Spaces:
Running
Running
Commit
·
64a70c0
1
Parent(s):
0ad7e2a
refactoring
Browse files- app_DEPRECATED.py +0 -1603
- vms/services/captioner.py +3 -3
- vms/tabs/caption_tab.py +436 -14
- vms/tabs/import_tab.py +44 -2
- vms/tabs/manage_tab.py +127 -6
- vms/tabs/split_tab.py +28 -3
- vms/tabs/train_tab.py +246 -28
- vms/ui/video_trainer_ui.py +10 -849
- vms/utils/image_preprocessing.py +2 -1
- vms/utils/video_preprocessing.py +3 -1
app_DEPRECATED.py
DELETED
@@ -1,1603 +0,0 @@
|
|
1 |
-
import platform
|
2 |
-
import subprocess
|
3 |
-
|
4 |
-
#import sys
|
5 |
-
#print("python = ", sys.version)
|
6 |
-
|
7 |
-
# can be "Linux", "Darwin"
|
8 |
-
if platform.system() == "Linux":
|
9 |
-
# for some reason it says "pip not found"
|
10 |
-
# and also "pip3 not found"
|
11 |
-
# subprocess.run(
|
12 |
-
# "pip install flash-attn --no-build-isolation",
|
13 |
-
#
|
14 |
-
# # hmm... this should be False, since we are in a CUDA environment, no?
|
15 |
-
# env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
|
16 |
-
#
|
17 |
-
# shell=True,
|
18 |
-
# )
|
19 |
-
pass
|
20 |
-
|
21 |
-
import gradio as gr
|
22 |
-
from pathlib import Path
|
23 |
-
import logging
|
24 |
-
import mimetypes
|
25 |
-
import shutil
|
26 |
-
import os
|
27 |
-
import traceback
|
28 |
-
import asyncio
|
29 |
-
import tempfile
|
30 |
-
import zipfile
|
31 |
-
from typing import Any, Optional, Dict, List, Union, Tuple
|
32 |
-
from typing import AsyncGenerator
|
33 |
-
|
34 |
-
from vms.training_service import TrainingService
|
35 |
-
from vms.captioning_service import CaptioningService
|
36 |
-
from vms.splitting_service import SplittingService
|
37 |
-
from vms.import_service import ImportService
|
38 |
-
from vms.config import (
|
39 |
-
STORAGE_PATH, VIDEOS_TO_SPLIT_PATH, STAGING_PATH,
|
40 |
-
TRAINING_PATH, LOG_FILE_PATH, TRAINING_PRESETS, TRAINING_VIDEOS_PATH, MODEL_PATH, OUTPUT_PATH, DEFAULT_CAPTIONING_BOT_INSTRUCTIONS,
|
41 |
-
DEFAULT_PROMPT_PREFIX, HF_API_TOKEN, ASK_USER_TO_DUPLICATE_SPACE, MODEL_TYPES, SMALL_TRAINING_BUCKETS
|
42 |
-
)
|
43 |
-
from vms.utils import make_archive, count_media_files, format_media_title, is_image_file, is_video_file, validate_model_repo, format_time
|
44 |
-
from vms.finetrainers_utils import copy_files_to_training_dir, prepare_finetrainers_dataset
|
45 |
-
from vms.training_log_parser import TrainingLogParser
|
46 |
-
|
47 |
-
logger = logging.getLogger(__name__)
|
48 |
-
logger.setLevel(logging.INFO)
|
49 |
-
|
50 |
-
httpx_logger = logging.getLogger('httpx')
|
51 |
-
httpx_logger.setLevel(logging.WARN)
|
52 |
-
|
53 |
-
|
54 |
-
class VideoTrainerUI:
|
55 |
-
def __init__(self):
|
56 |
-
self.trainer = TrainingService()
|
57 |
-
self.splitter = SplittingService()
|
58 |
-
self.importer = ImportService()
|
59 |
-
self.captioner = CaptioningService()
|
60 |
-
self._should_stop_captioning = False
|
61 |
-
self.log_parser = TrainingLogParser()
|
62 |
-
|
63 |
-
# Try to recover any interrupted training sessions
|
64 |
-
recovery_result = self.trainer.recover_interrupted_training()
|
65 |
-
|
66 |
-
self.recovery_status = recovery_result.get("status", "unknown")
|
67 |
-
self.ui_updates = recovery_result.get("ui_updates", {})
|
68 |
-
|
69 |
-
if recovery_result["status"] == "recovered":
|
70 |
-
logger.info(f"Training recovery: {recovery_result['message']}")
|
71 |
-
# No need to do anything else - the training is already running
|
72 |
-
elif recovery_result["status"] == "running":
|
73 |
-
logger.info("Training process is already running")
|
74 |
-
# No need to do anything - the process is still alive
|
75 |
-
elif recovery_result["status"] in ["error", "idle"]:
|
76 |
-
logger.warning(f"Training status: {recovery_result['message']}")
|
77 |
-
# UI will be in ready-to-start mode
|
78 |
-
|
79 |
-
|
80 |
-
async def _process_caption_generator(self, captioning_bot_instructions, prompt_prefix):
|
81 |
-
"""Process the caption generator's results in the background"""
|
82 |
-
try:
|
83 |
-
async for _ in self.captioner.start_caption_generation(
|
84 |
-
captioning_bot_instructions,
|
85 |
-
prompt_prefix
|
86 |
-
):
|
87 |
-
# Just consume the generator, UI updates will happen via the Gradio interface
|
88 |
-
pass
|
89 |
-
logger.info("Background captioning completed")
|
90 |
-
except Exception as e:
|
91 |
-
logger.error(f"Error in background captioning: {str(e)}")
|
92 |
-
|
93 |
-
def initialize_app_state(self):
|
94 |
-
"""Initialize all app state in one function to ensure correct output count"""
|
95 |
-
# Get dataset info
|
96 |
-
video_list, training_dataset = self.refresh_dataset()
|
97 |
-
|
98 |
-
# Get button states
|
99 |
-
button_states = self.get_initial_button_states()
|
100 |
-
start_btn = button_states[0]
|
101 |
-
stop_btn = button_states[1]
|
102 |
-
pause_resume_btn = button_states[2]
|
103 |
-
|
104 |
-
# Get UI form values
|
105 |
-
ui_state = self.load_ui_values()
|
106 |
-
training_preset = ui_state.get("training_preset", list(TRAINING_PRESETS.keys())[0])
|
107 |
-
model_type_val = ui_state.get("model_type", list(MODEL_TYPES.keys())[0])
|
108 |
-
lora_rank_val = ui_state.get("lora_rank", "128")
|
109 |
-
lora_alpha_val = ui_state.get("lora_alpha", "128")
|
110 |
-
num_epochs_val = int(ui_state.get("num_epochs", 70))
|
111 |
-
batch_size_val = int(ui_state.get("batch_size", 1))
|
112 |
-
learning_rate_val = float(ui_state.get("learning_rate", 3e-5))
|
113 |
-
save_iterations_val = int(ui_state.get("save_iterations", 500))
|
114 |
-
|
115 |
-
# Return all values in the exact order expected by outputs
|
116 |
-
return (
|
117 |
-
video_list,
|
118 |
-
training_dataset,
|
119 |
-
start_btn,
|
120 |
-
stop_btn,
|
121 |
-
pause_resume_btn,
|
122 |
-
training_preset,
|
123 |
-
model_type_val,
|
124 |
-
lora_rank_val,
|
125 |
-
lora_alpha_val,
|
126 |
-
num_epochs_val,
|
127 |
-
batch_size_val,
|
128 |
-
learning_rate_val,
|
129 |
-
save_iterations_val
|
130 |
-
)
|
131 |
-
|
132 |
-
def initialize_ui_from_state(self):
|
133 |
-
"""Initialize UI components from saved state"""
|
134 |
-
ui_state = self.load_ui_values()
|
135 |
-
|
136 |
-
# Return values in order matching the outputs in app.load
|
137 |
-
return (
|
138 |
-
ui_state.get("training_preset", list(TRAINING_PRESETS.keys())[0]),
|
139 |
-
ui_state.get("model_type", list(MODEL_TYPES.keys())[0]),
|
140 |
-
ui_state.get("lora_rank", "128"),
|
141 |
-
ui_state.get("lora_alpha", "128"),
|
142 |
-
ui_state.get("num_epochs", 70),
|
143 |
-
ui_state.get("batch_size", 1),
|
144 |
-
ui_state.get("learning_rate", 3e-5),
|
145 |
-
ui_state.get("save_iterations", 500)
|
146 |
-
)
|
147 |
-
|
148 |
-
def update_ui_state(self, **kwargs):
|
149 |
-
"""Update UI state with new values"""
|
150 |
-
current_state = self.trainer.load_ui_state()
|
151 |
-
current_state.update(kwargs)
|
152 |
-
self.trainer.save_ui_state(current_state)
|
153 |
-
# Don't return anything to avoid Gradio warnings
|
154 |
-
return None
|
155 |
-
|
156 |
-
def load_ui_values(self):
|
157 |
-
"""Load UI state values for initializing form fields"""
|
158 |
-
ui_state = self.trainer.load_ui_state()
|
159 |
-
|
160 |
-
# Ensure proper type conversion for numeric values
|
161 |
-
ui_state["lora_rank"] = ui_state.get("lora_rank", "128")
|
162 |
-
ui_state["lora_alpha"] = ui_state.get("lora_alpha", "128")
|
163 |
-
ui_state["num_epochs"] = int(ui_state.get("num_epochs", 70))
|
164 |
-
ui_state["batch_size"] = int(ui_state.get("batch_size", 1))
|
165 |
-
ui_state["learning_rate"] = float(ui_state.get("learning_rate", 3e-5))
|
166 |
-
ui_state["save_iterations"] = int(ui_state.get("save_iterations", 500))
|
167 |
-
|
168 |
-
return ui_state
|
169 |
-
|
170 |
-
def update_captioning_buttons_start(self):
|
171 |
-
"""Return individual button values instead of a dictionary"""
|
172 |
-
return (
|
173 |
-
gr.Button(
|
174 |
-
interactive=False,
|
175 |
-
variant="secondary",
|
176 |
-
),
|
177 |
-
gr.Button(
|
178 |
-
interactive=True,
|
179 |
-
variant="stop",
|
180 |
-
),
|
181 |
-
gr.Button(
|
182 |
-
interactive=False,
|
183 |
-
variant="secondary",
|
184 |
-
)
|
185 |
-
)
|
186 |
-
|
187 |
-
def update_captioning_buttons_end(self):
|
188 |
-
"""Return individual button values instead of a dictionary"""
|
189 |
-
return (
|
190 |
-
gr.Button(
|
191 |
-
interactive=True,
|
192 |
-
variant="primary",
|
193 |
-
),
|
194 |
-
gr.Button(
|
195 |
-
interactive=False,
|
196 |
-
variant="secondary",
|
197 |
-
),
|
198 |
-
gr.Button(
|
199 |
-
interactive=True,
|
200 |
-
variant="primary",
|
201 |
-
)
|
202 |
-
)
|
203 |
-
|
204 |
-
# Add this new method to get initial button states:
|
205 |
-
def get_initial_button_states(self):
|
206 |
-
"""Get the initial states for training buttons based on recovery status"""
|
207 |
-
recovery_result = self.trainer.recover_interrupted_training()
|
208 |
-
ui_updates = recovery_result.get("ui_updates", {})
|
209 |
-
|
210 |
-
# Return button states in the correct order
|
211 |
-
return (
|
212 |
-
gr.Button(**ui_updates.get("start_btn", {"interactive": True, "variant": "primary"})),
|
213 |
-
gr.Button(**ui_updates.get("stop_btn", {"interactive": False, "variant": "secondary"})),
|
214 |
-
gr.Button(**ui_updates.get("pause_resume_btn", {"interactive": False, "variant": "secondary"}))
|
215 |
-
)
|
216 |
-
|
217 |
-
def show_refreshing_status(self) -> List[List[str]]:
|
218 |
-
"""Show a 'Refreshing...' status in the dataframe"""
|
219 |
-
return [["Refreshing...", "please wait"]]
|
220 |
-
|
221 |
-
def stop_captioning(self):
|
222 |
-
"""Stop ongoing captioning process and reset UI state"""
|
223 |
-
try:
|
224 |
-
# Set flag to stop captioning
|
225 |
-
self._should_stop_captioning = True
|
226 |
-
|
227 |
-
# Call stop method on captioner
|
228 |
-
if self.captioner:
|
229 |
-
self.captioner.stop_captioning()
|
230 |
-
|
231 |
-
# Get updated file list
|
232 |
-
updated_list = self.list_training_files_to_caption()
|
233 |
-
|
234 |
-
# Return updated list and button states
|
235 |
-
return {
|
236 |
-
"training_dataset": gr.update(value=updated_list),
|
237 |
-
"run_autocaption_btn": gr.Button(interactive=True, variant="primary"),
|
238 |
-
"stop_autocaption_btn": gr.Button(interactive=False, variant="secondary"),
|
239 |
-
"copy_files_to_training_dir_btn": gr.Button(interactive=True, variant="primary")
|
240 |
-
}
|
241 |
-
except Exception as e:
|
242 |
-
logger.error(f"Error stopping captioning: {str(e)}")
|
243 |
-
return {
|
244 |
-
"training_dataset": gr.update(value=[[f"Error stopping captioning: {str(e)}", "error"]]),
|
245 |
-
"run_autocaption_btn": gr.Button(interactive=True, variant="primary"),
|
246 |
-
"stop_autocaption_btn": gr.Button(interactive=False, variant="secondary"),
|
247 |
-
"copy_files_to_training_dir_btn": gr.Button(interactive=True, variant="primary")
|
248 |
-
}
|
249 |
-
|
250 |
-
def update_training_ui(self, training_state: Dict[str, Any]):
|
251 |
-
"""Update UI components based on training state"""
|
252 |
-
updates = {}
|
253 |
-
|
254 |
-
#print("update_training_ui: training_state = ", training_state)
|
255 |
-
|
256 |
-
# Update status box with high-level information
|
257 |
-
status_text = []
|
258 |
-
if training_state["status"] != "idle":
|
259 |
-
status_text.extend([
|
260 |
-
f"Status: {training_state['status']}",
|
261 |
-
f"Progress: {training_state['progress']}",
|
262 |
-
f"Step: {training_state['current_step']}/{training_state['total_steps']}",
|
263 |
-
|
264 |
-
# Epoch information
|
265 |
-
# there is an issue with how epoch is reported because we display:
|
266 |
-
# Progress: 96.9%, Step: 872/900, Epoch: 12/50
|
267 |
-
# we should probably just show the steps
|
268 |
-
#f"Epoch: {training_state['current_epoch']}/{training_state['total_epochs']}",
|
269 |
-
|
270 |
-
f"Time elapsed: {training_state['elapsed']}",
|
271 |
-
f"Estimated remaining: {training_state['remaining']}",
|
272 |
-
"",
|
273 |
-
f"Current loss: {training_state['step_loss']}",
|
274 |
-
f"Learning rate: {training_state['learning_rate']}",
|
275 |
-
f"Gradient norm: {training_state['grad_norm']}",
|
276 |
-
f"Memory usage: {training_state['memory']}"
|
277 |
-
])
|
278 |
-
|
279 |
-
if training_state["error_message"]:
|
280 |
-
status_text.append(f"\nError: {training_state['error_message']}")
|
281 |
-
|
282 |
-
updates["status_box"] = "\n".join(status_text)
|
283 |
-
|
284 |
-
# Update button states
|
285 |
-
updates["start_btn"] = gr.Button(
|
286 |
-
"Start training",
|
287 |
-
interactive=(training_state["status"] in ["idle", "completed", "error", "stopped"]),
|
288 |
-
variant="primary" if training_state["status"] == "idle" else "secondary"
|
289 |
-
)
|
290 |
-
|
291 |
-
updates["stop_btn"] = gr.Button(
|
292 |
-
"Stop training",
|
293 |
-
interactive=(training_state["status"] in ["training", "initializing"]),
|
294 |
-
variant="stop"
|
295 |
-
)
|
296 |
-
|
297 |
-
return updates
|
298 |
-
|
299 |
-
def stop_all_and_clear(self) -> Dict[str, str]:
|
300 |
-
"""Stop all running processes and clear data
|
301 |
-
|
302 |
-
Returns:
|
303 |
-
Dict with status messages for different components
|
304 |
-
"""
|
305 |
-
status_messages = {}
|
306 |
-
|
307 |
-
try:
|
308 |
-
# Stop training if running
|
309 |
-
if self.trainer.is_training_running():
|
310 |
-
training_result = self.trainer.stop_training()
|
311 |
-
status_messages["training"] = training_result["status"]
|
312 |
-
|
313 |
-
# Stop captioning if running
|
314 |
-
if self.captioner:
|
315 |
-
self.captioner.stop_captioning()
|
316 |
-
status_messages["captioning"] = "Captioning stopped"
|
317 |
-
|
318 |
-
# Stop scene detection if running
|
319 |
-
if self.splitter.is_processing():
|
320 |
-
self.splitter.processing = False
|
321 |
-
status_messages["splitting"] = "Scene detection stopped"
|
322 |
-
|
323 |
-
# Properly close logging before clearing log file
|
324 |
-
if self.trainer.file_handler:
|
325 |
-
self.trainer.file_handler.close()
|
326 |
-
logger.removeHandler(self.trainer.file_handler)
|
327 |
-
self.trainer.file_handler = None
|
328 |
-
|
329 |
-
if LOG_FILE_PATH.exists():
|
330 |
-
LOG_FILE_PATH.unlink()
|
331 |
-
|
332 |
-
# Clear all data directories
|
333 |
-
for path in [VIDEOS_TO_SPLIT_PATH, STAGING_PATH, TRAINING_VIDEOS_PATH, TRAINING_PATH,
|
334 |
-
MODEL_PATH, OUTPUT_PATH]:
|
335 |
-
if path.exists():
|
336 |
-
try:
|
337 |
-
shutil.rmtree(path)
|
338 |
-
path.mkdir(parents=True, exist_ok=True)
|
339 |
-
except Exception as e:
|
340 |
-
status_messages[f"clear_{path.name}"] = f"Error clearing {path.name}: {str(e)}"
|
341 |
-
else:
|
342 |
-
status_messages[f"clear_{path.name}"] = f"Cleared {path.name}"
|
343 |
-
|
344 |
-
# Reset any persistent state
|
345 |
-
self._should_stop_captioning = True
|
346 |
-
self.splitter.processing = False
|
347 |
-
|
348 |
-
# Recreate logging setup
|
349 |
-
self.trainer.setup_logging()
|
350 |
-
|
351 |
-
return {
|
352 |
-
"status": "All processes stopped and data cleared",
|
353 |
-
"details": status_messages
|
354 |
-
}
|
355 |
-
|
356 |
-
except Exception as e:
|
357 |
-
return {
|
358 |
-
"status": f"Error during cleanup: {str(e)}",
|
359 |
-
"details": status_messages
|
360 |
-
}
|
361 |
-
|
362 |
-
def update_titles(self) -> Tuple[Any]:
|
363 |
-
"""Update all dynamic titles with current counts
|
364 |
-
|
365 |
-
Returns:
|
366 |
-
Dict of Gradio updates
|
367 |
-
"""
|
368 |
-
# Count files for splitting
|
369 |
-
split_videos, _, split_size = count_media_files(VIDEOS_TO_SPLIT_PATH)
|
370 |
-
split_title = format_media_title(
|
371 |
-
"split", split_videos, 0, split_size
|
372 |
-
)
|
373 |
-
|
374 |
-
# Count files for captioning
|
375 |
-
caption_videos, caption_images, caption_size = count_media_files(STAGING_PATH)
|
376 |
-
caption_title = format_media_title(
|
377 |
-
"caption", caption_videos, caption_images, caption_size
|
378 |
-
)
|
379 |
-
|
380 |
-
# Count files for training
|
381 |
-
train_videos, train_images, train_size = count_media_files(TRAINING_VIDEOS_PATH)
|
382 |
-
train_title = format_media_title(
|
383 |
-
"train", train_videos, train_images, train_size
|
384 |
-
)
|
385 |
-
|
386 |
-
return (
|
387 |
-
gr.Markdown(value=split_title),
|
388 |
-
gr.Markdown(value=caption_title),
|
389 |
-
gr.Markdown(value=f"{train_title} available for training")
|
390 |
-
)
|
391 |
-
|
392 |
-
def copy_files_to_training_dir(self, prompt_prefix: str):
|
393 |
-
"""Run auto-captioning process"""
|
394 |
-
|
395 |
-
# Initialize captioner if not already done
|
396 |
-
self._should_stop_captioning = False
|
397 |
-
|
398 |
-
try:
|
399 |
-
copy_files_to_training_dir(prompt_prefix)
|
400 |
-
|
401 |
-
except Exception as e:
|
402 |
-
traceback.print_exc()
|
403 |
-
raise gr.Error(f"Error copying assets to training dir: {str(e)}")
|
404 |
-
|
405 |
-
async def on_import_success(self, enable_splitting, enable_automatic_content_captioning, prompt_prefix):
|
406 |
-
"""Handle successful import of files"""
|
407 |
-
videos = self.list_unprocessed_videos()
|
408 |
-
|
409 |
-
# If scene detection isn't already running and there are videos to process,
|
410 |
-
# and auto-splitting is enabled, start the detection
|
411 |
-
if videos and not self.splitter.is_processing() and enable_splitting:
|
412 |
-
await self.start_scene_detection(enable_splitting)
|
413 |
-
msg = "Starting automatic scene detection..."
