Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Abhishek Thakur
commited on
Commit
·
327a449
1
Parent(s):
9f9c9d7
autotrain spacerunner
Browse files- app.py +125 -96
- requirements.autotrain +21 -0
- script.py +0 -0
app.py
CHANGED
|
@@ -1,12 +1,14 @@
|
|
| 1 |
import os
|
| 2 |
-
is_spaces = True if os.environ.get('SPACE_ID') else False
|
| 3 |
|
| 4 |
-
if(
|
|
|
|
|
|
|
| 5 |
import spaces
|
| 6 |
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
|
| 7 |
import sys
|
| 8 |
|
| 9 |
from dotenv import load_dotenv
|
|
|
|
| 10 |
load_dotenv()
|
| 11 |
|
| 12 |
# Add the current working directory to the Python path
|
|
@@ -22,11 +24,14 @@ import json
|
|
| 22 |
import yaml
|
| 23 |
from slugify import slugify
|
| 24 |
from transformers import AutoProcessor, AutoModelForCausalLM
|
| 25 |
-
|
|
|
|
|
|
|
| 26 |
from toolkit.job import get_job
|
| 27 |
|
| 28 |
MAX_IMAGES = 150
|
| 29 |
|
|
|
|
| 30 |
def load_captioning(uploaded_images, concept_sentence):
|
| 31 |
updates = []
|
| 32 |
if len(uploaded_images) <= 1:
|
|
@@ -34,11 +39,9 @@ def load_captioning(uploaded_images, concept_sentence):
|
|
| 34 |
"Please upload at least 2 images to train your model (the ideal number with default settings is between 4-30)"
|
| 35 |
)
|
| 36 |
elif len(uploaded_images) > MAX_IMAGES:
|
| 37 |
-
raise gr.Error(
|
| 38 |
-
f"For now, only {MAX_IMAGES} or less images are allowed for training"
|
| 39 |
-
)
|
| 40 |
# Update for the captioning_area
|
| 41 |
-
#for _ in range(3):
|
| 42 |
updates.append(gr.update(visible=True))
|
| 43 |
# Update visibility and image for each captioning row and image
|
| 44 |
for i in range(1, MAX_IMAGES + 1):
|
|
@@ -50,23 +53,25 @@ def load_captioning(uploaded_images, concept_sentence):
|
|
| 50 |
|
| 51 |
# Update for image component - display image if available, otherwise hide
|
| 52 |
image_value = uploaded_images[i - 1] if visible else None
|
| 53 |
-
|
| 54 |
updates.append(gr.update(value=image_value, visible=visible))
|
| 55 |
|
| 56 |
-
#Update value of captioning area
|
| 57 |
text_value = "[trigger]" if visible and concept_sentence else None
|
| 58 |
updates.append(gr.update(value=text_value, visible=visible))
|
| 59 |
|
| 60 |
-
#Update for the sample caption area
|
| 61 |
updates.append(gr.update(visible=True))
|
| 62 |
updates.append(gr.update(placeholder=f'A photo of {concept_sentence} holding a sign that reads "Hello friend"'))
|
| 63 |
-
updates.append(gr.update(placeholder=f
|
| 64 |
-
updates.append(gr.update(placeholder=f
|
| 65 |
return updates
|
| 66 |
|
| 67 |
-
|
|
|
|
| 68 |
load_captioning = spaces.GPU()(load_captioning)
|
| 69 |
|
|
|
|
| 70 |
def create_dataset(*inputs):
|
| 71 |
print("Creating dataset")
|
| 72 |
images = inputs[0]
|
|
@@ -74,56 +79,60 @@ def create_dataset(*inputs):
|
|
| 74 |
if not os.path.exists(destination_folder):
