
std_repaint_btn = gr.Button("Repaint")
with gr.Accordion("✏️ Edit Lyrics", open=False):
std_edit_lyrics = gr.Textbox(label="New Lyrics", lines=4)
std_edit_btn = gr.Button("Edit Lyrics")
# Event handlers
std_generate_btn.click(
fn=standard_generate,
inputs=[std_prompt, std_lyrics, std_duration, std_temperature,
std_top_p, std_seed, std_style, std_use_lora, std_lora_path],
outputs=[std_audio_out, std_info]
)
std_var_btn.click(
fn=standard_variation,
inputs=[std_audio_out, std_var_strength],
outputs=[std_audio_out, std_info]
)
std_repaint_btn.click(
fn=standard_repaint,
inputs=[std_audio_out, std_repaint_start, std_repaint_end, std_repaint_prompt],
outputs=[std_audio_out, std_info]
)
std_edit_btn.click(
fn=standard_lyric_edit,
inputs=[std_audio_out, std_edit_lyrics],
outputs=[std_audio_out, std_info]
)
# ============ TAB 2: CUSTOM TIMELINE ============
with gr.Tab("⏱️ Timeline Workflow"):
gr.Markdown("""
### Custom Timeline-based Generation
Generate 32-second clips that seamlessly blend together on a master timeline.
""")
# Session state for timeline
timeline_state = gr.State(value=None)
with gr.Row():
with gr.Column():
tl_prompt = gr.Textbox(
label="Prompt",
placeholder="Describe this section...",
lines=3
)
tl_lyrics = gr.Textbox(
label="Lyrics for this clip",
placeholder="Enter lyrics for this 32s section...",
lines=4
)
gr.Markdown("*How far back to reference for style guidance*")
tl_context_length = gr.Slider(
minimum=0, maximum=120, value=30, step=10,
label="Context Length (seconds)"
)
with gr.Row():
tl_style = gr.Dropdown(
choices=["auto", "pop", "rock", "jazz", "electronic"],
value="auto",
label="Style"
)
tl_temperature = gr.Slider(
minimum=0.5, maximum=1.0, value=0.7, step=0.05,
label="Temperature"
)
tl_seed = gr.Number(label="Seed (-1 for random)", value=-1)
with gr.Row():
tl_generate_btn = gr.Button("🎵 Generate Clip", variant="primary", size="lg")
tl_extend_btn = gr.Button("➕ Extend", size="lg")
tl_reset_btn = gr.Button("🔄 Reset Timeline", variant="secondary")
tl_info = gr.Textbox(label="Status", lines=2)
with gr.Column():
tl_clip_audio = gr.Audio(label="Latest Clip")
tl_full_audio = gr.Audio(label="Full Timeline")
tl_timeline_viz = gr.Image(label="Timeline Visualization")
with gr.Accordion("🎨 Inpaint Timeline Region", open=False):
tl_inpaint_start = gr.Number(label="Start Time (s)", value=0)
tl_inpaint_end = gr.Number(label="End Time (s)", value=10)
tl_inpaint_prompt = gr.Textbox(label="New Prompt", lines=2)
tl_inpaint_btn = gr.Button("Inpaint Region")
# Event handlers
tl_generate_btn.click(
fn=timeline_generate,
inputs=[tl_prompt, tl_lyrics, tl_context_length, tl_style,
tl_temperature, tl_seed, timeline_state],
outputs=[tl_clip_audio, tl_full_audio, tl_timeline_viz, timeline_state, tl_info]
)
tl_extend_btn.click(
fn=timeline_extend,
inputs=[tl_prompt, tl_lyrics, tl_context_length, timeline_state],
outputs=[tl_clip_audio, tl_full_audio, tl_timeline_viz, timeline_state, tl_info]
)
tl_reset_btn.click(
fn=timeline_reset,
inputs=[timeline_state],
outputs=[tl_clip_audio, tl_full_audio, tl_info, timeline_state]
)
tl_inpaint_btn.click(
fn=timeline_inpaint,
inputs=[tl_inpaint_start, tl_inpaint_end, tl_inpaint_prompt, timeline_state],
outputs=[tl_full_audio, tl_timeline_viz, timeline_state, tl_info]
)
# ============ TAB 3: LORA TRAINING STUDIO ============
with gr.Tab("🎓 LoRA Training Studio"):
gr.Markdown("""
### Train Custom LoRA Models
Step-by-step wizard: provide audio data, auto-label with LLM, preprocess, and train.
""")
training_state = gr.State(value={})
lora_progress = gr.Textbox(
label="Progress",
value="Total: 0 | Labeled: 0 | Preprocessed: 0 | Remaining: 0",
interactive=False,
)
with gr.Tabs():
# ---------- Sub-tab 1: Data Source ----------
with gr.Tab("1. Data Source"):
gr.Markdown(
"Download audio from a HuggingFace dataset repo. "
"Labels and progress are synced back to the repo automatically."
)
lora_hf_id = gr.Textbox(
label="Dataset ID",
placeholder="username/dataset-name",
)
lora_custom_tag = gr.Textbox(
label="Custom Tag (trigger word for LoRA)",
placeholder="lofi, synthwave, jazz-piano…",
)
with gr.Row():
lora_hf_max = gr.Slider(
minimum=1, maximum=500, value=50, step=1,
label="Batch size",
)
lora_hf_offset = gr.Number(
label="Offset (auto-increments)",
value=0,
precision=0,
)
lora_hf_btn = gr.Button(
"Download Batch & Scan", variant="primary"
)
lora_source_status = gr.Textbox(
label="Status", lines=2, interactive=False
)
# ---------- Sub-tab 2: Label & Review ----------
with gr.Tab("2. Label & Review"):
gr.Markdown(
"Auto-label samples using the LLM, then review and edit metadata."