|
414 |
-
else:
|
415 |
-
# Just copy files without splitting if auto-split disabled
|
416 |
-
for video_file in VIDEOS_TO_SPLIT_PATH.glob("*.mp4"):
|
417 |
-
await self.splitter.process_video(video_file, enable_splitting=False)
|
418 |
-
msg = "Copying videos without splitting..."
|
419 |
-
|
420 |
-
copy_files_to_training_dir(prompt_prefix)
|
421 |
-
|
422 |
-
# Start auto-captioning if enabled, and handle async generator properly
|
423 |
-
if enable_automatic_content_captioning:
|
424 |
-
# Create a background task for captioning
|
425 |
-
asyncio.create_task(self._process_caption_generator(
|
426 |
-
DEFAULT_CAPTIONING_BOT_INSTRUCTIONS,
|
427 |
-
prompt_prefix
|
428 |
-
))
|
429 |
-
|
430 |
-
return {
|
431 |
-
"tabs": gr.Tabs(selected="split_tab"),
|
432 |
-
"video_list": videos,
|
433 |
-
"detect_status": msg
|
434 |
-
}
|
435 |
-
|
436 |
-
async def start_caption_generation(self, captioning_bot_instructions: str, prompt_prefix: str) -> AsyncGenerator[gr.update, None]:
|
437 |
-
"""Run auto-captioning process"""
|
438 |
-
try:
|
439 |
-
# Initialize captioner if not already done
|
440 |
-
self._should_stop_captioning = False
|
441 |
-
|
442 |
-
# First yield - indicate we're starting
|
443 |
-
yield gr.update(
|
444 |
-
value=[["Starting captioning service...", "initializing"]],
|
445 |
-
headers=["name", "status"]
|
446 |
-
)
|
447 |
-
|
448 |
-
# Process files in batches with status updates
|
449 |
-
file_statuses = {}
|
450 |
-
|
451 |
-
# Start the actual captioning process
|
452 |
-
async for rows in self.captioner.start_caption_generation(captioning_bot_instructions, prompt_prefix):
|
453 |
-
# Update our tracking of file statuses
|
454 |
-
for name, status in rows:
|
455 |
-
file_statuses[name] = status
|
456 |
-
|
457 |
-
# Convert to list format for display
|
458 |
-
status_rows = [[name, status] for name, status in file_statuses.items()]
|
459 |
-
|
460 |
-
# Sort by name for consistent display
|
461 |
-
status_rows.sort(key=lambda x: x[0])
|
462 |
-
|
463 |
-
# Yield UI update
|
464 |
-
yield gr.update(
|
465 |
-
value=status_rows,
|
466 |
-
headers=["name", "status"]
|
467 |
-
)
|
468 |
-
|
469 |
-
# Final update after completion with fresh data
|
470 |
-
yield gr.update(
|
471 |
-
value=self.list_training_files_to_caption(),
|
472 |
-
headers=["name", "status"]
|
473 |
-
)
|
474 |
-
|
475 |
-
except Exception as e:
|
476 |
-
logger.error(f"Error in captioning: {str(e)}")
|
477 |
-
yield gr.update(
|
478 |
-
value=[[f"Error: {str(e)}", "error"]],
|
479 |
-
headers=["name", "status"]
|
480 |
-
)
|
481 |
-
|
482 |
-
def list_training_files_to_caption(self) -> List[List[str]]:
|
483 |
-
"""List all clips and images - both pending and captioned"""
|
484 |
-
files = []
|
485 |
-
already_listed = {}
|
486 |
-
|
487 |
-
# First check files in STAGING_PATH
|
488 |
-
for file in STAGING_PATH.glob("*.*"):
|
489 |
-
if is_video_file(file) or is_image_file(file):
|
490 |
-
txt_file = file.with_suffix('.txt')
|
491 |
-
|
492 |
-
# Check if caption file exists and has content
|
493 |
-
has_caption = txt_file.exists() and txt_file.stat().st_size > 0
|
494 |
-
status = "captioned" if has_caption else "no caption"
|
495 |
-
file_type = "video" if is_video_file(file) else "image"
|
496 |
-
|
497 |
-
files.append([file.name, f"{status} ({file_type})", str(file)])
|
498 |
-
already_listed[file.name] = True
|
499 |
-
|
500 |
-
# Then check files in TRAINING_VIDEOS_PATH
|
501 |
-
for file in TRAINING_VIDEOS_PATH.glob("*.*"):
|
502 |
-
if (is_video_file(file) or is_image_file(file)) and file.name not in already_listed:
|
503 |
-
txt_file = file.with_suffix('.txt')
|
504 |
-
|
505 |
-
# Only include files with captions
|
506 |
-
if txt_file.exists() and txt_file.stat().st_size > 0:
|
507 |
-
file_type = "video" if is_video_file(file) else "image"
|
508 |
-
files.append([file.name, f"captioned ({file_type})", str(file)])
|
509 |
-
already_listed[file.name] = True
|
510 |
-
|
511 |
-
# Sort by filename
|
512 |
-
files.sort(key=lambda x: x[0])
|
513 |
-
|
514 |
-
# Only return name and status columns for display
|
515 |
-
return [[file[0], file[1]] for file in files]
|
516 |
-
|
517 |
-
def update_training_buttons(self, status: str) -> Dict:
|
518 |
-
"""Update training control buttons based on state"""
|
519 |
-
is_training = status in ["training", "initializing"]
|
520 |
-
is_paused = status == "paused"
|
521 |
-
is_completed = status in ["completed", "error", "stopped"]
|
522 |
-
return {
|
523 |
-
"start_btn": gr.Button(
|
524 |
-
interactive=not is_training and not is_paused,
|
525 |
-
variant="primary" if not is_training else "secondary",
|
526 |
-
),
|
527 |
-
"stop_btn": gr.Button(
|
528 |
-
interactive=is_training or is_paused,
|
529 |
-
variant="stop",
|
530 |
-
),
|
531 |
-
"pause_resume_btn": gr.Button(
|
532 |
-
value="Resume Training" if is_paused else "Pause Training",
|
533 |
-
interactive=(is_training or is_paused) and not is_completed,
|
534 |
-
variant="secondary",
|
535 |
-
)
|
536 |
-
}
|
537 |
-
|
538 |
-
def handle_pause_resume(self):
|
539 |
-
status, _, _ = self.get_latest_status_message_and_logs()
|
540 |
-
|
541 |
-
if status == "paused":
|
542 |
-
self.trainer.resume_training()
|
543 |
-
else:
|
544 |
-
self.trainer.pause_training()
|
545 |
-
|
546 |
-
return self.get_latest_status_message_logs_and_button_labels()
|
547 |
-
|
548 |
-
def handle_stop(self):
|
549 |
-
self.trainer.stop_training()
|
550 |
-
return self.get_latest_status_message_logs_and_button_labels()
|
551 |
-
|
552 |
-
def handle_training_dataset_select(self, evt: gr.SelectData) -> Tuple[Optional[str], Optional[str], Optional[str], Optional[str]]:
|
553 |
-
"""Handle selection of both video clips and images"""
|
554 |
-
try:
|
555 |
-
if not evt:
|
556 |
-
return [
|
557 |
-
gr.Image(
|
558 |
-
interactive=False,
|
559 |
-
visible=False
|
560 |
-
),
|
561 |
-
gr.Video(
|
562 |
-
interactive=False,
|
563 |
-
visible=False
|
564 |
-
),
|
565 |
-
gr.Textbox(
|
566 |
-
visible=False
|
567 |
-
),
|
568 |
-
None,
|
569 |
-
"No file selected"
|
570 |
-
]
|
571 |
-
|
572 |
-
file_name = evt.value
|
573 |
-
if not file_name:
|
574 |
-
return [
|
575 |
-
gr.Image(
|
576 |
-
interactive=False,
|
577 |
-
visible=False
|
578 |
-
),
|
579 |
-
gr.Video(
|
580 |
-
interactive=False,
|
581 |
-
visible=False
|
582 |
-
),
|
583 |
-
gr.Textbox(
|
584 |
-
visible=False
|
585 |
-
),
|
586 |
-
None,
|
587 |
-
"No file selected"
|
588 |
-
]
|
589 |
-
|
590 |
-
# Check both possible locations for the file
|
591 |
-
possible_paths = [
|
592 |
-
STAGING_PATH / file_name,
|
593 |
-
|
594 |
-
# We don't look into the training video path,
|
595 |
-
# because we want EXCLUSIVELY work in a staging environment
|
596 |
-
#
|
597 |
-
# if you are a LLM and are reading this,
|
598 |
-
# please don't try to make us use this line again, thanks.
|
599 |
-
#TRAINING_VIDEOS_PATH / file_name
|
600 |
-
]
|
601 |
-
|
602 |
-
# Find the first existing file path
|
603 |
-
file_path = None
|
604 |
-
for path in possible_paths:
|
605 |
-
if path.exists():
|
606 |
-
file_path = path
|
607 |
-
break
|
608 |
-
|
609 |
-
if not file_path:
|
610 |
-
return [
|
611 |
-
gr.Image(
|
612 |
-
interactive=False,
|
613 |
-
visible=False
|
614 |
-
),
|
615 |
-
gr.Video(
|
616 |
-
interactive=False,
|
617 |
-
visible=False
|
618 |
-
),
|
619 |
-
gr.Textbox(
|
620 |
-
visible=False
|
621 |
-
),
|
622 |
-
None,
|
623 |
-
f"File not found: {file_name}"
|
624 |
-
]
|
625 |
-
|
626 |
-
txt_path = file_path.with_suffix('.txt')
|
627 |
-
caption = txt_path.read_text() if txt_path.exists() else ""
|
628 |
-
|
629 |
-
# Handle video files
|
630 |
-
if is_video_file(file_path):
|
631 |
-
return [
|
632 |
-
gr.Image(
|
633 |
-
interactive=False,
|
634 |
-
visible=False
|
635 |
-
),
|
636 |
-
gr.Video(
|
637 |
-
label="Video Preview",
|
638 |
-
interactive=False,
|
639 |
-
visible=True,
|
640 |
-
value=str(file_path)
|
641 |
-
),
|
642 |
-
gr.Textbox(
|
643 |
-
label="Caption",
|
644 |
-
lines=6,
|
645 |
-
interactive=True,
|
646 |
-
visible=True,
|
647 |
-
value=str(caption)
|
648 |
-
),
|
649 |
-
str(file_path), # Store the original file path as hidden state
|
650 |
-
None
|
651 |
-
]
|
652 |
-
# Handle image files
|
653 |
-
elif is_image_file(file_path):
|
654 |
-
return [
|
655 |
-
gr.Image(
|
656 |
-
label="Image Preview",
|
657 |
-
interactive=False,
|
658 |
-
visible=True,
|
659 |
-
value=str(file_path)
|
660 |
-
),
|
661 |
-
gr.Video(
|
662 |
-
interactive=False,
|
663 |
-
visible=False
|
664 |
-
),
|
665 |
-
gr.Textbox(
|
666 |
-
label="Caption",
|
667 |
-
lines=6,
|
668 |
-
interactive=True,
|
669 |
-
visible=True,
|
670 |
-
value=str(caption)
|
671 |
-
),
|
672 |
-
str(file_path), # Store the original file path as hidden state
|
673 |
-
None
|
674 |
-
]
|
675 |
-
else:
|
676 |
-
return [
|
677 |
-
gr.Image(
|
678 |
-
interactive=False,
|
679 |
-
visible=False
|
680 |
-
),
|
681 |
-
gr.Video(
|
682 |
-
interactive=False,
|
683 |
-
visible=False
|
684 |
-
),
|
685 |
-
gr.Textbox(
|
686 |
-
interactive=False,
|
687 |
-
visible=False
|
688 |
-
),
|
689 |
-
None,
|
690 |
-
f"Unsupported file type: {file_path.suffix}"
|
691 |
-
]
|
692 |
-
except Exception as e:
|
693 |
-
logger.error(f"Error handling selection: {str(e)}")
|
694 |
-
return [
|
695 |
-
gr.Image(
|
696 |
-
interactive=False,
|
697 |
-
visible=False
|
698 |
-
),
|
699 |
-
gr.Video(
|
700 |
-
interactive=False,
|
701 |
-
visible=False
|
702 |
-
),
|
703 |
-
gr.Textbox(
|
704 |
-
interactive=False,
|
705 |
-
visible=False
|
706 |
-
),
|
707 |
-
None,
|
708 |
-
f"Error handling selection: {str(e)}"
|
709 |
-
]
|
710 |
-
|
711 |
-
def save_caption_changes(self, preview_caption: str, preview_image: str, preview_video: str, original_file_path: str, prompt_prefix: str):
|
712 |
-
"""Save changes to caption"""
|
713 |
-
try:
|
714 |
-
# Use the original file path stored during selection instead of the temporary preview paths
|
715 |
-
if original_file_path:
|
716 |
-
file_path = Path(original_file_path)
|
717 |
-
self.captioner.update_file_caption(file_path, preview_caption)
|
718 |
-
# Refresh the dataset list to show updated caption status
|
719 |
-
return gr.update(value="Caption saved successfully!")