|
| 75 |
os.makedirs(destination_folder)
|
| 76 |
|
| 77 |
-
jsonl_file_path = os.path.join(destination_folder,
|
| 78 |
-
with open(jsonl_file_path,
|
| 79 |
for index, image in enumerate(images):
|
| 80 |
new_image_path = shutil.copy(image, destination_folder)
|
| 81 |
-
|
| 82 |
original_caption = inputs[index + 1]
|
| 83 |
file_name = os.path.basename(new_image_path)
|
| 84 |
|
| 85 |
data = {"file_name": file_name, "prompt": original_caption}
|
| 86 |
|
| 87 |
jsonl_file.write(json.dumps(data) + "\n")
|
| 88 |
-
|
| 89 |
return destination_folder
|
| 90 |
|
|
|
|
| 91 |
def run_captioning(images, concept_sentence, *captions):
|
| 92 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 93 |
torch_dtype = torch.float16
|
| 94 |
-
model = AutoModelForCausalLM.from_pretrained(
|
|
|
|
|
|
|
| 95 |
processor = AutoProcessor.from_pretrained("microsoft/Florence-2-large", trust_remote_code=True)
|
| 96 |
|
| 97 |
captions = list(captions)
|
| 98 |
for i, image_path in enumerate(images):
|
| 99 |
print(captions[i])
|
| 100 |
if isinstance(image_path, str): # If image is a file path
|
| 101 |
-
image = Image.open(image_path).convert(
|
| 102 |
-
|
| 103 |
prompt = "<DETAILED_CAPTION>"
|
| 104 |
inputs = processor(text=prompt, images=image, return_tensors="pt").to(device, torch_dtype)
|
| 105 |
-
|
| 106 |
generated_ids = model.generate(
|
| 107 |
-
input_ids=inputs["input_ids"],
|
| 108 |
-
pixel_values=inputs["pixel_values"],
|
| 109 |
-
max_new_tokens=1024,
|
| 110 |
-
num_beams=3
|
| 111 |
)
|
| 112 |
-
|
| 113 |
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
| 114 |
-
parsed_answer = processor.post_process_generation(
|
| 115 |
-
|
| 116 |
-
|
|
|
|
|
|
|
| 117 |
caption_text = f"{caption_text} [trigger]"
|
| 118 |
captions[i] = caption_text
|
| 119 |
-
|
| 120 |
-
|
| 121 |
yield captions
|
| 122 |
model.to("cpu")
|
| 123 |
del model
|
| 124 |
del processor
|
| 125 |
-
|
|
|
|
| 126 |
def start_training(
|
|
|
|
|
|
|
| 127 |
lora_name,
|
| 128 |
concept_sentence,
|
| 129 |
steps,
|
|
@@ -144,57 +153,76 @@ def start_training(
|
|
| 144 |
config = yaml.safe_load(f)
|
| 145 |
|
| 146 |
# Update the config with user inputs
|
| 147 |
-
config[
|
| 148 |
-
config[
|
| 149 |
-
config[
|
| 150 |
-
config[
|
| 151 |
-
config[
|
| 152 |
-
config[
|
| 153 |
-
config[
|
| 154 |
-
config[
|
| 155 |
-
if
|
| 156 |
-
config[
|
| 157 |
-
if
|
| 158 |
-
config[
|
| 159 |
-
config[
|
| 160 |
-
config[
|
| 161 |
-
if
|
| 162 |
-
config[
|
| 163 |
-
if
|
| 164 |
-
config[
|
| 165 |
-
if
|
| 166 |
-
config[
|
| 167 |
else:
|
| 168 |
-
config[
|
| 169 |
# Save the updated config
|
| 170 |
config_path = f"config/{slugged_lora_name}.yaml"
|
| 171 |
with open(config_path, "w") as f:
|
| 172 |
yaml.dump(config, f)
|
| 173 |
-
if
|
| 174 |
print("Started training with spacerunner...")