)
lora_label_batch_size = gr.Slider(
minimum=1, maximum=500, value=50, step=1,
label="Label batch size (samples per run)",
)
lora_label_btn = gr.Button(
"Label Batch (+ auto-save)",
variant="primary",
)
lora_label_status = gr.Textbox(
label="Label Status", lines=3, interactive=False
)
lora_review_df = gr.Dataframe(
headers=DATAFRAME_HEADERS,
label="Sample Review (editable: BPM, Key, Caption)",
interactive=True,
wrap=True,
)
with gr.Row():
lora_save_btn = gr.Button("Save Edits")
lora_save_dataset_btn = gr.Button(
"Save Dataset to JSON", variant="secondary"
)
lora_save_status = gr.Textbox(
label="Save Status", interactive=False
)
# ---------- Sub-tab 3: Preprocess ----------
with gr.Tab("3. Preprocess"):
gr.Markdown(
"Encode audio through VAE and text encoders to create training tensors."
)
lora_preprocess_batch_size = gr.Slider(
minimum=1, maximum=500, value=50, step=1,
label="Preprocess batch size (samples per run)",
)
lora_preprocess_btn = gr.Button(
"Preprocess Batch (+ auto-save)", variant="primary"
)
lora_preprocess_status = gr.Textbox(
label="Preprocess Status", lines=3, interactive=False
)
# ---------- Sub-tab 4: Train ----------
with gr.Tab("4. Train"):
gr.Markdown("Configure and run LoRA training.")
with gr.Row():
with gr.Column():
lora_model_name = gr.Textbox(
label="Model Name",
value="my_lora",
placeholder="my_lora",
)
with gr.Row():
lora_lr = gr.Number(
label="Learning Rate", value=1e-4
)
lora_batch_size = gr.Slider(
minimum=1, maximum=8, value=1, step=1,
label="Batch Size",
)
with gr.Row():
lora_epochs = gr.Slider(
minimum=1, maximum=500, value=100, step=1,
label="Epochs",
)
lora_grad_accum = gr.Slider(
minimum=1, maximum=16, value=4, step=1,
label="Gradient Accumulation",
)
with gr.Row():
lora_rank = gr.Slider(
minimum=4, maximum=128, value=8, step=4,
label="LoRA Rank",
)
lora_alpha = gr.Slider(
minimum=4, maximum=128, value=16, step=4,
label="LoRA Alpha",
)
with gr.Row():
lora_train_btn = gr.Button(
"Start Training",
variant="primary",
size="lg",
)
lora_stop_btn = gr.Button(
"Stop Training",
variant="stop",
size="lg",
)
with gr.Column():
lora_train_status = gr.Textbox(
label="Training Status",
lines=4,
interactive=False,
)
lora_model_path = gr.Textbox(
label="Model Path",
interactive=False,
)
lora_dl_btn = gr.Button("Download Model")
lora_dl_file = gr.File(label="Download")
gr.Markdown("""
#### Tips
- Upload 10+ audio samples for best results
- Keep samples consistent in style/quality
- Higher rank = more capacity but slower training
- Default settings (rank=8, lr=1e-4, 100 epochs) are a good starting point
""")
# ---------- Event handlers ----------
# Data Source
lora_hf_btn.click(
fn=lora_download_hf,
inputs=[lora_hf_id, lora_custom_tag, lora_hf_max, lora_hf_offset, training_state],
outputs=[lora_source_status, training_state, lora_hf_offset, lora_progress],
)
# Label & Review
lora_label_btn.click(
fn=lora_auto_label,
inputs=[lora_label_batch_size, training_state],
outputs=[lora_review_df, lora_label_status, training_state, lora_progress],
)
lora_save_btn.click(
fn=lora_save_edits,
inputs=[lora_review_df, training_state],
outputs=[lora_save_status],
)
lora_save_dataset_btn.click(
fn=lora_save_dataset_to_json,
inputs=[training_state],
outputs=[lora_save_status],
)
# Preprocess
lora_preprocess_btn.click(
fn=lora_preprocess,
inputs=[lora_preprocess_batch_size, training_state],
outputs=[lora_preprocess_status, lora_progress],
)
# Train
lora_train_btn.click(
fn=lora_train_real,
inputs=[
lora_lr, lora_batch_size, lora_epochs,
lora_rank, lora_alpha, lora_grad_accum,
lora_model_name, training_state,
],
outputs=[lora_model_path, lora_train_status],
)
lora_stop_btn.click(
fn=lora_stop_training,
inputs=[],
outputs=[lora_train_status],
)
lora_dl_btn.click(
fn=lora_download_model,
inputs=[lora_model_path],
outputs=[lora_dl_file],
)
gr.Markdown("""
---
### About
ACE-Step 1.5 Custom Edition by Gamahea | Based on [ACE-Step](https://ace-step.github.io/)
""")
return app
# ==================== MAIN ====================
if __name__ == "__main__":
logger.info("Starting ACE-Step 1.5 Custom Edition...")
try:
# Create and launch app
app = create_ui()
# Monkey patch the get_api_info method to prevent JSON schema errors
original_get_api_info = app.get_api_info
def safe_get_api_info(*args, **kwargs):
"""Patched get_api_info that returns minimal info to avoid schema errors"""
try:
return original_get_api_info(*args, **kwargs)
except (TypeError, AttributeError, KeyError) as e:
logger.warning(f"API info generation failed, returning minimal info: {e}")
return {
"named_endpoints": {},
"unnamed_endpoints": {}
}
app.get_api_info = safe_get_api_info
logger.info("✓ Patched get_api_info method")
# Launch the app
app.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
show_error=True
)
except Exception as e:
logger.error(f"Failed to launch app: {e}")
import traceback
traceback.print_exc()
raise