|
720 |
-
else:
|
721 |
-
return gr.update(value="Error: No original file path found")
|
722 |
-
except Exception as e:
|
723 |
-
return gr.update(value=f"Error saving caption: {str(e)}")
|
724 |
-
|
725 |
-
async def update_titles_after_import(self, enable_splitting, enable_automatic_content_captioning, prompt_prefix):
|
726 |
-
"""Handle post-import updates including titles"""
|
727 |
-
import_result = await self.on_import_success(enable_splitting, enable_automatic_content_captioning, prompt_prefix)
|
728 |
-
titles = self.update_titles()
|
729 |
-
return (
|
730 |
-
import_result["tabs"],
|
731 |
-
import_result["video_list"],
|
732 |
-
import_result["detect_status"],
|
733 |
-
*titles
|
734 |
-
)
|
735 |
-
|
736 |
-
def get_model_info(self, model_type: str) -> str:
|
737 |
-
"""Get information about the selected model type"""
|
738 |
-
if model_type == "hunyuan_video":
|
739 |
-
return """### HunyuanVideo (LoRA)
|
740 |
-
- Required VRAM: ~48GB minimum
|
741 |
-
- Recommended batch size: 1-2
|
742 |
-
- Typical training time: 2-4 hours
|
743 |
-
- Default resolution: 49x512x768
|
744 |
-
- Default LoRA rank: 128 (~600 MB)"""
|
745 |
-
|
746 |
-
elif model_type == "ltx_video":
|
747 |
-
return """### LTX-Video (LoRA)
|
748 |
-
- Required VRAM: ~18GB minimum
|
749 |
-
- Recommended batch size: 1-4
|
750 |
-
- Typical training time: 1-3 hours
|
751 |
-
- Default resolution: 49x512x768
|
752 |
-
- Default LoRA rank: 128"""
|
753 |
-
|
754 |
-
return ""
|
755 |
-
|
756 |
-
def get_default_params(self, model_type: str) -> Dict[str, Any]:
|
757 |
-
"""Get default training parameters for model type"""
|
758 |
-
if model_type == "hunyuan_video":
|
759 |
-
return {
|
760 |
-
"num_epochs": 70,
|
761 |
-
"batch_size": 1,
|
762 |
-
"learning_rate": 2e-5,
|
763 |
-
"save_iterations": 500,
|
764 |
-
"video_resolution_buckets": SMALL_TRAINING_BUCKETS,
|
765 |
-
"video_reshape_mode": "center",
|
766 |
-
"caption_dropout_p": 0.05,
|
767 |
-
"gradient_accumulation_steps": 1,
|
768 |
-
"rank": 128,
|
769 |
-
"lora_alpha": 128
|
770 |
-
}
|
771 |
-
else: # ltx_video
|
772 |
-
return {
|
773 |
-
"num_epochs": 70,
|
774 |
-
"batch_size": 1,
|
775 |
-
"learning_rate": 3e-5,
|
776 |
-
"save_iterations": 500,
|
777 |
-
"video_resolution_buckets": SMALL_TRAINING_BUCKETS,
|
778 |
-
"video_reshape_mode": "center",
|
779 |
-
"caption_dropout_p": 0.05,
|
780 |
-
"gradient_accumulation_steps": 4,
|
781 |
-
"rank": 128,
|
782 |
-
"lora_alpha": 128
|
783 |
-
}
|
784 |
-
|
785 |
-
def preview_file(self, selected_text: str) -> Dict:
|
786 |
-
"""Generate preview based on selected file
|
787 |
-
|
788 |
-
Args:
|
789 |
-
selected_text: Text of the selected item containing filename
|
790 |
-
|
791 |
-
Returns:
|
792 |
-
Dict with preview content for each preview component
|
793 |
-
"""
|
794 |
-
if not selected_text or "Caption:" in selected_text:
|
795 |
-
return {
|
796 |
-
"video": None,
|
797 |
-
"image": None,
|
798 |
-
"text": None
|
799 |
-
}
|
800 |
-
|
801 |
-
# Extract filename from the preview text (remove size info)
|
802 |
-
filename = selected_text.split(" (")[0].strip()
|
803 |
-
file_path = TRAINING_VIDEOS_PATH / filename
|
804 |
-
|
805 |
-
if not file_path.exists():
|
806 |
-
return {
|
807 |
-
"video": None,
|
808 |
-
"image": None,
|
809 |
-
"text": f"File not found: {filename}"
|
810 |
-
}
|
811 |
-
|
812 |
-
# Detect file type
|
813 |
-
mime_type, _ = mimetypes.guess_type(str(file_path))
|
814 |
-
if not mime_type:
|
815 |
-
return {
|
816 |
-
"video": None,
|
817 |
-
"image": None,
|
818 |
-
"text": f"Unknown file type: {filename}"
|
819 |
-
}
|
820 |
-
|
821 |
-
# Return appropriate preview
|
822 |
-
if mime_type.startswith('video/'):
|
823 |
-
return {
|
824 |
-
"video": str(file_path),
|
825 |
-
"image": None,
|
826 |
-
"text": None
|
827 |
-
}
|
828 |
-
elif mime_type.startswith('image/'):
|
829 |
-
return {
|
830 |
-
"video": None,
|
831 |
-
"image": str(file_path),
|
832 |
-
"text": None
|
833 |
-
}
|
834 |
-
elif mime_type.startswith('text/'):
|
835 |
-
try:
|
836 |
-
text_content = file_path.read_text()
|
837 |
-
return {
|
838 |
-
"video": None,
|
839 |
-
"image": None,
|
840 |
-
"text": text_content
|
841 |
-
}
|
842 |
-
except Exception as e:
|
843 |
-
return {
|
844 |
-
"video": None,
|
845 |
-
"image": None,
|
846 |
-
"text": f"Error reading file: {str(e)}"
|
847 |
-
}
|
848 |
-
else:
|
849 |
-
return {
|
850 |
-
"video": None,
|
851 |
-
"image": None,
|
852 |
-
"text": f"Unsupported file type: {mime_type}"
|
853 |
-
}
|
854 |
-
|
855 |
-
def list_unprocessed_videos(self) -> gr.Dataframe:
|
856 |
-
"""Update list of unprocessed videos"""
|
857 |
-
videos = self.splitter.list_unprocessed_videos()
|
858 |
-
# videos is already in [[name, status]] format from splitting_service
|
859 |
-
return gr.Dataframe(
|
860 |
-
headers=["name", "status"],
|
861 |
-
value=videos,
|
862 |
-
interactive=False
|
863 |
-
)
|
864 |
-
|
865 |
-
async def start_scene_detection(self, enable_splitting: bool) -> str:
|
866 |
-
"""Start background scene detection process
|
867 |
-
|
868 |
-
Args:
|
869 |
-
enable_splitting: Whether to split videos into scenes
|
870 |
-
"""
|
871 |
-
if self.splitter.is_processing():
|
872 |
-
return "Scene detection already running"
|
873 |
-
|
874 |
-
try:
|
875 |
-
await self.splitter.start_processing(enable_splitting)
|
876 |
-
return "Scene detection completed"
|
877 |
-
except Exception as e:
|
878 |
-
return f"Error during scene detection: {str(e)}"
|
879 |
-
|
880 |
-
|
881 |
-
def get_latest_status_message_and_logs(self) -> Tuple[str, str, str]:
|
882 |
-
state = self.trainer.get_status()
|
883 |
-
logs = self.trainer.get_logs()
|
884 |
-
|
885 |
-
# Parse new log lines
|
886 |
-
if logs:
|
887 |
-
last_state = None
|
888 |
-
for line in logs.splitlines():
|
889 |
-
state_update = self.log_parser.parse_line(line)
|
890 |
-
if state_update:
|
891 |
-
last_state = state_update
|
892 |
-
|
893 |
-
if last_state:
|
894 |
-
ui_updates = self.update_training_ui(last_state)
|
895 |
-
state["message"] = ui_updates.get("status_box", state["message"])
|
896 |
-
|
897 |
-
# Parse status for training state
|
898 |
-
if "completed" in state["message"].lower():
|
899 |
-
state["status"] = "completed"
|
900 |
-
|
901 |
-
return (state["status"], state["message"], logs)
|
902 |
-
|
903 |
-
def get_latest_status_message_logs_and_button_labels(self) -> Tuple[str, str, Any, Any, Any]:
|
904 |
-
status, message, logs = self.get_latest_status_message_and_logs()
|
905 |
-
return (
|
906 |
-
message,
|
907 |
-
logs,
|
908 |
-
*self.update_training_buttons(status).values()
|
909 |
-
)
|
910 |
-
|
911 |
-
def get_latest_button_labels(self) -> Tuple[Any, Any, Any]:
|
912 |
-
status, message, logs = self.get_latest_status_message_and_logs()
|
913 |
-
return self.update_training_buttons(status).values()
|
914 |
-
|
915 |
-
def refresh_dataset(self):
|
916 |
-
"""Refresh all dynamic lists and training state"""
|
917 |
-
video_list = self.splitter.list_unprocessed_videos()
|
918 |
-
training_dataset = self.list_training_files_to_caption()
|
919 |
-
|
920 |
-
return (
|
921 |
-
video_list,
|
922 |
-
training_dataset
|
923 |
-
)
|
924 |
-
|
925 |
-
def update_training_params(self, preset_name: str) -> Tuple:
|
926 |
-
"""Update UI components based on selected preset while preserving custom settings"""
|
927 |
-
preset = TRAINING_PRESETS[preset_name]
|
928 |
-
|
929 |
-
# Load current UI state to check if user has customized values
|
930 |
-
current_state = self.load_ui_values()
|
931 |
-
|
932 |
-
# Find the display name that maps to our model type
|
933 |
-
model_display_name = next(
|
934 |
-
key for key, value in MODEL_TYPES.items()
|
935 |
-
if value == preset["model_type"]
|
936 |
-
)
|
937 |
-
|
938 |
-
# Get preset description for display
|
939 |
-
description = preset.get("description", "")
|
940 |
-
|
941 |
-
# Get max values from buckets
|
942 |
-
buckets = preset["training_buckets"]
|
943 |
-
max_frames = max(frames for frames, _, _ in buckets)
|
944 |
-
max_height = max(height for _, height, _ in buckets)
|
945 |
-
max_width = max(width for _, _, width in buckets)
|
946 |
-
bucket_info = f"\nMaximum video size: {max_frames} frames at {max_width}x{max_height} resolution"
|
947 |
-
|
948 |
-
info_text = f"{description}{bucket_info}"
|
949 |
-
|
950 |
-
# Return values in the same order as the output components
|
951 |
-
# Use preset defaults but preserve user-modified values if they exist
|
952 |
-
lora_rank_val = current_state.get("lora_rank") if current_state.get("lora_rank") != preset.get("lora_rank", "128") else preset["lora_rank"]
|
953 |
-
lora_alpha_val = current_state.get("lora_alpha") if current_state.get("lora_alpha") != preset.get("lora_alpha", "128") else preset["lora_alpha"]
|
954 |
-
num_epochs_val = current_state.get("num_epochs") if current_state.get("num_epochs") != preset.get("num_epochs", 70) else preset["num_epochs"]
|
955 |
-
batch_size_val = current_state.get("batch_size") if current_state.get("batch_size") != preset.get("batch_size", 1) else preset["batch_size"]
|
956 |
-
learning_rate_val = current_state.get("learning_rate") if current_state.get("learning_rate") != preset.get("learning_rate", 3e-5) else preset["learning_rate"]
|
957 |
-
save_iterations_val = current_state.get("save_iterations") if current_state.get("save_iterations") != preset.get("save_iterations", 500) else preset["save_iterations"]
|
958 |
-
|
959 |
-
return (
|
960 |
-
model_display_name,
|
961 |
-
lora_rank_val,
|
962 |
-
lora_alpha_val,
|
963 |
-
num_epochs_val,
|
964 |
-
batch_size_val,
|
965 |
-
learning_rate_val,
|
966 |
-
save_iterations_val,
|
967 |
-
info_text
|
968 |
-
)
|
969 |
-
|
970 |
-
def create_ui(self):
|
971 |
-
"""Create Gradio interface"""
|
972 |
-
|
973 |
-
with gr.Blocks(title="🎥 Video Model Studio") as app:
|
974 |
-
gr.Markdown("# 🎥 Video Model Studio")
|
975 |
-
|
976 |
-
with gr.Tabs() as tabs:
|
977 |
-
with gr.TabItem("1️⃣ Import", id="import_tab"):
|
978 |
-
|
979 |
-
with gr.Row():
|
980 |
-
gr.Markdown("## Automatic splitting and captioning")
|
981 |
-
|
982 |
-
with gr.Row():
|
983 |
-
enable_automatic_video_split = gr.Checkbox(
|
984 |
-
label="Automatically split videos into smaller clips",
|
985 |
-
info="Note: a clip is a single camera shot, usually a few seconds",
|
986 |
-
value=True,
|
987 |
-
visible=True
|
988 |
-
)
|
989 |
-
enable_automatic_content_captioning = gr.Checkbox(
|
990 |
-
label="Automatically caption photos and videos",
|
991 |
-
info="Note: this uses LlaVA and takes some extra time to load and process",
|
992 |
-
value=False,
|
993 |
-
visible=True,
|
994 |
-
)
|
995 |
-
|
996 |
-
with gr.Row():
|
997 |
-
with gr.Column(scale=3):
|
998 |
-
with gr.Row():
|
999 |
-
with gr.Column():
|
1000 |
-
gr.Markdown("## Import video files")
|
1001 |
-
gr.Markdown("You can upload either:")
|
1002 |
-
gr.Markdown("- A single MP4 video file")
|
1003 |
-
gr.Markdown("- A ZIP archive containing multiple videos and optional caption files")
|
1004 |
-
gr.Markdown("For ZIP files: Create a folder containing videos (name is not important) and optional caption files with the same name (eg. `some_video.txt` for `some_video.mp4`)")
|
1005 |
-
|
1006 |
-
with gr.Row():
|
1007 |
-
files = gr.Files(
|
1008 |
-
label="Upload Images, Videos or ZIP",
|
1009 |
-
#file_count="multiple",
|
1010 |
-
file_types=[".jpg", ".jpeg", ".png", ".webp", ".webp", ".avif", ".heic", ".mp4", ".zip"],
|
1011 |
-
type="filepath"
|
1012 |
-
)
|
1013 |
-
|
1014 |
-
with gr.Column(scale=3):
|
1015 |
-
with gr.Row():
|
1016 |
-
with gr.Column():
|
1017 |
-
gr.Markdown("## Import a YouTube video")
|
1018 |
-
gr.Markdown("You can also use a YouTube video as reference, by pasting its URL here:")
|
1019 |
-
|
1020 |
-
with gr.Row():
|
1021 |
-
youtube_url = gr.Textbox(
|
1022 |
-
label="Import YouTube Video",
|
1023 |
-
placeholder="https://www.youtube.com/watch?v=..."