|
| 175 |
-
|
| 176 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 177 |
else:
|
| 178 |
-
#run the job locally
|
| 179 |
job = get_job(config_path)
|
| 180 |
job.run()
|
| 181 |
job.cleanup()
|
| 182 |
|
| 183 |
return f"Training completed successfully. Model saved as {slugged_lora_name}"
|
| 184 |
|
|
|
|
| 185 |
theme = gr.themes.Monochrome(
|
| 186 |
text_size=gr.themes.Size(lg="18px", md="15px", sm="13px", xl="22px", xs="12px", xxl="24px", xxs="9px"),
|
| 187 |
-
font=[gr.themes.GoogleFont(
|
| 188 |
)
|
| 189 |
-
css =
|
| 190 |
#component-1{text-align:center}
|
| 191 |
.main_ui_logged_out{opacity: 0.3; pointer-events: none}
|
| 192 |
.tabitem{border: 0px}
|
| 193 |
-
|
|
|
|
| 194 |
|
| 195 |
def swap_visibilty(profile: gr.OAuthProfile | None):
|
| 196 |
print(profile)
|
| 197 |
-
if
|
| 198 |
if profile is None:
|
| 199 |
return gr.update(elem_classes=["main_ui_logged_out"])
|
| 200 |
else:
|
|
@@ -202,19 +230,26 @@ def swap_visibilty(profile: gr.OAuthProfile | None):
|
|
| 202 |
return gr.update(elem_classes=["main_ui_logged_in"])
|
| 203 |
else:
|
| 204 |
return gr.update(elem_classes=["main_ui_logged_in"])
|
| 205 |
-
|
|
|
|
| 206 |
with gr.Blocks(theme=theme, css=css) as demo:
|
| 207 |
-
gr.Markdown(
|
| 208 |
-
|
| 209 |
-
|
|
|
|
|
|
|
| 210 |
gr.LoginButton("Sign in with Hugging Face to train your LoRA on Spaces", visible=is_spaces)
|
| 211 |
with gr.Tab("Train on Spaces" if is_spaces else "Train locally"):
|
| 212 |
with gr.Column() as main_ui:
|
| 213 |
with gr.Row():
|
| 214 |
-
lora_name = gr.Textbox(
|
| 215 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 216 |
# label="What are you training?", choices=["object", "style", "character", "face", "custom"]
|
| 217 |
-
#)
|
| 218 |
concept_sentence = gr.Textbox(
|
| 219 |
label="Trigger word/sentence",
|
| 220 |
info="Trigger word or sentence to be used",
|
|
@@ -233,9 +268,11 @@ with gr.Blocks(theme=theme, css=css) as demo:
|
|
| 233 |
)
|
| 234 |
with gr.Column(scale=3, visible=False) as captioning_area:
|
| 235 |
with gr.Column():
|
| 236 |
-
gr.Markdown(
|
|
|
|
| 237 |
You can optionally add a custom caption for each image (or use an AI model for this). [trigger] will represent your concept sentence/trigger word.
|
| 238 |
-
"""
|
|
|
|
| 239 |
do_captioning = gr.Button("Add AI captions with Florence-2")
|
| 240 |
output_components = [captioning_area]
|
| 241 |
caption_list = []
|
|
@@ -251,28 +288,30 @@ with gr.Blocks(theme=theme, css=css) as demo:
|
|
| 251 |
scale=2,
|
| 252 |
show_label=False,
|
| 253 |
show_share_button=False,
|
| 254 |
-
show_download_button=False
|
| 255 |
)
|
| 256 |
locals()[f"caption_{i}"] = gr.Textbox(
|
| 257 |
label=f"Caption {i}", scale=15, interactive=True
|
| 258 |
)
|
| 259 |
-
|
| 260 |
output_components.append(locals()[f"captioning_row_{i}"])
|
| 261 |
output_components.append(locals()[f"image_{i}"])
|
| 262 |
output_components.append(locals()[f"caption_{i}"])
|
| 263 |
caption_list.append(locals()[f"caption_{i}"])
|
| 264 |
-
|
| 265 |
with gr.Accordion("Advanced options", open=False):
|
| 266 |
steps = gr.Number(label="Steps", value=1000, minimum=1, maximum=10000, step=1)
|
| 267 |
lr = gr.Number(label="Learning Rate", value=4e-4, minimum=1e-6, maximum=1e-3, step=1e-6)
|
| 268 |
rank = gr.Number(label="LoRA Rank", value=16, minimum=4, maximum=128, step=4)
|
| 269 |
-
|
| 270 |
with gr.Accordion("Sample prompts", visible=False) as sample:
|
| 271 |
-
gr.Markdown(
|
|
|
|
|
|
|
| 272 |
sample_1 = gr.Textbox(label="Test prompt 1")
|
| 273 |
sample_2 = gr.Textbox(label="Test prompt 2")
|
| 274 |
sample_3 = gr.Textbox(label="Test prompt 3")
|
| 275 |
-
|
| 276 |
output_components.append(sample)
|
| 277 |
output_components.append(sample_1)
|
| 278 |
output_components.append(sample_2)
|
|
@@ -281,7 +320,8 @@ with gr.Blocks(theme=theme, css=css) as demo:
|
|
| 281 |
progress_area = gr.Markdown("")
|
| 282 |
|
| 283 |
with gr.Tab("Train locally" if is_spaces else "Instructions"):
|
| 284 |
-
gr.Markdown(
|
|
|
|
| 285 |
```bash
|
| 286 |
git clone https://huggingface.co/spaces/flux-train/flux-lora-trainer
|
| 287 |
cd flux-lora-trainer
|
|
@@ -312,23 +352,14 @@ with gr.Blocks(theme=theme, css=css) as demo:
|
|
| 312 |
python app.py
|
| 313 |
```
|
| 314 |
If you prefer command line, you can run Ostris' [AI Toolkit](https://github.com/ostris/ai-toolkit) yourself directly.
