2024-10-12 12:17:41 +08:00
|
|
|
# Copyright © 2024 Apple Inc.
|
|
|
|
|
|
|
|
import argparse
|
|
|
|
import time
|
|
|
|
from functools import partial
|
|
|
|
from pathlib import Path
|
|
|
|
|
|
|
|
import mlx.core as mx
|
|
|
|
import mlx.nn as nn
|
|
|
|
import mlx.optimizers as optim
|
|
|
|
import numpy as np
|
|
|
|
from mlx.nn.utils import average_gradients
|
|
|
|
from mlx.utils import tree_flatten, tree_map, tree_reduce
|
|
|
|
from PIL import Image
|
|
|
|
|
2024-11-09 09:15:19 +08:00
|
|
|
from flux import FluxPipeline, Trainer, load_dataset, save_config
|
2024-10-12 12:17:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
def generate_progress_images(iteration, flux, args):
|
|
|
|
"""Generate images to monitor the progress of the finetuning."""
|
|
|
|
out_dir = Path(args.output_dir)
|
|
|
|
out_dir.mkdir(parents=True, exist_ok=True)
|
|
|
|
out_file = out_dir / f"{iteration:07d}_progress.png"
|
|
|
|
print(f"Generating {str(out_file)}", flush=True)
|
|
|
|
|
|
|
|
# Generate some images and arrange them in a grid
|
|
|
|
n_rows = 2
|
|
|
|
n_images = 4
|
|
|
|
x = flux.generate_images(
|
|
|
|
args.progress_prompt,
|
|
|
|
n_images,
|
|
|
|
args.progress_steps,
|
|
|
|
)
|
|
|
|
x = mx.pad(x, [(0, 0), (4, 4), (4, 4), (0, 0)])
|
|
|
|
B, H, W, C = x.shape
|
|
|
|
x = x.reshape(n_rows, B // n_rows, H, W, C).transpose(0, 2, 1, 3, 4)
|
|
|
|
x = x.reshape(n_rows * H, B // n_rows * W, C)
|
|
|
|
x = mx.pad(x, [(4, 4), (4, 4), (0, 0)])
|
|
|
|
x = (x * 255).astype(mx.uint8)
|
|
|
|
|
|
|
|
# Save them to disc
|
|
|
|
im = Image.fromarray(np.array(x))
|
|
|
|
im.save(out_file)
|
|
|
|
|
|
|
|
|
2024-11-09 09:15:19 +08:00
|
|
|
def save_adapters(adapter_name, flux, args):
|
2024-10-12 12:17:41 +08:00
|
|
|
out_dir = Path(args.output_dir)
|
|
|
|
out_dir.mkdir(parents=True, exist_ok=True)
|
2024-11-09 09:15:19 +08:00
|
|
|
out_file = out_dir / adapter_name
|
2024-10-12 12:17:41 +08:00
|
|
|
print(f"Saving {str(out_file)}")
|
|
|
|
|
|
|
|
mx.save_safetensors(
|
|
|
|
str(out_file),
|
|
|
|
dict(tree_flatten(flux.flow.trainable_parameters())),
|
|
|
|
metadata={
|
|
|
|
"lora_rank": str(args.lora_rank),
|
|
|
|
"lora_blocks": str(args.lora_blocks),
|
|
|
|
},
|
|
|
|
)
|
|
|
|
|
|
|
|
|
2024-10-16 01:37:45 +08:00
|
|
|
def setup_arg_parser():
|
|
|
|
"""Set up and return the argument parser."""
|
2024-10-12 12:17:41 +08:00
|
|
|
parser = argparse.ArgumentParser(
|
|
|
|
description="Finetune Flux to generate images with a specific subject"
|
|
|
|
)
|
|
|
|
|
|
|
|
parser.add_argument(
|
|
|
|
"--model",
|
|
|
|
default="dev",
|
|
|
|
choices=[
|
|
|
|
"dev",
|
|
|
|
"schnell",
|
|
|
|
],
|
|
|
|
help="Which flux model to train",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--guidance", type=float, default=4.0, help="The guidance factor to use."