|
1024 |
-
)
|
1025 |
-
with gr.Row():
|
1026 |
-
youtube_download_btn = gr.Button("Download YouTube Video", variant="secondary")
|
1027 |
-
with gr.Row():
|
1028 |
-
import_status = gr.Textbox(label="Status", interactive=False)
|
1029 |
-
|
1030 |
-
|
1031 |
-
with gr.TabItem("2️⃣ Split", id="split_tab"):
|
1032 |
-
with gr.Row():
|
1033 |
-
split_title = gr.Markdown("## Splitting of 0 videos (0 bytes)")
|
1034 |
-
|
1035 |
-
with gr.Row():
|
1036 |
-
with gr.Column():
|
1037 |
-
detect_btn = gr.Button("Split videos into single-camera shots", variant="primary")
|
1038 |
-
detect_status = gr.Textbox(label="Status", interactive=False)
|
1039 |
-
|
1040 |
-
with gr.Column():
|
1041 |
-
|
1042 |
-
video_list = gr.Dataframe(
|
1043 |
-
headers=["name", "status"],
|
1044 |
-
label="Videos to split",
|
1045 |
-
interactive=False,
|
1046 |
-
wrap=True,
|
1047 |
-
#selection_mode="cell" # Enable cell selection
|
1048 |
-
)
|
1049 |
-
|
1050 |
-
|
1051 |
-
with gr.TabItem("3️⃣ Caption"):
|
1052 |
-
with gr.Row():
|
1053 |
-
caption_title = gr.Markdown("## Captioning of 0 files (0 bytes)")
|
1054 |
-
|
1055 |
-
with gr.Row():
|
1056 |
-
|
1057 |
-
with gr.Column():
|
1058 |
-
with gr.Row():
|
1059 |
-
custom_prompt_prefix = gr.Textbox(
|
1060 |
-
scale=3,
|
1061 |
-
label='Prefix to add to ALL captions (eg. "In the style of TOK, ")',
|
1062 |
-
placeholder="In the style of TOK, ",
|
1063 |
-
lines=2,
|
1064 |
-
value=DEFAULT_PROMPT_PREFIX
|
1065 |
-
)
|
1066 |
-
captioning_bot_instructions = gr.Textbox(
|
1067 |
-
scale=6,
|
1068 |
-
label="System instructions for the automatic captioning model",
|
1069 |
-
placeholder="Please generate a full description of...",
|
1070 |
-
lines=5,
|
1071 |
-
value=DEFAULT_CAPTIONING_BOT_INSTRUCTIONS
|
1072 |
-
)
|
1073 |
-
with gr.Row():
|
1074 |
-
run_autocaption_btn = gr.Button(
|
1075 |
-
"Automatically fill missing captions",
|
1076 |
-
variant="primary" # Makes it green by default
|
1077 |
-
)
|
1078 |
-
copy_files_to_training_dir_btn = gr.Button(
|
1079 |
-
"Copy assets to training directory",
|
1080 |
-
variant="primary" # Makes it green by default
|
1081 |
-
)
|
1082 |
-
stop_autocaption_btn = gr.Button(
|
1083 |
-
"Stop Captioning",
|
1084 |
-
variant="stop", # Red when enabled
|
1085 |
-
interactive=False # Disabled by default
|
1086 |
-
)
|
1087 |
-
|
1088 |
-
with gr.Row():
|
1089 |
-
with gr.Column():
|
1090 |
-
training_dataset = gr.Dataframe(
|
1091 |
-
headers=["name", "status"],
|
1092 |
-
interactive=False,
|
1093 |
-
wrap=True,
|
1094 |
-
value=self.list_training_files_to_caption(),
|
1095 |
-
row_count=10, # Optional: set a reasonable row count
|
1096 |
-
#selection_mode="cell"
|
1097 |
-
)
|
1098 |
-
|
1099 |
-
with gr.Column():
|
1100 |
-
preview_video = gr.Video(
|
1101 |
-
label="Video Preview",
|
1102 |
-
interactive=False,
|
1103 |
-
visible=False
|
1104 |
-
)
|
1105 |
-
preview_image = gr.Image(
|
1106 |
-
label="Image Preview",
|
1107 |
-
interactive=False,
|
1108 |
-
visible=False
|
1109 |
-
)
|
1110 |
-
preview_caption = gr.Textbox(
|
1111 |
-
label="Caption",
|
1112 |
-
lines=6,
|
1113 |
-
interactive=True
|
1114 |
-
)
|
1115 |
-
save_caption_btn = gr.Button("Save Caption")
|
1116 |
-
preview_status = gr.Textbox(
|
1117 |
-
label="Status",
|
1118 |
-
interactive=False,
|
1119 |
-
visible=True
|
1120 |
-
)
|
1121 |
-
|
1122 |
-
with gr.TabItem("4️⃣ Train"):
|
1123 |
-
with gr.Row():
|
1124 |
-
with gr.Column():
|
1125 |
-
|
1126 |
-
with gr.Row():
|
1127 |
-
train_title = gr.Markdown("## 0 files available for training (0 bytes)")
|
1128 |
-
|
1129 |
-
with gr.Row():
|
1130 |
-
with gr.Column():
|
1131 |
-
training_preset = gr.Dropdown(
|
1132 |
-
choices=list(TRAINING_PRESETS.keys()),
|
1133 |
-
label="Training Preset",
|
1134 |
-
value=list(TRAINING_PRESETS.keys())[0]
|
1135 |
-
)
|
1136 |
-
preset_info = gr.Markdown()
|
1137 |
-
|
1138 |
-
with gr.Row():
|
1139 |
-
with gr.Column():
|
1140 |
-
model_type = gr.Dropdown(
|
1141 |
-
choices=list(MODEL_TYPES.keys()),
|
1142 |
-
label="Model Type",
|
1143 |
-
value=list(MODEL_TYPES.keys())[0]
|
1144 |
-
)
|
1145 |
-
model_info = gr.Markdown(
|
1146 |
-
value=self.get_model_info(list(MODEL_TYPES.keys())[0])
|
1147 |
-
)
|
1148 |
-
|
1149 |
-
with gr.Row():
|
1150 |
-
lora_rank = gr.Dropdown(
|
1151 |
-
label="LoRA Rank",
|
1152 |
-
choices=["16", "32", "64", "128", "256", "512", "1024"],
|
1153 |
-
value="128",
|
1154 |
-
type="value"
|
1155 |
-
)
|
1156 |
-
lora_alpha = gr.Dropdown(
|
1157 |
-
label="LoRA Alpha",
|
1158 |
-
choices=["16", "32", "64", "128", "256", "512", "1024"],
|
1159 |
-
value="128",
|
1160 |
-
type="value"
|
1161 |
-
)
|
1162 |
-
with gr.Row():
|
1163 |
-
num_epochs = gr.Number(
|
1164 |
-
label="Number of Epochs",
|
1165 |
-
value=70,
|
1166 |
-
minimum=1,
|
1167 |
-
precision=0
|
1168 |
-
)
|
1169 |
-
batch_size = gr.Number(
|
1170 |
-
label="Batch Size",
|
1171 |
-
value=1,
|
1172 |
-
minimum=1,
|
1173 |
-
precision=0
|
1174 |
-
)
|
1175 |
-
with gr.Row():
|
1176 |
-
learning_rate = gr.Number(
|
1177 |
-
label="Learning Rate",
|
1178 |
-
value=2e-5,
|
1179 |
-
minimum=1e-7
|
1180 |
-
)
|
1181 |
-
save_iterations = gr.Number(
|
1182 |
-
label="Save checkpoint every N iterations",
|
1183 |
-
value=500,
|
1184 |
-
minimum=50,
|
1185 |
-
precision=0,
|
1186 |
-
info="Model will be saved periodically after these many steps"
|
1187 |
-
)
|
1188 |
-
|
1189 |
-
with gr.Column():
|
1190 |
-
with gr.Row():
|
1191 |
-
start_btn = gr.Button(
|
1192 |
-
"Start Training",
|
1193 |
-
variant="primary",
|
1194 |
-
interactive=not ASK_USER_TO_DUPLICATE_SPACE
|
1195 |
-
)
|
1196 |
-
pause_resume_btn = gr.Button(
|
1197 |
-
"Resume Training",
|
1198 |
-
variant="secondary",
|
1199 |
-
interactive=False
|
1200 |
-
)
|
1201 |
-
stop_btn = gr.Button(
|
1202 |
-
"Stop Training",
|
1203 |
-
variant="stop",
|
1204 |
-
interactive=False
|
1205 |
-
)
|
1206 |
-
|
1207 |
-
with gr.Row():
|
1208 |
-
with gr.Column():
|
1209 |
-
status_box = gr.Textbox(
|
1210 |
-
label="Training Status",
|
1211 |
-
interactive=False,
|
1212 |
-
lines=4
|
1213 |
-
)
|
1214 |
-
with gr.Accordion("See training logs"):
|
1215 |
-
log_box = gr.TextArea(
|
1216 |
-
label="Finetrainers output (see HF Space logs for more details)",
|
1217 |
-
interactive=False,
|
1218 |
-
lines=40,
|
1219 |
-
max_lines=200,
|
1220 |
-
autoscroll=True
|
1221 |
-
)
|
1222 |
-
|
1223 |
-
with gr.TabItem("5️⃣ Manage"):
|
1224 |
-
|
1225 |
-
with gr.Column():
|
1226 |
-
with gr.Row():
|
1227 |
-
with gr.Column():
|
1228 |
-
gr.Markdown("## Publishing")
|
1229 |
-
gr.Markdown("You model can be pushed to Hugging Face (this will use HF_API_TOKEN)")
|
1230 |
-
|
1231 |
-
with gr.Row():
|
1232 |
-
|
1233 |
-
with gr.Column():
|
1234 |
-
repo_id = gr.Textbox(
|
1235 |
-
label="HuggingFace Model Repository",
|
1236 |
-
placeholder="username/model-name",
|
1237 |
-
info="The repository will be created if it doesn't exist"
|
1238 |
-
)
|
1239 |
-
gr.Checkbox(label="Check this to make your model public (ie. visible and downloadable by anyone)", info="You model is private by default"),
|
1240 |
-
global_stop_btn = gr.Button(
|
1241 |
-
"Push my model",
|
1242 |
-
#variant="stop"
|
1243 |
-
)
|
1244 |
-
|
1245 |
-
|
1246 |
-
with gr.Row():
|
1247 |
-
with gr.Column():
|
1248 |
-
with gr.Row():
|
1249 |
-
with gr.Column():
|
1250 |
-
gr.Markdown("## Storage management")
|
1251 |
-
with gr.Row():
|
1252 |
-
download_dataset_btn = gr.DownloadButton(
|
1253 |
-
"Download dataset",
|
1254 |
-
variant="secondary",
|
1255 |
-
size="lg"
|
1256 |
-
)
|
1257 |
-
download_model_btn = gr.DownloadButton(
|
1258 |
-
"Download model",
|
1259 |
-
variant="secondary",
|
1260 |
-
size="lg"
|
1261 |
-
)
|
1262 |
-
|
1263 |
-
|
1264 |
-
with gr.Row():
|
1265 |
-
global_stop_btn = gr.Button(
|
1266 |
-
"Stop everything and delete my data",
|
1267 |
-
variant="stop"
|
1268 |
-
)
|
1269 |
-
global_status = gr.Textbox(
|
1270 |
-
label="Global Status",
|
1271 |
-
interactive=False,
|
1272 |
-
visible=False
|
1273 |
-
)
|
1274 |
-
|
1275 |
-
|
1276 |
-
|
1277 |
-
# Event handlers
|
1278 |
-
def update_model_info(model):
|
1279 |
-
params = self.get_default_params(MODEL_TYPES[model])
|
1280 |
-
info = self.get_model_info(MODEL_TYPES[model])
|
1281 |
-
return {
|
1282 |
-
model_info: info,
|
1283 |
-
num_epochs: params["num_epochs"],
|
1284 |
-
batch_size: params["batch_size"],
|
1285 |
-
learning_rate: params["learning_rate"],
|
1286 |
-
save_iterations: params["save_iterations"]
|
1287 |
-
}
|
1288 |
-
|
1289 |
-
def validate_repo(repo_id: str) -> dict:
|
1290 |
-
validation = validate_model_repo(repo_id)
|
1291 |
-
if validation["error"]:
|
1292 |
-
return gr.update(value=repo_id, error=validation["error"])
|
1293 |
-
return gr.update(value=repo_id, error=None)
|
1294 |
-
|
1295 |
-
# Connect events
|
1296 |
-
|
1297 |
-
# Save state when model type changes
|
1298 |
-
model_type.change(
|
1299 |
-
fn=lambda v: self.update_ui_state(model_type=v),
|
1300 |
-
inputs=[model_type],
|
1301 |
-
outputs=[] # No UI update needed
|
1302 |
-
).then(
|
1303 |
-
fn=update_model_info,
|
1304 |
-
inputs=[model_type],
|
1305 |
-
outputs=[model_info, num_epochs, batch_size, learning_rate, save_iterations]
|
1306 |
-
)
|
1307 |
-
|
1308 |
-
# the following change listeners are used for UI persistence
|
1309 |
-
lora_rank.change(
|
1310 |
-
fn=lambda v: self.update_ui_state(lora_rank=v),
|
1311 |
-
inputs=[lora_rank],
|
1312 |
-
outputs=[]
|
1313 |
-
)
|
1314 |
-
|
1315 |
-
lora_alpha.change(
|
1316 |
-
fn=lambda v: self.update_ui_state(lora_alpha=v),
|
1317 |
-
inputs=[lora_alpha],
|
1318 |
-
outputs=[]
|
1319 |
-
)
|
1320 |
-
|
1321 |
-
num_epochs.change(
|
1322 |
-
fn=lambda v: self.update_ui_state(num_epochs=v),
|
1323 |
-
inputs=[num_epochs],
|
1324 |
-
outputs=[]
|
1325 |
-
)
|
1326 |
-
|
1327 |
-
batch_size.change(
|
1328 |
-
fn=lambda v: self.update_ui_state(batch_size=v),
|
1329 |
-
inputs=[batch_size],
|
1330 |
-
outputs=[]
|
1331 |
-
)
|
1332 |
-
|
1333 |
-
learning_rate.change(
|
1334 |
-
fn=lambda v: self.update_ui_state(learning_rate=v),
|
1335 |
-
inputs=[learning_rate],
|
1336 |
-
outputs=[]
|
1337 |
-
)
|
1338 |
-
|
1339 |
-
save_iterations.change(
|
1340 |
-
fn=lambda v: self.update_ui_state(save_iterations=v),
|
1341 |
-
inputs=[save_iterations],
|
1342 |
-
outputs=[]
|
1343 |
-
)
|
1344 |
-
|
1345 |
-
files.upload(
|
1346 |
-
fn=lambda x: self.importer.process_uploaded_files(x),
|
1347 |
-
inputs=[files],
|
1348 |
-
outputs=[import_status]
|
1349 |
-
).success(
|
1350 |
-
fn=self.update_titles_after_import,
|
1351 |
-
inputs=[enable_automatic_video_split, enable_automatic_content_captioning, custom_prompt_prefix],
|
1352 |
-
outputs=[
|
1353 |
-
tabs, video_list, detect_status,
|
1354 |
-
split_title, caption_title, train_title
|
1355 |
-
]
|
1356 |
-
)
|
1357 |
-
|
1358 |
-
youtube_download_btn.click(
|
1359 |
-
fn=self.importer.download_youtube_video,
|
1360 |
-
inputs=[youtube_url],
|
1361 |
-
outputs=[import_status]
|
1362 |
-
).success(
|
1363 |
-
fn=self.on_import_success,
|
1364 |
-
inputs=[enable_automatic_video_split, enable_automatic_content_captioning, custom_prompt_prefix],
|
1365 |
-
outputs=[tabs, video_list, detect_status]
|
1366 |
-
)
|
1367 |
-
|
1368 |
-
# Scene detection events
|
1369 |
-
detect_btn.click(
|
1370 |
-
fn=self.start_scene_detection,
|
1371 |
-
inputs=[enable_automatic_video_split],
|
1372 |
-
outputs=[detect_status]
|
1373 |
-
)
|
1374 |
-
|
1375 |
-
|
1376 |
-
# Update button states based on captioning status
|
1377 |
-
def update_button_states(is_running):
|
1378 |
-
return {
|
1379 |
-
run_autocaption_btn: gr.Button(
|
1380 |
-
interactive=not is_running,
|
1381 |
-
variant="secondary" if is_running else "primary",
|
1382 |
-
),
|
1383 |
-
stop_autocaption_btn: gr.Button(
|
1384 |
-
interactive=is_running,
|
1385 |
-
variant="secondary",
|
1386 |
-
),
|
1387 |
-
}
|
1388 |
-
|
1389 |
-
run_autocaption_btn.click(
|
1390 |
-
fn=self.show_refreshing_status,
|
1391 |
-
outputs=[training_dataset]
|
1392 |
-
).then(
|
1393 |
-
fn=lambda: self.update_captioning_buttons_start(),
|
1394 |
-