|
| 315 |
-
|
| 316 |
-
|
|
|
|
| 317 |
dataset_folder = gr.State()
|
| 318 |
|
| 319 |
-
images.upload(
|
| 320 |
-
load_captioning,
|
| 321 |
-
inputs=[images, concept_sentence],
|
| 322 |
-
outputs=output_components,
|
| 323 |
-
queue=False
|
| 324 |
-
)
|
| 325 |
|
| 326 |
-
start.click(
|
| 327 |
-
fn=create_dataset,
|
| 328 |
-
inputs=[images] + caption_list,
|
| 329 |
-
outputs=dataset_folder,
|
| 330 |
-
queue=False
|
| 331 |
-
).then(
|
| 332 |
fn=start_training,
|
| 333 |
inputs=[
|
| 334 |
lora_name,
|
|
@@ -342,14 +373,12 @@ with gr.Blocks(theme=theme, css=css) as demo:
|
|
| 342 |
sample_3,
|
| 343 |
],
|
| 344 |
outputs=progress_area,
|
| 345 |
-
queue=False
|
| 346 |
)
|
| 347 |
|
| 348 |
-
do_captioning.click(
|
| 349 |
-
fn=run_captioning, inputs=[images, concept_sentence] + caption_list, outputs=caption_list
|
| 350 |
-
)
|
| 351 |
demo.load(fn=swap_visibilty, outputs=main_ui, queue=False)
|
| 352 |
|
| 353 |
if __name__ == "__main__":
|
| 354 |
demo.queue()
|
| 355 |
-
demo.launch(share=True)
|
|
|
|
| 1 |
import os
|
|
|
|
| 2 |
|
| 3 |
+
is_spaces = True if os.environ.get("SPACE_ID") else False
|
| 4 |
+
|
| 5 |
+
if is_spaces:
|
| 6 |
import spaces
|
| 7 |
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
|
| 8 |
import sys
|
| 9 |
|
| 10 |
from dotenv import load_dotenv
|
| 11 |
+
|
| 12 |
load_dotenv()
|
| 13 |
|
| 14 |
# Add the current working directory to the Python path
|
|
|
|
| 24 |
import yaml
|
| 25 |
from slugify import slugify
|
| 26 |
from transformers import AutoProcessor, AutoModelForCausalLM
|
| 27 |
+
import subprocess
|
| 28 |
+
|
| 29 |
+
if not is_spaces:
|
| 30 |
from toolkit.job import get_job
|
| 31 |
|
| 32 |
MAX_IMAGES = 150
|
| 33 |
|
| 34 |
+
|
| 35 |
def load_captioning(uploaded_images, concept_sentence):
|
| 36 |
updates = []
|
| 37 |
if len(uploaded_images) <= 1:
|
|
|
|
| 39 |
"Please upload at least 2 images to train your model (the ideal number with default settings is between 4-30)"
|
| 40 |
)
|
| 41 |
elif len(uploaded_images) > MAX_IMAGES:
|
| 42 |
+
raise gr.Error(f"For now, only {MAX_IMAGES} or less images are allowed for training")
|
|
|
|
|
|
|
| 43 |
# Update for the captioning_area
|
| 44 |
+
# for _ in range(3):
|
| 45 |
updates.append(gr.update(visible=True))
|
| 46 |
# Update visibility and image for each captioning row and image
|
| 47 |
for i in range(1, MAX_IMAGES + 1):
|
|
|
|
| 53 |
|
| 54 |
# Update for image component - display image if available, otherwise hide
|
| 55 |
image_value = uploaded_images[i - 1] if visible else None
|
| 56 |
+
|
| 57 |
updates.append(gr.update(value=image_value, visible=visible))
|
| 58 |
|
| 59 |
+
# Update value of captioning area
|
| 60 |
text_value = "[trigger]" if visible and concept_sentence else None
|
| 61 |
updates.append(gr.update(value=text_value, visible=visible))
|
| 62 |
|
| 63 |
+
# Update for the sample caption area
|