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--iterations",
|
|
|
|
type=int,
|
|
|
|
default=600,
|
|
|
|
help="How many iterations to train for",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--batch-size",
|
|
|
|
type=int,
|
|
|
|
default=1,
|
|
|
|
help="The batch size to use when training the stable diffusion model",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--resolution",
|
|
|
|
type=lambda x: tuple(map(int, x.split("x"))),
|
|
|
|
default=(512, 512),
|
|
|
|
help="The resolution of the training images",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--num-augmentations",
|
|
|
|
type=int,
|
|
|
|
default=5,
|
|
|
|
help="Augment the images by random cropping and panning",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--progress-prompt",
|
|
|
|
required=True,
|
|
|
|
help="Use this prompt when generating images for evaluation",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--progress-steps",
|
|
|
|
type=int,
|
|
|
|
default=50,
|
|
|
|
help="Use this many steps when generating images for evaluation",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--progress-every",
|
|
|
|
type=int,
|
|
|
|
default=50,
|
|
|
|
help="Generate images every PROGRESS_EVERY steps",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--checkpoint-every",
|
|
|
|
type=int,
|
|
|
|
default=50,
|
|
|
|
help="Save the model every CHECKPOINT_EVERY steps",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--lora-blocks",
|
|
|
|
type=int,
|
|
|
|
default=-1,
|
|
|
|
help="Train the last LORA_BLOCKS transformer blocks",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--lora-rank", type=int, default=8, help="LoRA rank for finetuning"
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--warmup-steps", type=int, default=100, help="Learning rate warmup"
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--learning-rate", type=float, default="1e-4", help="Learning rate for training"
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--grad-accumulate",
|
|
|
|
type=int,
|
|
|
|
default=4,
|
|
|
|
help="Accumulate gradients for that many iterations before applying them",
|
|
|
|
)
|
|
|
|
parser.add_argument(
|
|
|
|
"--output-dir", default="mlx_output", help="Folder to save the checkpoints in"
|
|
|
|
)
|
|
|
|
|
|
|
|
parser.add_argument("dataset")
|
2024-10-16 01:37:45 +08:00
|
|
|
return parser
|
|
|
|
|
2024-10-12 12:17:41 +08:00
|
|
|
|
2024-10-16 01:37:45 +08:00
|
|
|
if __name__ == "__main__":
|
|
|
|
parser = setup_arg_parser()
|
2024-10-12 12:17:41 +08:00
|
|
|
args = parser.parse_args()
|
|
|
|
|
2024-11-09 09:15:19 +08:00
|
|
|
output_path = Path(args.output_dir)
|
|
|
|
output_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
save_config(vars(args), output_path / "adapter_config.json")
|
|
|
|
|
2024-10-12 12:17:41 +08:00
|
|
|
# Load the model and set it up for LoRA training. We use the same random
|
|
|
|
# state when creating the LoRA layers so all workers will have the same
|
|
|
|
# initial weights.
|
|
|
|
mx.random.seed(0x0F0F0F0F)
|
|
|
|
flux = FluxPipeline("flux-" + args.model)
|
|
|
|
flux.flow.freeze()
|
|
|
|
flux.linear_to_lora_layers(args.lora_rank, args.lora_blocks)
|
|
|
|
|
|
|
|
# Reset the seed to a different seed per worker if we are in distributed
|
|
|
|
# mode so that each worker is working on different data, diffusion step and
|
|
|
|
# random noise.
|
|
|
|
mx.random.seed(0xF0F0F0F0 + mx.distributed.init().rank())
|
|
|
|
|
|
|
|
# Report how many parameters we are training
|
|
|
|
trainable_params = tree_reduce(
|
|
|
|
lambda acc, x: acc + x.size, flux.flow.trainable_parameters(), 0
|
|
|
|
)
|
2024-10-16 01:37:45 +08:00
|
|
|
print(f"Training {trainable_params / 1024 ** 2:.3f}M parameters", flush=True)
|
2024-10-12 12:17:41 +08:00
|
|
|
|
|
|
|
# Set up the optimizer and training steps. The steps are a bit verbose to
|
|
|
|
# support gradient accumulation together with compilation.
|
|
|
|
warmup = optim.linear_schedule(0, args.learning_rate, args.warmup_steps)
|
|
|
|
cosine = optim.cosine_decay(
|
|
|
|
args.learning_rate, args.iterations // args.grad_accumulate
|
|
|
|
)
|
|
|
|
lr_schedule = optim.join_schedules([warmup, cosine], [args.warmup_steps])
|
|
|
|
optimizer = optim.Adam(learning_rate=lr_schedule)
|
|
|
|
state = [flux.flow.state, optimizer.state, mx.random.state]
|
|
|
|
|
|
|
|
@partial(mx.compile, inputs=state, outputs=state)
|
|
|
|
def single_step(x, t5_feat, clip_feat, guidance):
|
|
|
|
loss, grads = nn.value_and_grad(flux.flow, flux.training_loss)(
|
|
|
|
x, t5_feat, clip_feat, guidance
|
|
|
|
)
|
|
|
|
grads = average_gradients(grads)
|
|
|
|
optimizer.update(flux.flow, grads)
|
|
|
|
|
|
|
|
return loss
|
|
|
|
|
|
|
|
@partial(mx.compile, inputs=state, outputs=state)
|
|
|
|
def compute_loss_and_grads(x, t5_feat, clip_feat, guidance):
|
|
|
|
return nn.value_and_grad(flux.flow, flux.training_loss)(
|
|
|
|
x, t5_feat, clip_feat, guidance
|
|
|
|
)
|
|
|
|
|
|
|
|
@partial(mx.compile, inputs=state, outputs=state)
|
|
|
|
def compute_loss_and_accumulate_grads(x, t5_feat, clip_feat, guidance, prev_grads):
|
|
|
|
loss, grads = nn.value_and_grad(flux.flow, flux.training_loss)(
|
|
|
|
x, t5_feat, clip_feat, guidance
|
|
|
|
)
|
|
|
|
grads = tree_map(lambda a, b: a + b, prev_grads, grads)
|
|
|
|
return loss, grads
|
|
|
|
|
|
|
|
@partial(mx.compile, inputs=state, outputs=state)
|
|
|
|
def grad_accumulate_and_step(x, t5_feat, clip_feat, guidance, prev_grads):
|
|
|
|
loss, grads = nn.value_and_grad(flux.flow, flux.training_loss)(
|
|
|
|
x, t5_feat, clip_feat, guidance
|
|
|
|
)
|
|
|
|
grads = tree_map(
|
|
|
|
lambda a, b: (a + b) / args.grad_accumulate,
|
|
|
|
prev_grads,
|
|
|
|
grads,
|
|
|
|
)
|
|
|
|
grads = average_gradients(grads)
|
|
|
|
optimizer.update(flux.flow, grads)