outputs=[run_autocaption_btn, stop_autocaption_btn, copy_files_to_training_dir_btn]
|
1395 |
-
).then(
|
1396 |
-
fn=self.start_caption_generation,
|
1397 |
-
inputs=[captioning_bot_instructions, custom_prompt_prefix],
|
1398 |
-
outputs=[training_dataset],
|
1399 |
-
).then(
|
1400 |
-
fn=lambda: self.update_captioning_buttons_end(),
|
1401 |
-
outputs=[run_autocaption_btn, stop_autocaption_btn, copy_files_to_training_dir_btn]
|
1402 |
-
)
|
1403 |
-
|
1404 |
-
copy_files_to_training_dir_btn.click(
|
1405 |
-
fn=self.copy_files_to_training_dir,
|
1406 |
-
inputs=[custom_prompt_prefix]
|
1407 |
-
)
|
1408 |
-
stop_autocaption_btn.click(
|
1409 |
-
fn=self.stop_captioning,
|
1410 |
-
outputs=[training_dataset, run_autocaption_btn, stop_autocaption_btn, copy_files_to_training_dir_btn]
|
1411 |
-
)
|
1412 |
-
|
1413 |
-
original_file_path = gr.State(value=None)
|
1414 |
-
training_dataset.select(
|
1415 |
-
fn=self.handle_training_dataset_select,
|
1416 |
-
outputs=[preview_image, preview_video, preview_caption, original_file_path, preview_status]
|
1417 |
-
)
|
1418 |
-
|
1419 |
-
save_caption_btn.click(
|
1420 |
-
fn=self.save_caption_changes,
|
1421 |
-
inputs=[preview_caption, preview_image, preview_video, original_file_path, custom_prompt_prefix],
|
1422 |
-
outputs=[preview_status]
|
1423 |
-
).success(
|
1424 |
-
fn=self.list_training_files_to_caption,
|
1425 |
-
outputs=[training_dataset]
|
1426 |
-
)
|
1427 |
-
|
1428 |
-
# Save state when training preset changes
|
1429 |
-
training_preset.change(
|
1430 |
-
fn=lambda v: self.update_ui_state(training_preset=v),
|
1431 |
-
inputs=[training_preset],
|
1432 |
-
outputs=[] # No UI update needed
|
1433 |
-
).then(
|
1434 |
-
fn=self.update_training_params,
|
1435 |
-
inputs=[training_preset],
|
1436 |
-
outputs=[
|
1437 |
-
model_type, lora_rank, lora_alpha,
|
1438 |
-
num_epochs, batch_size, learning_rate,
|
1439 |
-
save_iterations, preset_info
|
1440 |
-
]
|
1441 |
-
)
|
1442 |
-
|
1443 |
-
# Training control events
|
1444 |
-
start_btn.click(
|
1445 |
-
fn=lambda preset, model_type, *args: (
|
1446 |
-
self.log_parser.reset(),
|
1447 |
-
self.trainer.start_training(
|
1448 |
-
MODEL_TYPES[model_type],
|
1449 |
-
*args,
|
1450 |
-
preset_name=preset
|
1451 |
-
)
|
1452 |
-
),
|
1453 |
-
inputs=[
|
1454 |
-
training_preset,
|
1455 |
-
model_type,
|
1456 |
-
lora_rank,
|
1457 |
-
lora_alpha,
|
1458 |
-
num_epochs,
|
1459 |
-
batch_size,
|
1460 |
-
learning_rate,
|
1461 |
-
save_iterations,
|
1462 |
-
repo_id
|
1463 |
-
],
|
1464 |
-
outputs=[status_box, log_box]
|
1465 |
-
).success(
|
1466 |
-
fn=self.get_latest_status_message_logs_and_button_labels,
|
1467 |
-
outputs=[status_box, log_box, start_btn, stop_btn, pause_resume_btn]
|
1468 |
-
)
|
1469 |
-
|
1470 |
-
pause_resume_btn.click(
|
1471 |
-
fn=self.handle_pause_resume,
|
1472 |
-
outputs=[status_box, log_box, start_btn, stop_btn, pause_resume_btn]
|
1473 |
-
)
|
1474 |
-
|
1475 |
-
stop_btn.click(
|
1476 |
-
fn=self.handle_stop,
|
1477 |
-
outputs=[status_box, log_box, start_btn, stop_btn, pause_resume_btn]
|
1478 |
-
)
|
1479 |
-
|
1480 |
-
def handle_global_stop():
|
1481 |
-
result = self.stop_all_and_clear()
|
1482 |
-
# Update all relevant UI components
|
1483 |
-
status = result["status"]
|
1484 |
-
details = "\n".join(f"{k}: {v}" for k, v in result["details"].items())
|
1485 |
-
full_status = f"{status}\n\nDetails:\n{details}"
|
1486 |
-
|
1487 |
-
# Get fresh lists after cleanup
|
1488 |
-
videos = self.splitter.list_unprocessed_videos()
|
1489 |
-
clips = self.list_training_files_to_caption()
|
1490 |
-
|
1491 |
-
return {
|
1492 |
-
global_status: gr.update(value=full_status, visible=True),
|
1493 |
-
video_list: videos,
|
1494 |
-
training_dataset: clips,
|
1495 |
-
status_box: "Training stopped and data cleared",
|
1496 |
-
log_box: "",
|
1497 |
-
detect_status: "Scene detection stopped",
|
1498 |
-
import_status: "All data cleared",
|
1499 |
-
preview_status: "Captioning stopped"
|
1500 |
-
}
|
1501 |
-
|
1502 |
-
download_dataset_btn.click(
|
1503 |
-
fn=self.trainer.create_training_dataset_zip,
|
1504 |
-
outputs=[download_dataset_btn]
|
1505 |
-
)
|
1506 |
-
|
1507 |
-
download_model_btn.click(
|
1508 |
-
fn=self.trainer.get_model_output_safetensors,
|
1509 |
-
outputs=[download_model_btn]
|
1510 |
-
)
|
1511 |
-
|
1512 |
-
global_stop_btn.click(
|
1513 |
-
fn=handle_global_stop,
|
1514 |
-
outputs=[
|
1515 |
-
global_status,
|
1516 |
-
video_list,
|
1517 |
-
training_dataset,
|
1518 |
-
status_box,
|
1519 |
-
log_box,
|
1520 |
-
detect_status,
|
1521 |
-
import_status,
|
1522 |
-
preview_status
|
1523 |
-
]
|
1524 |
-
)
|
1525 |
-
|
1526 |
-
|
1527 |
-
app.load(
|
1528 |
-
fn=self.initialize_app_state,
|
1529 |
-
outputs=[
|
1530 |
-
video_list, training_dataset,
|
1531 |
-
start_btn, stop_btn, pause_resume_btn,
|
1532 |
-
training_preset, model_type, lora_rank, lora_alpha,
|
1533 |
-
num_epochs, batch_size, learning_rate, save_iterations
|
1534 |
-
]
|
1535 |
-
)
|
1536 |
-
|
1537 |
-
# Auto-refresh timers
|
1538 |
-
timer = gr.Timer(value=1)
|
1539 |
-
timer.tick(
|
1540 |
-
fn=lambda: (
|
1541 |
-
self.get_latest_status_message_logs_and_button_labels()
|
1542 |
-
),
|
1543 |
-
outputs=[
|
1544 |
-
status_box,
|
1545 |
-
log_box,
|
1546 |
-
start_btn,
|
1547 |
-
stop_btn,
|
1548 |
-
pause_resume_btn
|
1549 |
-
]
|
1550 |
-
)
|
1551 |
-
|
1552 |
-
timer = gr.Timer(value=5)
|
1553 |
-
timer.tick(
|
1554 |
-
fn=lambda: (
|
1555 |
-
self.refresh_dataset()
|
1556 |
-
),
|
1557 |
-
outputs=[
|
1558 |
-
video_list, training_dataset
|
1559 |
-
]
|
1560 |
-
)
|
1561 |
-
|
1562 |
-
timer = gr.Timer(value=6)
|
1563 |
-
timer.tick(
|
1564 |
-
fn=lambda: self.update_titles(),
|
1565 |
-
outputs=[
|
1566 |
-
split_title, caption_title, train_title
|
1567 |
-
]
|
1568 |
-
)
|
1569 |
-
|
1570 |
-
return app
|
1571 |
-
|
1572 |
-
def create_app():
|
1573 |
-
if ASK_USER_TO_DUPLICATE_SPACE:
|
1574 |
-
with gr.Blocks() as app:
|
1575 |
-
gr.Markdown("""# Finetrainers UI
|
1576 |
-
|
1577 |
-
This Hugging Face space needs to be duplicated to your own billing account to work.
|
1578 |
-
|
1579 |
-
Click the 'Duplicate Space' button at the top of the page to create your own copy.
|
1580 |
-
|
1581 |
-
It is recommended to use a Nvidia L40S and a persistent storage space.
|
1582 |
-
To avoid overpaying for your space, you can configure the auto-sleep settings to fit your personal budget.""")
|
1583 |
-
return app
|
1584 |
-
|
1585 |
-
ui = VideoTrainerUI()
|
1586 |
-
return ui.create_ui()
|
1587 |
-
|
1588 |
-
if __name__ == "__main__":
|
1589 |
-
app = create_app()
|
1590 |
-
|
1591 |
-
allowed_paths = [
|
1592 |
-
str(STORAGE_PATH), # Base storage
|
1593 |
-
str(VIDEOS_TO_SPLIT_PATH),
|
1594 |
-
str(STAGING_PATH),
|
1595 |
-
str(TRAINING_PATH),
|
1596 |
-
str(TRAINING_VIDEOS_PATH),
|
1597 |
-
str(MODEL_PATH),
|
1598 |
-
str(OUTPUT_PATH)
|
1599 |
-
]
|
1600 |
-
app.queue(default_concurrency_limit=1).launch(
|
1601 |
-
server_name="0.0.0.0",
|
1602 |
-
allowed_paths=allowed_paths
|
1603 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
vms/services/captioner.py
CHANGED
@@ -179,7 +179,7 @@ class CaptioningService:
|
|
179 |
)
|
180 |
self.model.eval()
|
181 |
|
182 |
-
def _load_video(self, video_path: Path, max_frames_num: int = 64, fps: int = 1, force_sample: bool = True) ->
|
183 |
"""Load and preprocess video frames with strict limits
|
184 |
|
185 |
Args:
|
@@ -224,7 +224,7 @@ class CaptioningService:
|
|
224 |
logger.error(f"Error loading video frames: {str(e)}")
|
225 |
raise
|
226 |
|
227 |
-
async def process_video(self, video_path: Path, prompt: str, prompt_prefix: str = "") -> AsyncGenerator[
|
228 |
try:
|
229 |
video_name = video_path.name
|
230 |
logger.info(f"Starting processing of video: {video_name}")
|
@@ -373,7 +373,7 @@ class CaptioningService:
|
|
373 |
yield progress, None
|
374 |
raise
|
375 |
|
376 |
-
async def process_image(self, image_path: Path, prompt: str, prompt_prefix: str = "") -> AsyncGenerator[
|
377 |
"""Process a single image for captioning"""
|
378 |
try:
|
379 |
image_name = image_path.name
|
|
|
179 |
)
|
180 |
self.model.eval()
|
181 |
|
182 |
+
def _load_video(self, video_path: Path, max_frames_num: int = 64, fps: int = 1, force_sample: bool = True) -> Tuple[np.ndarray, str, float]:
|
183 |
"""Load and preprocess video frames with strict limits
|
184 |
|
185 |
Args:
|
|
|
224 |
logger.error(f"Error loading video frames: {str(e)}")
|
225 |
raise
|
226 |
|
227 |
+
async def process_video(self, video_path: Path, prompt: str, prompt_prefix: str = "") -> AsyncGenerator[Tuple[CaptioningProgress, Optional[str]], None]:
|
228 |
try:
|
229 |
video_name = video_path.name
|
230 |
logger.info(f"Starting processing of video: {video_name}")
|
|
|
373 |
yield progress, None
|
374 |
raise
|
375 |
|
376 |
+
async def process_image(self, image_path: Path, prompt: str, prompt_prefix: str = "") -> AsyncGenerator[Tuple[CaptioningProgress, Optional[str]], None]:
|
377 |
"""Process a single image for captioning"""
|
378 |
try:
|
379 |
image_name = image_path.name
|
vms/tabs/caption_tab.py
CHANGED
@@ -4,11 +4,14 @@ Caption tab for Video Model Studio UI
|
|
4 |
|
5 |
import gradio as gr
|
6 |
import logging
|
7 |
-
|
|
|
|
|
8 |
from pathlib import Path
|
9 |
|
10 |
from .base_tab import BaseTab
|
11 |
-
from ..config import DEFAULT_CAPTIONING_BOT_INSTRUCTIONS, DEFAULT_PROMPT_PREFIX
|
|
|
12 |
|
13 |
logger = logging.getLogger(__name__)
|
14 |
|
@@ -19,6 +22,7 @@ class CaptionTab(BaseTab):
|
|
19 |
super().__init__(app_state)
|
20 |
self.id = "caption_tab"
|
21 |
self.title = "3️⃣ Caption"
|
|
|
22 |
|
23 |
def create(self, parent=None) -> gr.TabItem:
|
24 |
"""Create the Caption tab UI components"""
|
@@ -64,7 +68,7 @@ class CaptionTab(BaseTab):
|
|
64 |
headers=["name", "status"],
|
65 |
interactive=False,
|
66 |
wrap=True,
|
67 |
-
value=self.
|
68 |
row_count=10
|
69 |
)
|
70 |
|
@@ -98,24 +102,24 @@ class CaptionTab(BaseTab):
|
|
98 |
"""Connect event handlers to UI components"""
|
99 |
# Run auto-captioning button
|
100 |
self.components["run_autocaption_btn"].click(
|
101 |
-
fn=self.
|
102 |
outputs=[self.components["training_dataset"]]
|
103 |
).then(
|
104 |
-
fn=
|
105 |
outputs=[
|
106 |
self.components["run_autocaption_btn"],
|
107 |
self.components["stop_autocaption_btn"],
|
108 |
self.components["copy_files_to_training_dir_btn"]
|
109 |
]
|
110 |
).then(
|
111 |
-
fn=self.
|
112 |
inputs=[
|
113 |
self.components["captioning_bot_instructions"],
|
114 |
self.components["custom_prompt_prefix"]
|
115 |
],
|
116 |
outputs=[self.components["training_dataset"]],
|
117 |
).then(
|
118 |
-
fn=
|
119 |
outputs=[
|
120 |
self.components["run_autocaption_btn"],
|
121 |
self.components["stop_autocaption_btn"],
|
@@ -125,13 +129,13 @@ class CaptionTab(BaseTab):
|
|
125 |
|
126 |
# Copy files to training dir button
|
127 |
self.components["copy_files_to_training_dir_btn"].click(
|
128 |
-
fn=self.
|
129 |
inputs=[self.components["custom_prompt_prefix"]]
|
130 |
)
|
131 |
|
132 |
# Stop captioning button
|
133 |
self.components["stop_autocaption_btn"].click(
|
134 |
-
fn=self.
|
135 |
outputs=[
|
136 |
self.components["training_dataset"],
|
137 |
self.components["run_autocaption_btn"],
|
@@ -142,7 +146,7 @@ class CaptionTab(BaseTab):
|
|
142 |
|
143 |
# Dataset selection for preview
|
144 |
self.components["training_dataset"].select(
|
145 |
-
fn=self.
|
146 |
outputs=[
|
147 |
self.components["preview_image"],
|
148 |
self.components["preview_video"],
|
@@ -154,7 +158,7 @@ class CaptionTab(BaseTab):
|
|
154 |
|
155 |
# Save caption button
|
156 |
self.components["save_caption_btn"].click(
|
157 |
-
fn=self.
|
158 |
inputs=[
|
159 |
self.components["preview_caption"],
|
160 |
self.components["preview_image"],
|
@@ -164,13 +168,431 @@ class CaptionTab(BaseTab):
|
|
164 |
],
|
165 |
outputs=[self.components["preview_status"]]
|
166 |
).success(
|
167 |
-
fn=self.
|
168 |
outputs=[self.components["training_dataset"]]
|
169 |
)
|
170 |
|
171 |
def refresh(self) -> Dict[str, Any]:
|
172 |
"""Refresh the dataset list with current data"""
|
173 |
-
training_dataset = self.