| 64 |
updates.append(gr.update(visible=True))
|
| 65 |
updates.append(gr.update(placeholder=f'A photo of {concept_sentence} holding a sign that reads "Hello friend"'))
|
| 66 |
+
updates.append(gr.update(placeholder=f"A mountainous landscape in the style of {concept_sentence}"))
|
| 67 |
+
updates.append(gr.update(placeholder=f"A {concept_sentence} in a mall"))
|
| 68 |
return updates
|
| 69 |
|
| 70 |
+
|
| 71 |
+
if is_spaces:
|
| 72 |
load_captioning = spaces.GPU()(load_captioning)
|
| 73 |
|
| 74 |
+
|
| 75 |
def create_dataset(*inputs):
|
| 76 |
print("Creating dataset")
|
| 77 |
images = inputs[0]
|
|
|
|
| 79 |
if not os.path.exists(destination_folder):
|
| 80 |
os.makedirs(destination_folder)
|
| 81 |
|
| 82 |
+
jsonl_file_path = os.path.join(destination_folder, "metadata.jsonl")
|
| 83 |
+
with open(jsonl_file_path, "a") as jsonl_file:
|
| 84 |
for index, image in enumerate(images):
|
| 85 |
new_image_path = shutil.copy(image, destination_folder)
|
| 86 |
+
|
| 87 |
original_caption = inputs[index + 1]
|
| 88 |
file_name = os.path.basename(new_image_path)
|
| 89 |
|
| 90 |
data = {"file_name": file_name, "prompt": original_caption}
|
| 91 |
|
| 92 |
jsonl_file.write(json.dumps(data) + "\n")
|
| 93 |
+
|
| 94 |
return destination_folder
|
| 95 |
|
| 96 |
+
|
| 97 |
def run_captioning(images, concept_sentence, *captions):
|
| 98 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 99 |
torch_dtype = torch.float16
|
| 100 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 101 |
+
"microsoft/Florence-2-large", torch_dtype=torch_dtype, trust_remote_code=True
|
| 102 |
+
).to(device)
|
| 103 |
processor = AutoProcessor.from_pretrained("microsoft/Florence-2-large", trust_remote_code=True)
|
| 104 |
|
| 105 |
captions = list(captions)
|
| 106 |
for i, image_path in enumerate(images):
|
| 107 |
print(captions[i])
|
| 108 |
if isinstance(image_path, str): # If image is a file path
|
| 109 |
+
image = Image.open(image_path).convert("RGB")
|
| 110 |
+
|
| 111 |
prompt = "<DETAILED_CAPTION>"
|
| 112 |
inputs = processor(text=prompt, images=image, return_tensors="pt").to(device, torch_dtype)
|
| 113 |
+
|
| 114 |
generated_ids = model.generate(
|
| 115 |
+
input_ids=inputs["input_ids"], pixel_values=inputs["pixel_values"], max_new_tokens=1024, num_beams=3
|
|
|
|
|
|
|
|
|
|
| 116 |
)
|
| 117 |
+
|
| 118 |
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
| 119 |
+
parsed_answer = processor.post_process_generation(
|
| 120 |
+
generated_text, task=prompt, image_size=(image.width, image.height)
|
| 121 |
+
)
|
| 122 |
+
caption_text = parsed_answer["<DETAILED_CAPTION>"].replace("The image shows ", "")
|
| 123 |
+
if concept_sentence:
|
| 124 |
caption_text = f"{caption_text} [trigger]"
|
| 125 |
captions[i] = caption_text
|
| 126 |
+
|
|
|
|
| 127 |
yield captions
|
| 128 |
model.to("cpu")
|
| 129 |
del model
|
| 130 |
del processor
