Spaces:
Running
Running
File size: 13,105 Bytes
ed37502 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 | """LoRA training service — train custom LoRA models from reference images.
Wraps Kohya's sd-scripts for LoRA training with sensible defaults for
character LoRAs on SD 1.5 / RealisticVision. Manages the full pipeline:
dataset preparation, config generation, training launch, and output handling.
Requirements (installed automatically on first use):
- kohya sd-scripts (cloned from GitHub)
- accelerate, lion-pytorch, prodigy-optimizer
"""
from __future__ import annotations
import asyncio
import json
import logging
import os
import shutil
import subprocess
import sys
import time
import uuid
from dataclasses import dataclass, field
from pathlib import Path
from typing import Any
logger = logging.getLogger(__name__)
IS_HF_SPACES = os.environ.get("HF_SPACES") == "1" or os.environ.get("SPACE_ID") is not None
if IS_HF_SPACES:
TRAINING_BASE_DIR = Path("/app/data/training")
LORA_OUTPUT_DIR = Path("/app/data/loras")
else:
TRAINING_BASE_DIR = Path("D:/AI automation/content_engine/training")
LORA_OUTPUT_DIR = Path("D:/ComfyUI/Models/Lora")
SD_SCRIPTS_DIR = TRAINING_BASE_DIR / "sd-scripts"
def _default_base_model() -> str:
"""Get default base model path based on environment."""
if IS_HF_SPACES:
return "/app/models/realisticVisionV51_v51VAE.safetensors"
return "D:/ComfyUI/Models/StableDiffusion/realisticVisionV51_v51VAE.safetensors"
@dataclass
class TrainingConfig:
"""Configuration for a LoRA training job."""
name: str
base_model: str = "" # Set in __post_init__
resolution: int = 512
train_batch_size: int = 1
num_epochs: int = 10
learning_rate: float = 1e-4
network_rank: int = 32 # LoRA rank (dim)
network_alpha: int = 16
optimizer: str = "AdamW8bit" # AdamW8bit, Lion, Prodigy
lr_scheduler: str = "cosine_with_restarts"
max_train_steps: int | None = None # If set, overrides epochs
save_every_n_epochs: int = 2
clip_skip: int = 1
mixed_precision: str = "fp16"
seed: int = 42
caption_extension: str = ".txt"
trigger_word: str = ""
extra_args: dict[str, Any] = field(default_factory=dict)
def __post_init__(self):
if not self.base_model:
self.base_model = _default_base_model()
@dataclass
class TrainingJob:
"""Tracks state of a running or completed training job."""
id: str
name: str
config: TrainingConfig
status: str = "pending" # pending, preparing, training, completed, failed
progress: float = 0.0
current_epoch: int = 0
total_epochs: int = 0
current_step: int = 0
total_steps: int = 0
loss: float | None = None
started_at: float | None = None
completed_at: float | None = None
output_path: str | None = None
error: str | None = None
log_lines: list[str] = field(default_factory=list)
class LoRATrainer:
"""Manages LoRA training jobs using Kohya sd-scripts."""
def __init__(self):
self._jobs: dict[str, TrainingJob] = {}
self._processes: dict[str, asyncio.subprocess.Process] = {}
TRAINING_BASE_DIR.mkdir(parents=True, exist_ok=True)
@property
def sd_scripts_installed(self) -> bool:
return (SD_SCRIPTS_DIR / "train_network.py").exists()
async def install_sd_scripts(self) -> str:
"""Clone and set up Kohya sd-scripts. Returns status message."""
if self.sd_scripts_installed:
return "sd-scripts already installed"
SD_SCRIPTS_DIR.parent.mkdir(parents=True, exist_ok=True)
logger.info("Cloning kohya sd-scripts...")
proc = await asyncio.create_subprocess_exec(
"git", "clone", "--depth", "1",
"https://github.com/kohya-ss/sd-scripts.git",
str(SD_SCRIPTS_DIR),
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE,
)
stdout, stderr = await proc.communicate()
if proc.returncode != 0:
raise RuntimeError(f"Failed to clone sd-scripts: {stderr.decode()}")
# Install requirements
logger.info("Installing sd-scripts requirements...")
proc = await asyncio.create_subprocess_exec(
sys.executable, "-m", "pip", "install",
"accelerate", "lion-pytorch", "prodigy-optimizer",
"safetensors", "diffusers", "transformers",
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.PIPE,
)
await proc.communicate()
logger.info("sd-scripts installation complete")
return "sd-scripts installed successfully"
def prepare_dataset(self, job_id: str, image_paths: list[str], trigger_word: str = "") -> Path:
"""Prepare a training dataset directory with proper structure.
Creates: training/{job_id}/dataset/{num_repeats}_{trigger_word}/
Each image gets a caption file with the trigger word.