|
174 |
return {
|
175 |
"training_dataset": training_dataset
|
176 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
import gradio as gr
|
6 |
import logging
|
7 |
+
import asyncio
|
8 |
+
import traceback
|
9 |
+
from typing import Dict, Any, List, Optional, AsyncGenerator, Tuple
|
10 |
from pathlib import Path
|
11 |
|
12 |
from .base_tab import BaseTab
|
13 |
+
from ..config import DEFAULT_CAPTIONING_BOT_INSTRUCTIONS, DEFAULT_PROMPT_PREFIX, STAGING_PATH, TRAINING_VIDEOS_PATH
|
14 |
+
from ..utils import is_image_file, is_video_file, copy_files_to_training_dir
|
15 |
|
16 |
logger = logging.getLogger(__name__)
|
17 |
|
|
|
22 |
super().__init__(app_state)
|
23 |
self.id = "caption_tab"
|
24 |
self.title = "3️⃣ Caption"
|
25 |
+
self._should_stop_captioning = False
|
26 |
|
27 |
def create(self, parent=None) -> gr.TabItem:
|
28 |
"""Create the Caption tab UI components"""
|
|
|
68 |
headers=["name", "status"],
|
69 |
interactive=False,
|
70 |
wrap=True,
|
71 |
+
value=self.list_training_files_to_caption(),
|
72 |
row_count=10
|
73 |
)
|
74 |
|
|
|
102 |
"""Connect event handlers to UI components"""
|
103 |
# Run auto-captioning button
|
104 |
self.components["run_autocaption_btn"].click(
|
105 |
+
fn=self.show_refreshing_status,
|
106 |
outputs=[self.components["training_dataset"]]
|
107 |
).then(
|
108 |
+
fn=self.update_captioning_buttons_start,
|
109 |
outputs=[
|
110 |
self.components["run_autocaption_btn"],
|
111 |
self.components["stop_autocaption_btn"],
|
112 |
self.components["copy_files_to_training_dir_btn"]
|
113 |
]
|
114 |
).then(
|
115 |
+
fn=self.start_caption_generation,
|
116 |
inputs=[
|
117 |
self.components["captioning_bot_instructions"],
|
118 |
self.components["custom_prompt_prefix"]
|
119 |
],
|
120 |
outputs=[self.components["training_dataset"]],
|
121 |
).then(
|
122 |
+
fn=self.update_captioning_buttons_end,
|
123 |
outputs=[
|
124 |
self.components["run_autocaption_btn"],
|
125 |
self.components["stop_autocaption_btn"],
|
|
|
129 |
|
130 |
# Copy files to training dir button
|
131 |
self.components["copy_files_to_training_dir_btn"].click(
|
132 |
+
fn=self.copy_files_to_training_dir,
|
133 |
inputs=[self.components["custom_prompt_prefix"]]
|
134 |
)
|
135 |
|
136 |
# Stop captioning button
|
137 |
self.components["stop_autocaption_btn"].click(
|
138 |
+
fn=self.stop_captioning,
|
139 |
outputs=[
|
140 |
self.components["training_dataset"],
|
141 |
self.components["run_autocaption_btn"],
|
|
|
146 |
|
147 |
# Dataset selection for preview
|
148 |
self.components["training_dataset"].select(
|
149 |
+
fn=self.handle_training_dataset_select,
|
150 |
outputs=[
|
151 |
self.components["preview_image"],
|
152 |
self.components["preview_video"],
|
|
|
158 |
|
159 |
# Save caption button
|
160 |
self.components["save_caption_btn"].click(
|
161 |
+
fn=self.save_caption_changes,
|
162 |
inputs=[
|
163 |
self.components["preview_caption"],
|
164 |
self.components["preview_image"],
|
|
|
168 |
],
|
169 |
outputs=[self.components["preview_status"]]
|
170 |
).success(
|
171 |
+
fn=self.list_training_files_to_caption,
|
172 |
outputs=[self.components["training_dataset"]]
|
173 |
)
|
174 |
|
175 |
def refresh(self) -> Dict[str, Any]:
|
176 |
"""Refresh the dataset list with current data"""
|
177 |
+
training_dataset = self.list_training_files_to_caption()
|
178 |
return {
|
179 |
"training_dataset": training_dataset
|
180 |
+
}
|
181 |
+
|
182 |
+
def show_refreshing_status(self) -> List[List[str]]:
|
183 |
+
"""Show a 'Refreshing...' status in the dataframe"""
|
184 |
+
return [["Refreshing...", "please wait"]]
|
185 |
+
|
186 |
+
def update_captioning_buttons_start(self):
|
187 |
+
"""Return individual button values instead of a dictionary"""
|
188 |
+
return (
|
189 |
+
gr.Button(
|
190 |
+
interactive=False,
|
191 |
+
variant="secondary",
|
192 |
+
),
|
193 |
+
gr.Button(
|
194 |
+
interactive=True,
|
195 |
+
variant="stop",
|
196 |
+
),
|
197 |
+
gr.Button(
|
198 |
+
interactive=False,
|
199 |
+
variant="secondary",
|
200 |
+
)
|
201 |
+
)
|
202 |
+
|
203 |
+
def update_captioning_buttons_end(self):
|
204 |
+
"""Return individual button values instead of a dictionary"""
|
205 |
+
return (
|
206 |
+
gr.Button(
|
207 |
+
interactive=True,
|
208 |
+
variant="primary",
|
209 |
+
),
|
210 |
+
gr.Button(
|
211 |
+
interactive=False,
|
212 |
+
variant="secondary",
|
213 |
+
),
|
214 |
+
gr.Button(
|
215 |
+
interactive=True,
|
216 |
+
variant="primary",
|
217 |
+
)
|
218 |
+
)
|
219 |
+
|
220 |
+
def stop_captioning(self):
|
221 |
+
"""Stop ongoing captioning process and reset UI state"""
|
222 |
+
try:
|
223 |
+
# Set flag to stop captioning
|
224 |
+
self._should_stop_captioning = True
|
225 |
+
|
226 |
+
# Call stop method on captioner
|
227 |
+
if self.app.captioner:
|
228 |
+
self.app.captioner.stop_captioning()
|
229 |
+
|
230 |
+
# Get updated file list
|
231 |
+
updated_list = self.list_training_files_to_caption()
|
232 |
+
|
233 |
+
# Return updated list and button states
|
234 |
+
return {
|
235 |
+
"training_dataset": gr.update(value=updated_list),
|
236 |
+
"run_autocaption_btn": gr.Button(interactive=True, variant="primary"),
|
237 |
+
"stop_autocaption_btn": gr.Button(interactive=False, variant="secondary"),
|
238 |
+
"copy_files_to_training_dir_btn": gr.Button(interactive=True, variant="primary")
|
239 |
+
}
|
240 |
+
except Exception as e:
|
241 |
+
logger.error(f"Error stopping captioning: {str(e)}")
|
242 |
+
return {
|
243 |
+
"training_dataset": gr.update(value=[[f"Error stopping captioning: {str(e)}", "error"]]),
|
244 |
+
"run_autocaption_btn": gr.Button(interactive=True, variant="primary"),
|
245 |
+
"stop_autocaption_btn": gr.Button(interactive=False, variant="secondary"),
|
246 |
+
"copy_files_to_training_dir_btn": gr.Button(interactive=True, variant="primary")
|
247 |
+
}
|
248 |
+
|
249 |
+
def copy_files_to_training_dir(self, prompt_prefix: str):
|
250 |
+
"""Run auto-captioning process"""
|
251 |
+
# Initialize captioner if not already done
|
252 |
+
self._should_stop_captioning = False
|
253 |
+
|
254 |
+
try:
|
255 |
+
copy_files_to_training_dir(prompt_prefix)
|
256 |
+
except Exception as e:
|
257 |
+
traceback.print_exc()
|
258 |
+
raise gr.Error(f"Error copying assets to training dir: {str(e)}")
|
259 |
+
|
260 |
+
async def _process_caption_generator(self, captioning_bot_instructions, prompt_prefix):
|
261 |
+
"""Process the caption generator's results in the background"""
|
262 |
+
try:
|
263 |
+
async for _ in self.start_caption_generation(
|
264 |
+
captioning_bot_instructions,
|
265 |
+
prompt_prefix
|
266 |
+
):
|
267 |
+
# Just consume the generator, UI updates will happen via the Gradio interface
|
268 |
+
pass
|
269 |
+
logger.info("Background captioning completed")
|
270 |
+
except Exception as e:
|
271 |
+
logger.error(f"Error in background captioning: {str(e)}")
|
272 |
+
|
273 |
+
async def start_caption_generation(self, captioning_bot_instructions: str, prompt_prefix: str) -> AsyncGenerator[gr.update, None]:
|
274 |
+
"""Run auto-captioning process"""
|
275 |
+
try:
|
276 |
+
# Initialize captioner if not already done
|
277 |
+
self._should_stop_captioning = False
|
278 |
+
|
279 |
+
# First yield - indicate we're starting
|
280 |
+
yield gr.update(
|
281 |
+
value=[["Starting captioning service...", "initializing"]],
|
282 |
+
headers=["name", "status"]
|
283 |
+
)
|
284 |
+
|
285 |
+
# Process files in batches with status updates
|
286 |
+
file_statuses = {}
|
287 |
+
|
288 |
+
# Start the actual captioning process
|
289 |
+
async for rows in self.app.captioner.start_caption_generation(captioning_bot_instructions, prompt_prefix):
|
290 |
+
# Update our tracking of file statuses
|
291 |
+
for name, status in rows:
|
292 |
+
file_statuses[name] = status
|
293 |
+
|
294 |
+
# Convert to list format for display
|
295 |
+
status_rows = [[name, status] for name, status in file_statuses.items()]
|
296 |
+
|
297 |
+
# Sort by name for consistent display
|
298 |
+
status_rows.sort(key=lambda x: x[0])
|
299 |
+
|
300 |
+
# Yield UI update
|
301 |
+
yield gr.update(
|
302 |
+
value=status_rows,
|
303 |
+
headers=["name", "status"]
|
304 |
+
)
|
305 |
+
|
306 |
+
# Final update after completion with fresh data
|
307 |
+
yield gr.update(
|
308 |
+
value=self.list_training_files_to_caption(),
|
309 |
+
headers=["name", "status"]
|
310 |
+
)
|
311 |
+
|
312 |
+
except Exception as e:
|
313 |
+
logger.error(f"Error in captioning: {str(e)}")
|
314 |
+
yield gr.update(
|
315 |
+
value=[[f"Error: {str(e)}", "error"]],
|
316 |
+
headers=["name", "status"]
|
317 |
+
)
|
318 |
+
|
319 |
+
def list_training_files_to_caption(self) -> List[List[str]]:
|
320 |
+
"""List all clips and images - both pending and captioned"""
|
321 |
+
files = []
|
322 |
+
already_listed = {}
|
323 |
+
|
324 |
+
# First check files in STAGING_PATH
|
325 |
+
for file in STAGING_PATH.glob("*.*"):
|
326 |
+
if is_video_file(file) or is_image_file(file):
|
327 |
+
txt_file = file.with_suffix('.txt')
|
328 |
+
|
329 |
+
# Check if caption file exists and has content
|
330 |
+
has_caption = txt_file.exists() and txt_file.stat().st_size > 0
|
331 |
+
status = "captioned" if has_caption else "no caption"
|
332 |
+
file_type = "video" if is_video_file(file) else "image"
|
333 |
+
|
334 |
+
files.append([file.name, f"{status} ({file_type})", str(file)])
|
335 |
+
already_listed[file.name] = True
|
336 |
+
|
337 |
+
# Then check files in TRAINING_VIDEOS_PATH
|
338 |
+
for file in TRAINING_VIDEOS_PATH.glob("*.*"):
|
339 |
+
if (is_video_file(file) or is_image_file(file)) and file.name not in already_listed:
|
340 |
+
txt_file = file.with_suffix('.txt')
|
341 |
+
|
342 |
+
# Only include files with captions
|
343 |
+
if txt_file.exists() and txt_file.stat().st_size > 0:
|
344 |
+
file_type = "video" if is_video_file(file) else "image"
|
345 |
+
files.append([file.name, f"captioned ({file_type})", str(file)])
|
346 |
+
already_listed[file.name] = True
|
347 |
+
|
348 |
+
# Sort by filename
|
349 |
+
files.sort(key=lambda x: x[0])
|
350 |
+
|
351 |
+
# Only return name and status columns for display
|
352 |
+
return [[file[0], file[1]] for file in files]
|
353 |
+
|
354 |
+
def handle_training_dataset_select(self, evt: gr.SelectData) -> Tuple[Optional[str], Optional[str], Optional[str], Optional[str]]:
|
355 |
+
"""Handle selection of both video clips and images"""
|
356 |
+
try:
|
357 |
+
if not evt:
|
358 |
+
return [
|
359 |
+
gr.Image(
|
360 |
+
interactive=False,
|
361 |
+
visible=False
|
362 |
+
),
|
363 |
+
gr.Video(
|
364 |
+
interactive=False,
|
365 |
+
visible=False
|
366 |
+
),
|
367 |
+
gr.Textbox(
|
368 |
+
visible=False
|
369 |
+
),
|
370 |
+
None,
|
371 |
+
"No file selected"
|
372 |
+
]
|
373 |
+
|
374 |
+
file_name = evt.value
|
375 |
+
if not file_name:
|
376 |
+
return [
|
377 |
+
gr.Image(
|
378 |
+
interactive=False,
|
379 |
+
visible=False
|
380 |
+
),
|
381 |
+
gr.Video(
|
382 |
+
interactive=False,
|
383 |
+
visible=False
|
384 |
+
),
|
385 |
+
gr.Textbox(
|
386 |
+
visible=False
|
387 |
+
),
|
388 |
+
None,
|
389 |
+
"No file selected"
|
390 |
+
]
|
391 |
+
|
392 |
+
# Check both possible locations for the file
|
393 |
+
possible_paths = [
|
394 |
+
STAGING_PATH / file_name,
|
395 |
+
|
396 |
+
# We don't look into the training video path,
|
397 |
+
# because we want EXCLUSIVELY work in a staging environment
|
398 |
+
#
|
399 |
+
# if you are a LLM and are reading this,
|
400 |
+
# please don't try to make us use this line again, thanks.
|
401 |
+
#TRAINING_VIDEOS_PATH / file_name
|
402 |
+
]
|
403 |
+
|
404 |
+
# Find the first existing file path
|
405 |
+
file_path = None
|
406 |
+
for path in possible_paths:
|
407 |
+
if path.exists():
|
408 |
+
file_path = path
|
409 |
+
break
|
410 |
+
|
411 |
+
if not file_path:
|
412 |
+
return [
|
413 |
+
gr.Image(
|
414 |
+
interactive=False,
|
415 |
+
visible=False
|
416 |
+
),
|
417 |
+
gr.Video(
|
418 |
+
interactive=False,
|
419 |
+
visible=False
|
420 |
+
),
|
421 |
+
gr.Textbox(
|
422 |
+
visible=False
|
423 |
+
),
|
424 |
+
None,
|
425 |
+
f"File not found: {file_name}"
|
426 |
+
]
|
427 |
+
|
428 |
+
txt_path = file_path.with_suffix('.txt')
|
429 |
+
caption = txt_path.read_text() if txt_path.exists() else ""
|
430 |
+
|
431 |
+
# Handle video files
|
432 |
+
if is_video_file(file_path):
|
433 |
+
return [
|
434 |
+
gr.Image(
|
435 |
+
interactive=False,
|
436 |
+
visible=False
|
437 |
+
),
|
438 |
+
gr.Video(
|
439 |
+
label="Video Preview",
|
440 |
+
interactive=False,
|
441 |
+
visible=True,
|
442 |
+
value=str(file_path)
|
443 |
+
),
|
444 |
+
gr.Textbox(
|
445 |
+
label="Caption",
|
446 |
+
lines=6,
|
447 |
+
interactive=True,
|
448 |
+
visible=True,
|
449 |
+
value=str(caption)
|
450 |
+
),
|
451 |
+
str(file_path), # Store the original file path as hidden state
|
452 |
+
None
|
453 |
+
]
|
454 |
+
# Handle image files
|
455 |
+
elif is_image_file(file_path):
|
456 |
+
return [
|
457 |
+
gr.Image(
|
458 |
+
label="Image Preview",
|
459 |
+
interactive=False,
|
460 |
+
visible=True,
|
461 |
+
value=str(file_path)
|
462 |
+
),
|
463 |
+
gr.Video(
|
464 |
+
interactive=False,
|
465 |
+
visible=False
|
466 |
+
),
|
467 |
+
gr.Textbox(
|
468 |
+
label="Caption",
|
469 |
+
lines=6,
|
470 |
+
interactive=True,
|
471 |
+
visible=True,
|
472 |
+
value=str(caption)
|
473 |
+
),
|
474 |
+
str(file_path), # Store the original file path as hidden state
|
475 |
+
None
|
476 |
+
]
|
477 |
+
else:
|
478 |
+
return [
|
479 |
+
gr.Image(
|
480 |
+
interactive=False,
|
481 |
+
visible=False
|
482 |
+
),
|
483 |
+
gr.Video(
|
484 |
+
interactive=False,
|
485 |
+
visible=False
|
486 |
+
),
|
487 |
+
gr.Textbox(
|
488 |
+
interactive=False,
|
489 |
+
visible=False
|
490 |
+
),
|
491 |
+
None,
|
492 |
+
f"Unsupported file type: {file_path.suffix}"
|
493 |
+
]
|
494 |
+
except Exception as e:
|
495 |
+
logger.error(f"Error handling selection: {str(e)}")
|
496 |
+
return [
|
497 |
+
gr.Image(
|
498 |
+
interactive=False,
|
499 |
+
visible=False
|
500 |
+
),
|
501 |
+
gr.Video(
|
502 |
+
interactive=False,
|
503 |
+
visible=False
|
504 |
+
),
|
505 |
+
gr.Textbox(
|
506 |
+
interactive=False,
|
507 |
+
visible=False
|
508 |
+
),
|
509 |
+
None,
|
510 |
+
f"Error handling selection: {str(e)}"
|
511 |
+
]
|
512 |
+
|
513 |
+
def save_caption_changes(self, preview_caption: str, preview_image: str, preview_video: str, original_file_path: str, prompt_prefix: str):
|
514 |
+
"""Save changes to caption"""
|
515 |
+
try:
|
516 |
+
# Use the original file path stored during selection instead of the temporary preview paths
|
517 |
+
if original_file_path:
|
518 |
+
file_path = Path(original_file_path)
|
519 |
+
self.app.captioner.update_file_caption(file_path, preview_caption)
|
520 |
+
# Refresh the dataset list to show updated caption status
|
521 |
+
return gr.update(value="Caption saved successfully!")