|
| 131 |
+
|
| 132 |
+
|
| 133 |
def start_training(
|
| 134 |
+
profile: gr.OAuthProfile | None,
|
| 135 |
+
oauth_token: gr.OAuthToken | None,
|
| 136 |
lora_name,
|
| 137 |
concept_sentence,
|
| 138 |
steps,
|
|
|
|
| 153 |
config = yaml.safe_load(f)
|
| 154 |
|
| 155 |
# Update the config with user inputs
|
| 156 |
+
config["config"]["name"] = slugged_lora_name
|
| 157 |
+
config["config"]["process"][0]["model"]["low_vram"] = True
|
| 158 |
+
config["config"]["process"][0]["train"]["skip_first_sample"] = True
|
| 159 |
+
config["config"]["process"][0]["train"]["steps"] = int(steps)
|
| 160 |
+
config["config"]["process"][0]["train"]["lr"] = float(lr)
|
| 161 |
+
config["config"]["process"][0]["network"]["linear"] = int(rank)
|
| 162 |
+
config["config"]["process"][0]["network"]["linear_alpha"] = int(rank)
|
| 163 |
+
config["config"]["process"][0]["datasets"][0]["folder_path"] = dataset_folder
|
| 164 |
+
if concept_sentence:
|
| 165 |
+
config["config"]["process"][0]["trigger_word"] = concept_sentence
|
| 166 |
+
if sample_1 or sample_2 or sample_2:
|
| 167 |
+
config["config"]["process"][0]["train"]["disable_sampling"] = False
|
| 168 |
+
config["config"]["process"][0]["sample"]["sample_every"] = steps
|
| 169 |
+
config["config"]["process"][0]["sample"]["prompts"] = []
|
| 170 |
+
if sample_1:
|
| 171 |
+
config["config"]["process"][0]["sample"]["prompts"].append(sample_1)
|
| 172 |
+
if sample_2:
|
| 173 |
+
config["config"]["process"][0]["sample"]["prompts"].append(sample_2)
|
| 174 |
+
if sample_3:
|
| 175 |
+
config["config"]["process"][0]["sample"]["prompts"].append(sample_3)
|
| 176 |
else:
|
| 177 |
+
config["config"]["process"][0]["train"]["disable_sampling"] = True
|
| 178 |
# Save the updated config
|
| 179 |
config_path = f"config/{slugged_lora_name}.yaml"
|
| 180 |
with open(config_path, "w") as f:
|
| 181 |
yaml.dump(config, f)
|
| 182 |
+
if is_spaces:
|
| 183 |
print("Started training with spacerunner...")
|
| 184 |
+
# copy config to dataset_folder
|
| 185 |
+
shutil.copy(config_path, dataset_folder)
|
| 186 |
+
# get location of this script
|
| 187 |
+
script_location = os.path.dirname(os.path.abspath(__file__))
|
| 188 |
+
# copy script.py from current directory to dataset_folder
|
| 189 |
+
shutil.copy(script_location + "/script.py", dataset_folder)
|
| 190 |
+
# copy requirements.autotrain to dataset_folder as requirements.txt
|
| 191 |
+
shutil.copy(script_location + "/requirements.autotrain", dataset_folder + "/requirements.txt")
|
| 192 |
+
# command to run autotrain spacerunner
|
| 193 |
+
cmd = f"autotrain spacerunner --project-name {slugged_lora_name} --script-path {dataset_folder}"
|
| 194 |
+
cmd += f" --username {profile.name} --token {oauth_token} --backend spaces-l4x1"
|
| 195 |
+
outcome = subprocess.run(cmd)
|
| 196 |
+
if outcome.returncode == 0:
|
| 197 |
+
return f"""# Your training has started.