|
|
|
|
|
|
|
|
return loss
|
|
|
|
|
|
|
|
# We simply route to the appropriate step based on whether we have
|
|
|
|
# gradients from a previous step and whether we should be performing an
|
|
|
|
# update or simply computing and accumulating gradients in this step.
|
|
|
|
def step(x, t5_feat, clip_feat, guidance, prev_grads, perform_step):
|
|
|
|
if prev_grads is None:
|
|
|
|
if perform_step:
|
|
|
|
return single_step(x, t5_feat, clip_feat, guidance), None
|
|
|
|
else:
|
|
|
|
return compute_loss_and_grads(x, t5_feat, clip_feat, guidance)
|
|
|
|
else:
|
|
|
|
if perform_step:
|
|
|
|
return (
|
|
|
|
grad_accumulate_and_step(
|
|
|
|
x, t5_feat, clip_feat, guidance, prev_grads
|
|
|
|
),
|
|
|
|
None,
|
|
|
|
)
|
|
|
|
else:
|
|
|
|
return compute_loss_and_accumulate_grads(
|
|
|
|
x, t5_feat, clip_feat, guidance, prev_grads
|
|
|
|
)
|
|
|
|
|
2024-10-16 01:37:45 +08:00
|
|
|
dataset = load_dataset(args.dataset)
|
|
|
|
trainer = Trainer(flux, dataset, args)
|
|
|
|
trainer.encode_dataset()
|
|
|
|
|
2024-10-12 12:17:41 +08:00
|
|
|
guidance = mx.full((args.batch_size,), args.guidance, dtype=flux.dtype)
|
|
|
|
|
|
|
|
# An initial generation to compare
|
|
|
|
generate_progress_images(0, flux, args)
|
|
|
|
|
|
|
|
grads = None
|
|
|
|
losses = []
|
|
|
|
tic = time.time()
|
2024-10-16 01:37:45 +08:00
|
|
|
for i, batch in zip(range(args.iterations), trainer.iterate(args.batch_size)):
|
2024-10-12 12:17:41 +08:00
|
|
|
loss, grads = step(*batch, guidance, grads, (i + 1) % args.grad_accumulate == 0)
|
|
|
|
mx.eval(loss, grads, state)
|
|
|
|
losses.append(loss.item())
|
|
|
|
|
|
|
|
if (i + 1) % 10 == 0:
|
|
|
|
toc = time.time()
|
|
|
|
peak_mem = mx.metal.get_peak_memory() / 1024**3
|
|
|
|
print(
|
2024-10-16 01:37:45 +08:00
|
|
|
f"Iter: {i + 1} Loss: {sum(losses) / 10:.3f} "
|
2024-10-12 12:17:41 +08:00
|
|
|
f"It/s: {10 / (toc - tic):.3f} "
|
|
|
|
f"Peak mem: {peak_mem:.3f} GB",
|
|
|
|
flush=True,
|
|
|
|
)
|
|
|
|
|
|
|
|
if (i + 1) % args.progress_every == 0:
|
|
|
|
generate_progress_images(i + 1, flux, args)
|
|
|
|
|
|
|
|
if (i + 1) % args.checkpoint_every == 0:
|
2024-11-09 09:15:19 +08:00
|
|
|
save_adapters(f"{i + 1:07d}_adapters.safetensors", flux, args)
|
2024-10-12 12:17:41 +08:00
|
|
|
|
|
|
|
if (i + 1) % 10 == 0:
|
|
|
|
losses = []
|
|
|
|
tic = time.time()
|
2024-11-09 09:15:19 +08:00
|
|
|
|
|
|
|
save_adapters("final_adapters.safetensors", flux, args)
|
2024-12-09 15:01:53 +08:00
|
|
|
print("Training successful.")
|