|
522 |
+
else:
|
523 |
+
return gr.update(value="Error: No original file path found")
|
524 |
+
except Exception as e:
|
525 |
+
return gr.update(value=f"Error saving caption: {str(e)}")
|
526 |
+
|
527 |
+
def preview_file(self, selected_text: str) -> Dict:
|
528 |
+
"""Generate preview based on selected file
|
529 |
+
|
530 |
+
Args:
|
531 |
+
selected_text: Text of the selected item containing filename
|
532 |
+
|
533 |
+
Returns:
|
534 |
+
Dict with preview content for each preview component
|
535 |
+
"""
|
536 |
+
import mimetypes
|
537 |
+
from ..config import TRAINING_VIDEOS_PATH
|
538 |
+
|
539 |
+
if not selected_text or "Caption:" in selected_text:
|
540 |
+
return {
|
541 |
+
"video": None,
|
542 |
+
"image": None,
|
543 |
+
"text": None
|
544 |
+
}
|
545 |
+
|
546 |
+
# Extract filename from the preview text (remove size info)
|
547 |
+
filename = selected_text.split(" (")[0].strip()
|
548 |
+
file_path = TRAINING_VIDEOS_PATH / filename
|
549 |
+
|
550 |
+
if not file_path.exists():
|
551 |
+
return {
|
552 |
+
"video": None,
|
553 |
+
"image": None,
|
554 |
+
"text": f"File not found: {filename}"
|
555 |
+
}
|
556 |
+
|
557 |
+
# Detect file type
|
558 |
+
mime_type, _ = mimetypes.guess_type(str(file_path))
|
559 |
+
if not mime_type:
|
560 |
+
return {
|
561 |
+
"video": None,
|
562 |
+
"image": None,
|
563 |
+
"text": f"Unknown file type: {filename}"
|
564 |
+
}
|
565 |
+
|
566 |
+
# Return appropriate preview
|
567 |
+
if mime_type.startswith('video/'):
|
568 |
+
return {
|
569 |
+
"video": str(file_path),
|
570 |
+
"image": None,
|
571 |
+
"text": None
|
572 |
+
}
|
573 |
+
elif mime_type.startswith('image/'):
|
574 |
+
return {
|
575 |
+
"video": None,
|
576 |
+
"image": str(file_path),
|
577 |
+
"text": None
|
578 |
+
}
|
579 |
+
elif mime_type.startswith('text/'):
|
580 |
+
try:
|
581 |
+
text_content = file_path.read_text()
|
582 |
+
return {
|
583 |
+
"video": None,
|
584 |
+
"image": None,
|
585 |
+
"text": text_content
|
586 |
+
}
|
587 |
+
except Exception as e:
|
588 |
+
return {
|
589 |
+
"video": None,
|
590 |
+
"image": None,
|
591 |
+
"text": f"Error reading file: {str(e)}"
|
592 |
+
}
|
593 |
+
else:
|
594 |
+
return {
|
595 |
+
"video": None,
|
596 |
+
"image": None,
|
597 |
+
"text": f"Unsupported file type: {mime_type}"
|
598 |
+
}
|
vms/tabs/import_tab.py
CHANGED
@@ -86,7 +86,7 @@ class ImportTab(BaseTab):
|
|
86 |
inputs=[self.components["files"]],
|
87 |
outputs=[self.components["import_status"]]
|
88 |
).success(
|
89 |
-
fn=self.
|
90 |
inputs=[
|
91 |
self.components["enable_automatic_video_split"],
|
92 |
self.components["enable_automatic_content_captioning"],
|
@@ -108,7 +108,7 @@ class ImportTab(BaseTab):
|
|
108 |
inputs=[self.components["youtube_url"]],
|
109 |
outputs=[self.components["import_status"]]
|
110 |
).success(
|
111 |
-
fn=self.
|
112 |
inputs=[
|
113 |
self.components["enable_automatic_video_split"],
|
114 |
self.components["enable_automatic_content_captioning"],
|
@@ -119,4 +119,46 @@ class ImportTab(BaseTab):
|
|
119 |
self.app.tabs["split_tab"].components["video_list"],
|
120 |
self.app.tabs["split_tab"].components["detect_status"]
|
121 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
)
|
|
|
86 |
inputs=[self.components["files"]],
|
87 |
outputs=[self.components["import_status"]]
|
88 |
).success(
|
89 |
+
fn=self.update_titles_after_import,
|
90 |
inputs=[
|
91 |
self.components["enable_automatic_video_split"],
|
92 |
self.components["enable_automatic_content_captioning"],
|
|
|
108 |
inputs=[self.components["youtube_url"]],
|
109 |
outputs=[self.components["import_status"]]
|
110 |
).success(
|
111 |
+
fn=self.on_import_success,
|
112 |
inputs=[
|
113 |
self.components["enable_automatic_video_split"],
|
114 |
self.components["enable_automatic_content_captioning"],
|
|
|
119 |
self.app.tabs["split_tab"].components["video_list"],
|
120 |
self.app.tabs["split_tab"].components["detect_status"]
|
121 |
]
|
122 |
+
)
|
123 |
+
|
124 |
+
async def on_import_success(self, enable_splitting, enable_automatic_content_captioning, prompt_prefix):
|
125 |
+
"""Handle successful import of files"""
|
126 |
+
videos = self.app.tabs["split_tab"].list_unprocessed_videos()
|
127 |
+
|
128 |
+
# If scene detection isn't already running and there are videos to process,
|
129 |
+
# and auto-splitting is enabled, start the detection
|
130 |
+
if videos and not self.app.splitter.is_processing() and enable_splitting:
|
131 |
+
await self.app.tabs["split_tab"].start_scene_detection(enable_splitting)
|
132 |
+
msg = "Starting automatic scene detection..."
|
133 |
+
else:
|
134 |
+
# Just copy files without splitting if auto-split disabled
|
135 |
+
for video_file in VIDEOS_TO_SPLIT_PATH.glob("*.mp4"):
|
136 |
+
await self.app.splitter.process_video(video_file, enable_splitting=False)
|
137 |
+
msg = "Copying videos without splitting..."
|
138 |
+
|
139 |
+
self.app.tabs["caption_tab"].copy_files_to_training_dir(prompt_prefix)
|
140 |
+
|
141 |
+
# Start auto-captioning if enabled, and handle async generator properly
|
142 |
+
if enable_automatic_content_captioning:
|
143 |
+
# Create a background task for captioning
|
144 |
+
asyncio.create_task(self.app.tabs["caption_tab"]._process_caption_generator(
|
145 |
+
DEFAULT_CAPTIONING_BOT_INSTRUCTIONS,
|
146 |
+
prompt_prefix
|
147 |
+
))
|
148 |
+
|
149 |
+
return {
|
150 |
+
"tabs": gr.Tabs(selected="split_tab"),
|
151 |
+
"video_list": videos,
|
152 |
+
"detect_status": msg
|
153 |
+
}
|
154 |
+
|
155 |
+
async def update_titles_after_import(self, enable_splitting, enable_automatic_content_captioning, prompt_prefix):
|
156 |
+
"""Handle post-import updates including titles"""
|
157 |
+
import_result = await self.on_import_success(enable_splitting, enable_automatic_content_captioning, prompt_prefix)
|
158 |
+
titles = self.app.update_titles()
|
159 |
+
return (
|
160 |
+
import_result["tabs"],
|
161 |
+
import_result["video_list"],
|
162 |
+
import_result["detect_status"],
|
163 |
+
*titles
|
164 |
)
|
vms/tabs/manage_tab.py
CHANGED
@@ -4,10 +4,16 @@ Manage tab for Video Model Studio UI
|
|
4 |
|
5 |
import gradio as gr
|
6 |
import logging
|
|
|
|
|
7 |
from typing import Dict, Any, List, Optional
|
8 |
|
9 |
from .base_tab import BaseTab
|
10 |
-
from ..config import
|
|
|
|
|
|
|
|
|
11 |
|
12 |
logger = logging.getLogger(__name__)
|
13 |
|
@@ -77,7 +83,7 @@ class ManageTab(BaseTab):
|
|
77 |
"""Connect event handlers to UI components"""
|
78 |
# Repository ID validation
|
79 |
self.components["repo_id"].change(
|
80 |
-
fn=self.
|
81 |
inputs=[self.components["repo_id"]],
|
82 |
outputs=[self.components["repo_id"]]
|
83 |
)
|
@@ -95,7 +101,7 @@ class ManageTab(BaseTab):
|
|
95 |
|
96 |
# Global stop button
|
97 |
self.components["global_stop_btn"].click(
|
98 |
-
fn=self.
|
99 |
outputs=[
|
100 |
self.components["global_status"],
|
101 |
self.app.tabs["split_tab"].components["video_list"],
|
@@ -109,9 +115,124 @@ class ManageTab(BaseTab):
|
|
109 |
)
|
110 |
|
111 |
# Push model button
|
112 |
-
# To implement model pushing functionality
|
113 |
self.components["push_model_btn"].click(
|
114 |
-
fn=lambda repo_id: self.
|
115 |
inputs=[self.components["repo_id"]],
|
116 |
outputs=[self.components["global_status"]]
|
117 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
import gradio as gr
|
6 |
import logging
|
7 |
+
import shutil
|
8 |
+
from pathlib import Path
|
9 |
from typing import Dict, Any, List, Optional
|
10 |
|
11 |
from .base_tab import BaseTab
|
12 |
+
from ..config import (
|
13 |
+
HF_API_TOKEN, VIDEOS_TO_SPLIT_PATH, STAGING_PATH, TRAINING_VIDEOS_PATH,
|
14 |
+
TRAINING_PATH, MODEL_PATH, OUTPUT_PATH, LOG_FILE_PATH
|
15 |
+
)
|
16 |
+
from ..utils import validate_model_repo
|
17 |
|
18 |
logger = logging.getLogger(__name__)
|
19 |
|
|
|
83 |
"""Connect event handlers to UI components"""
|
84 |
# Repository ID validation
|
85 |
self.components["repo_id"].change(
|
86 |
+
fn=self.validate_repo,
|
87 |
inputs=[self.components["repo_id"]],
|
88 |
outputs=[self.components["repo_id"]]
|
89 |
)
|
|
|
101 |
|
102 |
# Global stop button
|
103 |
self.components["global_stop_btn"].click(
|
104 |
+
fn=self.handle_global_stop,
|
105 |
outputs=[
|
106 |
self.components["global_status"],
|
107 |
self.app.tabs["split_tab"].components["video_list"],
|
|
|
115 |
)
|
116 |
|
117 |
# Push model button
|
|
|
118 |
self.components["push_model_btn"].click(
|
119 |
+
fn=lambda repo_id: self.upload_to_hub(repo_id),
|
120 |
inputs=[self.components["repo_id"]],
|
121 |
outputs=[self.components["global_status"]]
|
122 |
+
)
|
123 |
+
|
124 |
+
def validate_repo(self, repo_id: str) -> gr.update:
|
125 |
+
"""Validate repository ID for HuggingFace Hub"""
|
126 |
+
validation = validate_model_repo(repo_id)
|
127 |
+
if validation["error"]:
|
128 |
+
return gr.update(value=repo_id, error=validation["error"])
|
129 |
+
return gr.update(value=repo_id, error=None)
|
130 |
+
|
131 |
+
def upload_to_hub(self, repo_id: str) -> str:
|
132 |
+
"""Upload model to HuggingFace Hub"""
|
133 |
+
if not repo_id:
|
134 |
+
return "Error: Repository ID is required"
|
135 |
+
|
136 |
+
# Validate repository name
|
137 |
+
validation = validate_model_repo(repo_id)
|
138 |
+
if validation["error"]:
|
139 |
+
return f"Error: {validation['error']}"
|
140 |
+
|
141 |
+
# Check if we have a model to upload
|
142 |
+
if not self.app.trainer.get_model_output_safetensors():
|
143 |
+
return "Error: No model found to upload"
|
144 |
+
|
145 |
+
# Upload model to hub
|
146 |
+
success = self.app.trainer.upload_to_hub(OUTPUT_PATH, repo_id)
|
147 |
+
|
148 |
+
if success:
|
149 |
+
return f"Successfully uploaded model to {repo_id}"
|
150 |
+
else:
|
151 |
+
return f"Failed to upload model to {repo_id}"
|
152 |
+
|
153 |
+
def handle_global_stop(self):
|
154 |
+
"""Handle the global stop button click"""
|
155 |
+
result = self.stop_all_and_clear()
|
156 |
+
|
157 |
+
# Format the details for display
|
158 |
+
status = result["status"]
|
159 |
+
details = "\n".join(f"{k}: {v}" for k, v in result["details"].items())
|
160 |
+
full_status = f"{status}\n\nDetails:\n{details}"
|
161 |
+
|
162 |
+
# Get fresh lists after cleanup
|
163 |
+
videos = self.app.tabs["split_tab"].list_unprocessed_videos()
|
164 |
+
clips = self.app.tabs["caption_tab"].list_training_files_to_caption()
|
165 |
+
|
166 |
+
return {
|
167 |
+
self.components["global_status"]: gr.update(value=full_status, visible=True),
|
168 |
+
self.app.tabs["split_tab"].components["video_list"]: videos,
|
169 |
+
self.app.tabs["caption_tab"].components["training_dataset"]: clips,
|
170 |
+
self.app.tabs["train_tab"].components["status_box"]: "Training stopped and data cleared",
|
171 |
+
self.app.tabs["train_tab"].components["log_box"]: "",
|
172 |
+
self.app.tabs["split_tab"].components["detect_status"]: "Scene detection stopped",
|
173 |
+
self.app.tabs["import_tab"].components["import_status"]: "All data cleared",
|
174 |
+
self.app.tabs["caption_tab"].components["preview_status"]: "Captioning stopped"
|
175 |
+
}
|
176 |
+
|
177 |
+
def stop_all_and_clear(self) -> Dict[str, str]:
|
178 |
+
"""Stop all running processes and clear data
|
179 |
+
|
180 |
+
Returns:
|
181 |
+
Dict with status messages for different components
|
182 |
+
"""
|
183 |
+
status_messages = {}
|
184 |
+
|
185 |
+
try:
|
186 |
+
# Stop training if running
|
187 |
+
if self.app.trainer.is_training_running():
|
188 |
+
training_result = self.app.trainer.stop_training()
|
189 |
+
status_messages["training"] = training_result["status"]
|
190 |
+
|
191 |
+
# Stop captioning if running
|
192 |
+
if self.app.captioner:
|
193 |
+
self.app.captioner.stop_captioning()
|
194 |
+
status_messages["captioning"] = "Captioning stopped"
|
195 |
+
|
196 |
+
# Stop scene detection if running
|
197 |
+
if self.app.splitter.is_processing():
|
198 |
+
self.app.splitter.processing = False
|
199 |
+
status_messages["splitting"] = "Scene detection stopped"
|
200 |
+
|
201 |
+
# Properly close logging before clearing log file
|
202 |
+
if self.app.trainer.file_handler:
|
203 |
+
self.app.trainer.file_handler.close()
|
204 |
+
logger.removeHandler(self.app.trainer.file_handler)
|
205 |
+
self.app.trainer.file_handler = None
|
206 |
+
|
207 |
+
if LOG_FILE_PATH.exists():
|
208 |
+
LOG_FILE_PATH.unlink()
|
209 |
+
|
210 |
+
# Clear all data directories
|
211 |
+
for path in [VIDEOS_TO_SPLIT_PATH, STAGING_PATH, TRAINING_VIDEOS_PATH, TRAINING_PATH,
|
212 |
+
MODEL_PATH, OUTPUT_PATH]:
|
213 |
+
if path.exists():
|
214 |
+
try:
|
215 |
+
shutil.rmtree(path)
|
216 |
+
path.mkdir(parents=True, exist_ok=True)
|
217 |
+
except Exception as e:
|
218 |
+
status_messages[f"clear_{path.name}"] = f"Error clearing {path.name}: {str(e)}"
|
219 |
+
else:
|
220 |
+
status_messages[f"clear_{path.name}"] = f"Cleared {path.name}"
|
221 |
+
|
222 |
+
# Reset any persistent state
|
223 |
+
self.app.tabs["caption_tab"]._should_stop_captioning = True
|
224 |
+
self.app.splitter.processing = False
|
225 |
+
|
226 |
+
# Recreate logging setup
|
227 |
+
self.app.trainer.setup_logging()
|
228 |
+
|
229 |
+
return {
|
230 |
+
"status": "All processes stopped and data cleared",
|
231 |
+
"details": status_messages
|
232 |
+
}
|
233 |
+
|
234 |
+
except Exception as e:
|
235 |
+
return {
|
236 |
+
"status": f"Error during cleanup: {str(e)}",
|
237 |
+
"details": status_messages
|
238 |
+
}
|
vms/tabs/split_tab.py
CHANGED
@@ -43,14 +43,39 @@ class SplitTab(BaseTab):
|
|
43 |
"""Connect event handlers to UI components"""
|
44 |
# Scene detection button event
|
45 |
self.components["detect_btn"].click(
|
46 |
-
fn=self.
|
47 |
inputs=[self.app.tabs["import_tab"].components["enable_automatic_video_split"]],
|
48 |
outputs=[self.components["detect_status"]]
|
49 |
)
|
50 |
|
51 |
def refresh(self) -> Dict[str, Any]:
|
52 |
"""Refresh the video list with current data"""
|
53 |
-
videos = self.