|
| 198 |
+
## - Training Status: <a href='https://huggingface.co/spaces/{profile.name}/autotrain-{slugged_lora_name}?logs=container'>{profile.name}/autotrain-{slugged_lora_name}</a> <small>(in the logs tab)</small>
|
| 199 |
+
## - Model page: <a href='https://huggingface.co/{profile.name}/{slugged_lora_name}'>{profile.name}/{slugged_lora_name}</a> <small>(will be available when training finishes)</small>"""
|
| 200 |
+
else:
|
| 201 |
+
print("Error: ", outcome.stderr)
|
| 202 |
+
raise gr.Error("Something went wrong. Make sure the name of your LoRA is unique and try again")
|
| 203 |
else:
|
| 204 |
+
# run the job locally
|
| 205 |
job = get_job(config_path)
|
| 206 |
job.run()
|
| 207 |
job.cleanup()
|
| 208 |
|
| 209 |
return f"Training completed successfully. Model saved as {slugged_lora_name}"
|
| 210 |
|
| 211 |
+
|
| 212 |
theme = gr.themes.Monochrome(
|
| 213 |
text_size=gr.themes.Size(lg="18px", md="15px", sm="13px", xl="22px", xs="12px", xxl="24px", xxs="9px"),
|
| 214 |
+
font=[gr.themes.GoogleFont("Source Sans Pro"), "ui-sans-serif", "system-ui", "sans-serif"],
|
| 215 |
)
|
| 216 |
+
css = """
|
| 217 |
#component-1{text-align:center}
|
| 218 |
.main_ui_logged_out{opacity: 0.3; pointer-events: none}
|
| 219 |
.tabitem{border: 0px}
|
| 220 |
+
"""
|
| 221 |
+
|
| 222 |
|
| 223 |
def swap_visibilty(profile: gr.OAuthProfile | None):
|
| 224 |
print(profile)
|
| 225 |
+
if is_spaces:
|
| 226 |
if profile is None:
|
| 227 |
return gr.update(elem_classes=["main_ui_logged_out"])
|
| 228 |
else:
|
|
|
|
| 230 |
return gr.update(elem_classes=["main_ui_logged_in"])
|
| 231 |
else:
|
| 232 |
return gr.update(elem_classes=["main_ui_logged_in"])
|
| 233 |
+
|
| 234 |
+
|
| 235 |
with gr.Blocks(theme=theme, css=css) as demo:
|
| 236 |
+
gr.Markdown(
|
| 237 |
+
"""# LoRA Ease for FLUX 🧞♂️
|
| 238 |
+
### Train a high quality FLUX LoRA in a breeze ༄ using [Ostris' AI Toolkit](https://github.com/ostris/ai-toolkit) and [AutoTrain Advanced](https://github.com/huggingface/autotrain-advanced)"""
|
| 239 |
+
)
|
| 240 |
+
if is_spaces:
|
| 241 |
gr.LoginButton("Sign in with Hugging Face to train your LoRA on Spaces", visible=is_spaces)
|
| 242 |
with gr.Tab("Train on Spaces" if is_spaces else "Train locally"):
|
| 243 |
with gr.Column() as main_ui:
|
| 244 |
with gr.Row():
|
| 245 |
+
lora_name = gr.Textbox(
|
| 246 |
+
label="The name of your LoRA",
|
| 247 |
+
info="This has to be a unique name",
|
| 248 |
+
placeholder="e.g.: Persian Miniature Painting style, Cat Toy",
|
| 249 |
+
)
|
| 250 |
+
# training_option = gr.Radio(
|
| 251 |
# label="What are you training?", choices=["object", "style", "character", "face", "custom"]
|
| 252 |
+
# )
|
| 253 |
concept_sentence = gr.Textbox(
|
| 254 |
label="Trigger word/sentence",
|
| 255 |
info="Trigger word or sentence to be used",
|
|
|
|
| 268 |
)
|
| 269 |
with gr.Column(scale=3, visible=False) as captioning_area:
|
| 270 |
with gr.Column():
|
| 271 |
+
gr.Markdown(
|
| 272 |
+
"""# Custom captioning
|
| 273 |
You can optionally add a custom caption for each image (or use an AI model for this). [trigger] will represent your concept sentence/trigger word.