"""
dataset_dir = TRAINING_BASE_DIR / job_id / "dataset"
# Convention: {repeats}_{concept_name}
repeats = 10
concept_dir = dataset_dir / f"{repeats}_{trigger_word or 'character'}"
concept_dir.mkdir(parents=True, exist_ok=True)
for img_path in image_paths:
src = Path(img_path)
if not src.exists():
logger.warning("Image not found: %s", img_path)
continue
dst = concept_dir / src.name
shutil.copy2(src, dst)
# Create caption file
caption_file = dst.with_suffix(".txt")
caption_file.write_text(trigger_word or "")
return dataset_dir
async def start_training(self, config: TrainingConfig, image_paths: list[str]) -> str:
"""Start a LoRA training job. Returns the job ID."""
job_id = str(uuid.uuid4())[:8]
if not self.sd_scripts_installed:
await self.install_sd_scripts()
job = TrainingJob(
id=job_id,
name=config.name,
config=config,
status="preparing",
total_epochs=config.num_epochs,
)
self._jobs[job_id] = job
# Prepare dataset
try:
dataset_dir = self.prepare_dataset(job_id, image_paths, config.trigger_word)
except Exception as e:
job.status = "failed"
job.error = f"Dataset preparation failed: {e}"
return job_id
# Create output directory
output_dir = TRAINING_BASE_DIR / job_id / "output"
output_dir.mkdir(parents=True, exist_ok=True)
# Build training command
cmd = self._build_training_command(config, dataset_dir, output_dir)
job.log_lines.append(f"Command: {' '.join(cmd)}")
# Launch training process
job.status = "training"
job.started_at = time.time()
asyncio.create_task(self._run_training(job_id, cmd, output_dir, config))
return job_id
def _build_training_command(
self, config: TrainingConfig, dataset_dir: Path, output_dir: Path
) -> list[str]:
"""Build the training command for Kohya sd-scripts."""
cmd = [
sys.executable,
str(SD_SCRIPTS_DIR / "train_network.py"),
f"--pretrained_model_name_or_path={config.base_model}",
f"--train_data_dir={dataset_dir}",
f"--output_dir={output_dir}",
f"--output_name={config.name}",
f"--resolution={config.resolution}",
f"--train_batch_size={config.train_batch_size}",
f"--max_train_epochs={config.num_epochs}",
f"--learning_rate={config.learning_rate}",
f"--network_module=networks.lora",
f"--network_dim={config.network_rank}",
f"--network_alpha={config.network_alpha}",
f"--optimizer_type={config.optimizer}",
f"--lr_scheduler={config.lr_scheduler}",
f"--save_every_n_epochs={config.save_every_n_epochs}",
f"--clip_skip={config.clip_skip}",
f"--mixed_precision={config.mixed_precision}",
f"--seed={config.seed}",
f"--caption_extension={config.caption_extension}",
"--cache_latents",
"--enable_bucket",
"--xformers",
"--save_model_as=safetensors",
]
if config.max_train_steps:
cmd.append(f"--max_train_steps={config.max_train_steps}")
return cmd
async def _run_training(
self, job_id: str, cmd: list[str], output_dir: Path, config: TrainingConfig
):
"""Run the training process and monitor progress."""
job = self._jobs[job_id]
try:
proc = await asyncio.create_subprocess_exec(
*cmd,
stdout=asyncio.subprocess.PIPE,
stderr=asyncio.subprocess.STDOUT,
cwd=str(SD_SCRIPTS_DIR),
)
self._processes[job_id] = proc
# Read output lines and parse progress
async for line_bytes in proc.stdout:
line = line_bytes.decode("utf-8", errors="replace").strip()
if not line:
continue
job.log_lines.append(line)
# Keep last 200 lines
if len(job.log_lines) > 200:
job.log_lines = job.log_lines[-200:]
# Parse progress from Kohya output
if "epoch" in line.lower() and "/" in line:
try:
# Look for patterns like "epoch 3/10"
parts = line.lower().split("epoch")
if len(parts) > 1:
ep_part = parts[1].strip().split()[0]
if "/" in ep_part:
current, total = ep_part.split("/")
job.current_epoch = int(current)
job.total_epochs = int(total)
job.progress = job.current_epoch / max(job.total_epochs, 1)
except (ValueError, IndexError):
pass
if "loss=" in line or "loss:" in line:
try:
loss_str = line.split("loss")[1].strip("=: ").split()[0].strip(",")
job.loss = float(loss_str)
except (ValueError, IndexError):
pass
if "steps:" in line.lower() or "step " in line.lower():
try:
import re
step_match = re.search(r"(\d+)/(\d+)", line)
if step_match:
job.current_step = int(step_match.group(1))
job.total_steps = int(step_match.group(2))
if job.total_steps > 0:
job.progress = job.current_step / job.total_steps
except (ValueError, IndexError):
pass
await proc.wait()
if proc.returncode == 0:
job.status = "completed"
job.progress = 1.0
job.completed_at = time.time()
# Find the output LoRA file and copy to ComfyUI
lora_file = output_dir / f"{config.name}.safetensors"
if lora_file.exists():
dest = LORA_OUTPUT_DIR / lora_file.name
LORA_OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
shutil.copy2(lora_file, dest)
job.output_path = str(dest)
logger.info("Training complete! LoRA saved to %s", dest)
else:
# Check for epoch-saved versions
for f in sorted(output_dir.glob("*.safetensors")):
dest = LORA_OUTPUT_DIR / f.name
shutil.copy2(f, dest)
job.output_path = str(dest)
logger.info("Training complete! Output in %s", output_dir)
else:
job.status = "failed"
job.error = f"Training process exited with code {proc.returncode}"
logger.error("Training failed: %s", job.error)
except Exception as e:
job.status = "failed"
job.error = str(e)
logger.error("Training error: %s", e, exc_info=True)
finally:
self._processes.pop(job_id, None)
def get_job(self, job_id: str) -> TrainingJob | None:
return self._jobs.get(job_id)
def list_jobs(self) -> list[TrainingJob]:
return list(self._jobs.values())
async def cancel_job(self, job_id: str) -> bool:
"""Cancel a running training job."""
proc = self._processes.get(job_id)
if proc:
proc.terminate()
self._processes.pop(job_id, None)
job = self._jobs.get(job_id)
if job:
job.status = "failed"
job.error = "Cancelled by user"
return True
return False
|