|
| 274 |
+
"""
|
| 275 |
+
)
|
| 276 |
do_captioning = gr.Button("Add AI captions with Florence-2")
|
| 277 |
output_components = [captioning_area]
|
| 278 |
caption_list = []
|
|
|
|
| 288 |
scale=2,
|
| 289 |
show_label=False,
|
| 290 |
show_share_button=False,
|
| 291 |
+
show_download_button=False,
|
| 292 |
)
|
| 293 |
locals()[f"caption_{i}"] = gr.Textbox(
|
| 294 |
label=f"Caption {i}", scale=15, interactive=True
|
| 295 |
)
|
| 296 |
+
|
| 297 |
output_components.append(locals()[f"captioning_row_{i}"])
|
| 298 |
output_components.append(locals()[f"image_{i}"])
|
| 299 |
output_components.append(locals()[f"caption_{i}"])
|
| 300 |
caption_list.append(locals()[f"caption_{i}"])
|
| 301 |
+
|
| 302 |
with gr.Accordion("Advanced options", open=False):
|
| 303 |
steps = gr.Number(label="Steps", value=1000, minimum=1, maximum=10000, step=1)
|
| 304 |
lr = gr.Number(label="Learning Rate", value=4e-4, minimum=1e-6, maximum=1e-3, step=1e-6)
|
| 305 |
rank = gr.Number(label="LoRA Rank", value=16, minimum=4, maximum=128, step=4)
|
| 306 |
+
|
| 307 |
with gr.Accordion("Sample prompts", visible=False) as sample:
|
| 308 |
+
gr.Markdown(
|
| 309 |
+
"Include sample prompts to test out your trained model. Don't forget to include your trigger word/sentence (optional)"
|
| 310 |
+
)
|
| 311 |
sample_1 = gr.Textbox(label="Test prompt 1")
|
| 312 |
sample_2 = gr.Textbox(label="Test prompt 2")
|
| 313 |
sample_3 = gr.Textbox(label="Test prompt 3")
|
| 314 |
+
|
| 315 |
output_components.append(sample)
|
| 316 |
output_components.append(sample_1)
|
| 317 |
output_components.append(sample_2)
|
|
|
|
| 320 |
progress_area = gr.Markdown("")
|
| 321 |
|
| 322 |
with gr.Tab("Train locally" if is_spaces else "Instructions"):
|
| 323 |
+
gr.Markdown(
|
| 324 |
+
f"""To use FLUX LoRA Ease locally with this UI, you can clone this repository (yes, HF Spaces are git repos!)
|
| 325 |
```bash
|
| 326 |
git clone https://huggingface.co/spaces/flux-train/flux-lora-trainer
|
| 327 |
cd flux-lora-trainer
|
|
|
|
| 352 |
python app.py
|
| 353 |
```
|
| 354 |
If you prefer command line, you can run Ostris' [AI Toolkit](https://github.com/ostris/ai-toolkit) yourself directly.
|
| 355 |
+
"""
|
| 356 |
+
)
|
| 357 |
+
|
| 358 |
dataset_folder = gr.State()
|
| 359 |
|
| 360 |
+
images.upload(load_captioning, inputs=[images, concept_sentence], outputs=output_components, queue=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 361 |
|
| 362 |
+
start.click(fn=create_dataset, inputs=[images] + caption_list, outputs=dataset_folder, queue=False).then(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 363 |
fn=start_training,
|
| 364 |
inputs=[
|
| 365 |
lora_name,
|
|
|
|
| 373 |
sample_3,
|
| 374 |
],
|
| 375 |
outputs=progress_area,
|
| 376 |
+
queue=False,
|
| 377 |
)
|
| 378 |
|
| 379 |
+
do_captioning.click(fn=run_captioning, inputs=[images, concept_sentence] + caption_list, outputs=caption_list)
|
|
|
|
|
|
|
| 380 |
demo.load(fn=swap_visibilty, outputs=main_ui, queue=False)
|
| 381 |
|
| 382 |
if __name__ == "__main__":
|
| 383 |
demo.queue()
|
| 384 |
+
demo.launch(share=True)
|
requirements.autotrain
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
git+https://github.com/huggingface/diffusers.git
|
| 2 |
+
lycoris-lora==1.8.3
|
| 3 |
+
flatten_json
|
| 4 |
+
pyyaml
|
| 5 |
+
oyaml
|
| 6 |
+
tensorboard
|
| 7 |
+
kornia
|
| 8 |
+
invisible-watermark
|
| 9 |
+
einops
|
| 10 |
+
toml
|
| 11 |
+
albumentations
|
| 12 |
+
pydantic
|
| 13 |
+
omegaconf
|
| 14 |
+
k-diffusion
|
| 15 |
+
open_clip_torch
|
| 16 |
+
prodigyopt
|
| 17 |
+
controlnet_aux==0.0.7
|
| 18 |
+
python-dotenv
|
| 19 |
+
lpips
|
| 20 |
+
pytorch_fid
|
| 21 |
+
optimum-quanto
|
script.py
ADDED
|
File without changes
|