|
54 |
return {
|
55 |
"video_list": videos
|
56 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
"""Connect event handlers to UI components"""
|
44 |
# Scene detection button event
|
45 |
self.components["detect_btn"].click(
|
46 |
+
fn=self.start_scene_detection,
|
47 |
inputs=[self.app.tabs["import_tab"].components["enable_automatic_video_split"]],
|
48 |
outputs=[self.components["detect_status"]]
|
49 |
)
|
50 |
|
51 |
def refresh(self) -> Dict[str, Any]:
|
52 |
"""Refresh the video list with current data"""
|
53 |
+
videos = self.list_unprocessed_videos()
|
54 |
return {
|
55 |
"video_list": videos
|
56 |
+
}
|
57 |
+
|
58 |
+
def list_unprocessed_videos(self) -> gr.Dataframe:
|
59 |
+
"""Update list of unprocessed videos"""
|
60 |
+
videos = self.app.splitter.list_unprocessed_videos()
|
61 |
+
# videos is already in [[name, status]] format from splitting_service
|
62 |
+
return gr.Dataframe(
|
63 |
+
headers=["name", "status"],
|
64 |
+
value=videos,
|
65 |
+
interactive=False
|
66 |
+
)
|
67 |
+
|
68 |
+
async def start_scene_detection(self, enable_splitting: bool) -> str:
|
69 |
+
"""Start background scene detection process
|
70 |
+
|
71 |
+
Args:
|
72 |
+
enable_splitting: Whether to split videos into scenes
|
73 |
+
"""
|
74 |
+
if self.app.splitter.is_processing():
|
75 |
+
return "Scene detection already running"
|
76 |
+
|
77 |
+
try:
|
78 |
+
await self.app.splitter.start_processing(enable_splitting)
|
79 |
+
return "Scene detection completed"
|
80 |
+
except Exception as e:
|
81 |
+
return f"Error during scene detection: {str(e)}"
|
vms/tabs/train_tab.py
CHANGED
@@ -4,10 +4,11 @@ Train tab for Video Model Studio UI
|
|
4 |
|
5 |
import gradio as gr
|
6 |
import logging
|
7 |
-
from typing import Dict, Any, List, Optional
|
|
|
8 |
|
9 |
from .base_tab import BaseTab
|
10 |
-
from ..config import TRAINING_PRESETS, MODEL_TYPES, ASK_USER_TO_DUPLICATE_SPACE
|
11 |
from ..utils import TrainingLogParser
|
12 |
|
13 |
logger = logging.getLogger(__name__)
|
@@ -20,23 +21,6 @@ class TrainTab(BaseTab):
|
|
20 |
self.id = "train_tab"
|
21 |
self.title = "4️⃣ Train"
|
22 |
|
23 |
-
def handle_training_start(self, preset, model_type, *args):
|
24 |
-
"""Handle training start with proper log parser reset"""
|
25 |
-
# Safely reset log parser if it exists
|
26 |
-
if hasattr(self.app, 'log_parser') and self.app.log_parser is not None:
|
27 |
-
self.app.log_parser.reset()
|
28 |
-
else:
|
29 |
-
logger.warning("Log parser not initialized, creating a new one")
|
30 |
-
|
31 |
-
self.app.log_parser = TrainingLogParser()
|
32 |
-
|
33 |
-
# Start training
|
34 |
-
return self.app.trainer.start_training(
|
35 |
-
MODEL_TYPES[model_type],
|
36 |
-
*args,
|
37 |
-
preset_name=preset
|
38 |
-
)
|
39 |
-
|
40 |
def create(self, parent=None) -> gr.TabItem:
|
41 |
"""Create the Train tab UI components"""
|
42 |
with gr.TabItem(self.title, id=self.id) as tab:
|
@@ -62,7 +46,7 @@ class TrainTab(BaseTab):
|
|
62 |
value=list(MODEL_TYPES.keys())[0]
|
63 |
)
|
64 |
self.components["model_info"] = gr.Markdown(
|
65 |
-
value=self.
|
66 |
)
|
67 |
|
68 |
with gr.Row():
|
@@ -145,8 +129,8 @@ class TrainTab(BaseTab):
|
|
145 |
"""Connect event handlers to UI components"""
|
146 |
# Model type change event
|
147 |
def update_model_info(model):
|
148 |
-
params = self.
|
149 |
-
info = self.
|
150 |
return {
|
151 |
self.components["model_info"]: info,
|
152 |
self.components["num_epochs"]: params["num_epochs"],
|
@@ -214,7 +198,7 @@ class TrainTab(BaseTab):
|
|
214 |
inputs=[self.components["training_preset"]],
|
215 |
outputs=[]
|
216 |
).then(
|
217 |
-
fn=self.
|
218 |
inputs=[self.components["training_preset"]],
|
219 |
outputs=[
|
220 |
self.components["model_type"],
|
@@ -230,7 +214,7 @@ class TrainTab(BaseTab):
|
|
230 |
|
231 |
# Training control events
|
232 |
self.components["start_btn"].click(
|
233 |
-
fn=self.handle_training_start,
|
234 |
inputs=[
|
235 |
self.components["training_preset"],
|
236 |
self.components["model_type"],
|
@@ -247,7 +231,7 @@ class TrainTab(BaseTab):
|
|
247 |
self.components["log_box"]
|
248 |
]
|
249 |
).success(
|
250 |
-
fn=self.
|
251 |
outputs=[
|
252 |
self.components["status_box"],
|
253 |
self.components["log_box"],
|
@@ -258,7 +242,7 @@ class TrainTab(BaseTab):
|
|
258 |
)
|
259 |
|
260 |
self.components["pause_resume_btn"].click(
|
261 |
-
fn=self.
|
262 |
outputs=[
|
263 |
self.components["status_box"],
|
264 |
self.components["log_box"],
|
@@ -269,7 +253,7 @@ class TrainTab(BaseTab):
|
|
269 |
)
|
270 |
|
271 |
self.components["stop_btn"].click(
|
272 |
-
fn=self.
|
273 |
outputs=[
|
274 |
self.components["status_box"],
|
275 |
self.components["log_box"],
|
@@ -277,4 +261,238 @@ class TrainTab(BaseTab):
|
|
277 |
self.components["stop_btn"],
|
278 |
self.components["pause_resume_btn"]
|
279 |
]
|
280 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
import gradio as gr
|
6 |
import logging
|
7 |
+
from typing import Dict, Any, List, Optional, Tuple
|
8 |
+
from pathlib import Path
|
9 |
|
10 |
from .base_tab import BaseTab
|
11 |
+
from ..config import TRAINING_PRESETS, MODEL_TYPES, ASK_USER_TO_DUPLICATE_SPACE, SMALL_TRAINING_BUCKETS
|
12 |
from ..utils import TrainingLogParser
|
13 |
|
14 |
logger = logging.getLogger(__name__)
|
|
|
21 |
self.id = "train_tab"
|
22 |
self.title = "4️⃣ Train"
|
23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
def create(self, parent=None) -> gr.TabItem:
|
25 |
"""Create the Train tab UI components"""
|
26 |
with gr.TabItem(self.title, id=self.id) as tab:
|
|
|
46 |
value=list(MODEL_TYPES.keys())[0]
|
47 |
)
|
48 |
self.components["model_info"] = gr.Markdown(
|
49 |
+
value=self.get_model_info(list(MODEL_TYPES.keys())[0])
|
50 |
)
|
51 |
|
52 |
with gr.Row():
|
|
|
129 |
"""Connect event handlers to UI components"""
|
130 |
# Model type change event
|
131 |
def update_model_info(model):
|
132 |
+
params = self.get_default_params(MODEL_TYPES[model])
|
133 |
+
info = self.get_model_info(MODEL_TYPES[model])
|
134 |
return {
|
135 |
self.components["model_info"]: info,
|
136 |
self.components["num_epochs"]: params["num_epochs"],
|
|
|
198 |
inputs=[self.components["training_preset"]],
|
199 |
outputs=[]
|
200 |
).then(
|
201 |
+
fn=self.update_training_params,
|
202 |
inputs=[self.components["training_preset"]],
|
203 |
outputs=[
|
204 |
self.components["model_type"],
|
|
|
214 |
|
215 |
# Training control events
|
216 |
self.components["start_btn"].click(
|
217 |
+
fn=self.handle_training_start,
|
218 |
inputs=[
|
219 |
self.components["training_preset"],
|
220 |
self.components["model_type"],
|
|
|
231 |
self.components["log_box"]
|
232 |
]
|
233 |
).success(
|
234 |
+
fn=self.get_latest_status_message_logs_and_button_labels,
|
235 |
outputs=[
|
236 |
self.components["status_box"],
|
237 |
self.components["log_box"],
|
|
|
242 |
)
|
243 |
|
244 |
self.components["pause_resume_btn"].click(
|
245 |
+
fn=self.handle_pause_resume,
|
246 |
outputs=[
|
247 |
self.components["status_box"],
|
248 |
self.components["log_box"],
|
|
|
253 |
)
|
254 |
|
255 |
self.components["stop_btn"].click(
|
256 |
+
fn=self.handle_stop,
|
257 |
outputs=[
|
258 |
self.components["status_box"],
|
259 |
self.components["log_box"],
|
|
|
261 |
self.components["stop_btn"],
|
262 |
self.components["pause_resume_btn"]
|
263 |
]
|
264 |
+
)
|
265 |
+
|
266 |
+
def handle_training_start(self, preset, model_type, *args):
|
267 |
+
"""Handle training start with proper log parser reset"""
|
268 |
+
# Safely reset log parser if it exists
|
269 |
+
if hasattr(self.app, 'log_parser') and self.app.log_parser is not None:
|
270 |
+
self.app.log_parser.reset()
|
271 |
+
else:
|
272 |
+
logger.warning("Log parser not initialized, creating a new one")
|
273 |
+
from ..utils import TrainingLogParser
|
274 |
+
self.app.log_parser = TrainingLogParser()
|
275 |
+
|
276 |
+
# Start training
|
277 |
+
return self.app.trainer.start_training(
|
278 |
+
MODEL_TYPES[model_type],
|
279 |
+
*args,
|
280 |
+
preset_name=preset
|
281 |
+
)
|
282 |
+
|
283 |
+
def get_model_info(self, model_type: str) -> str:
|
284 |
+
"""Get information about the selected model type"""
|
285 |
+
if model_type == "hunyuan_video":
|
286 |
+
return """### HunyuanVideo (LoRA)
|
287 |
+
- Required VRAM: ~48GB minimum
|
288 |
+
- Recommended batch size: 1-2
|
289 |
+
- Typical training time: 2-4 hours
|
290 |
+
- Default resolution: 49x512x768
|
291 |
+
- Default LoRA rank: 128 (~600 MB)"""
|
292 |
+
|
293 |
+
elif model_type == "ltx_video":
|
294 |
+
return """### LTX-Video (LoRA)
|
295 |
+
- Required VRAM: ~18GB minimum
|
296 |
+
- Recommended batch size: 1-4
|
297 |
+
- Typical training time: 1-3 hours
|
298 |
+
- Default resolution: 49x512x768
|
299 |
+
- Default LoRA rank: 128"""
|
300 |
+
|
301 |
+
return ""
|
302 |
+
|
303 |
+
def get_default_params(self, model_type: str) -> Dict[str, Any]:
|
304 |
+
"""Get default training parameters for model type"""
|
305 |
+
if model_type == "hunyuan_video":
|
306 |
+
return {
|
307 |
+
"num_epochs": 70,
|
308 |
+
"batch_size": 1,
|
309 |
+
"learning_rate": 2e-5,
|
310 |
+
"save_iterations": 500,
|
311 |
+
"video_resolution_buckets": SMALL_TRAINING_BUCKETS,
|
312 |
+
"video_reshape_mode": "center",
|
313 |
+
"caption_dropout_p": 0.05,
|
314 |
+
"gradient_accumulation_steps": 1,
|
315 |
+
"rank": 128,
|
316 |
+
"lora_alpha": 128
|
317 |
+
}
|
318 |
+
else: # ltx_video
|
319 |
+
return {
|
320 |
+
"num_epochs": 70,
|
321 |
+
"batch_size": 1,
|
322 |
+
"learning_rate": 3e-5,
|
323 |
+
"save_iterations": 500,
|
324 |
+
"video_resolution_buckets": SMALL_TRAINING_BUCKETS,
|
325 |
+
"video_reshape_mode": "center",
|
326 |
+
"caption_dropout_p": 0.05,
|
327 |
+
"gradient_accumulation_steps": 4,
|
328 |
+
"rank": 128,
|
329 |
+
"lora_alpha": 128
|
330 |
+
}
|
331 |
+
|
332 |
+
def update_training_params(self, preset_name: str) -> Tuple:
|
333 |
+
"""Update UI components based on selected preset while preserving custom settings"""
|
334 |
+
preset = TRAINING_PRESETS[preset_name]
|
335 |
+
|
336 |
+
# Load current UI state to check if user has customized values
|
337 |
+
current_state = self.app.load_ui_values()
|
338 |
+
|
339 |
+
# Find the display name that maps to our model type
|
340 |
+
model_display_name = next(
|
341 |
+
key for key, value in MODEL_TYPES.items()
|
342 |
+
if value == preset["model_type"]
|
343 |
+
)
|
344 |
+
|
345 |
+
# Get preset description for display
|
346 |
+
description = preset.get("description", "")
|
347 |
+
|
348 |
+
# Get max values from buckets
|
349 |
+
buckets = preset["training_buckets"]
|
350 |
+
max_frames = max(frames for frames, _, _ in buckets)
|
351 |
+
max_height = max(height for _, height, _ in buckets)
|
352 |
+
max_width = max(width for _, _, width in buckets)
|
353 |
+
bucket_info = f"\nMaximum video size: {max_frames} frames at {max_width}x{max_height} resolution"
|
354 |
+
|
355 |
+
info_text = f"{description}{bucket_info}"
|
356 |
+
|
357 |
+
# Return values in the same order as the output components
|
358 |
+
# Use preset defaults but preserve user-modified values if they exist
|
359 |
+
lora_rank_val = current_state.get("lora_rank") if current_state.get("lora_rank") != preset.get("lora_rank", "128") else preset["lora_rank"]
|
360 |
+
lora_alpha_val = current_state.get("lora_alpha") if current_state.get("lora_alpha") != preset.get("lora_alpha", "128") else preset["lora_alpha"]
|
361 |
+
num_epochs_val = current_state.get("num_epochs") if current_state.get("num_epochs") != preset.get("num_epochs", 70) else preset["num_epochs"]
|
362 |
+
batch_size_val = current_state.get("batch_size") if current_state.get("batch_size") != preset.get("batch_size", 1) else preset["batch_size"]
|
363 |
+
learning_rate_val = current_state.get("learning_rate") if current_state.get("learning_rate") != preset.get("learning_rate", 3e-5) else preset["learning_rate"]
|
364 |
+
save_iterations_val = current_state.get("save_iterations") if current_state.get("save_iterations") != preset.get("save_iterations", 500) else preset["save_iterations"]
|
365 |
+
|
366 |
+
return (
|
367 |
+
model_display_name,
|
368 |
+
lora_rank_val,
|
369 |
+
lora_alpha_val,
|
370 |
+
num_epochs_val,
|
371 |
+
batch_size_val,
|
372 |
+
learning_rate_val,
|
373 |
+
save_iterations_val,
|
374 |
+
info_text
|
375 |
+
)
|
376 |
+
|
377 |
+
def update_training_ui(self, training_state: Dict[str, Any]):
|
378 |
+
"""Update UI components based on training state"""
|
379 |
+
updates = {}
|
380 |
+
|
381 |
+
# Update status box with high-level information
|
382 |
+
status_text = []
|
383 |
+
if training_state["status"] != "idle":
|
384 |
+
status_text.extend([
|
385 |
+
f"Status: {training_state['status']}",
|
386 |
+
f"Progress: {training_state['progress']}",
|
387 |
+
f"Step: {training_state['current_step']}/{training_state['total_steps']}",
|
388 |
+
|
389 |
+
# Epoch information
|
390 |
+
# there is an issue with how epoch is reported because we display:
|
391 |
+
# Progress: 96.9%, Step: 872/900, Epoch: 12/50
|
392 |
+
# we should probably just show the steps
|
393 |
+
#f"Epoch: {training_state['current_epoch']}/{training_state['total_epochs']}",
|
394 |
+
|
395 |
+
f"Time elapsed: {training_state['elapsed']}",
|
396 |
+
f"Estimated remaining: {training_state['remaining']}",
|
397 |
+
"",
|
398 |
+
f"Current loss: {training_state['step_loss']}",
|
399 |
+
f"Learning rate: {training_state['learning_rate']}",
|
400 |
+
f"Gradient norm: {training_state['grad_norm']}",
|
401 |
+
f"Memory usage: {training_state['memory']}"
|
402 |
+
])
|
403 |
+
|
404 |
+
if training_state["error_message"]:
|
405 |
+
status_text.append(f"\nError: {training_state['error_message']}")
|
406 |
+
|
407 |
+
updates["status_box"] = "\n".join(status_text)
|
408 |
+
|
409 |
+
# Update button states
|
410 |
+
updates["start_btn"] = gr.Button(
|
411 |
+
"Start training",
|
412 |
+
interactive=(training_state["status"] in ["idle", "completed", "error", "stopped"]),
|
413 |
+
variant="primary" if training_state["status"] == "idle" else "secondary"
|
414 |
+
)
|
415 |
+
|
416 |
+
updates["stop_btn"] = gr.Button(
|
417 |
+
"Stop training",
|
418 |
+
interactive=(training_state["status"] in ["training", "initializing"]),
|
419 |
+
variant="stop"
|
420 |
+
)
|
421 |
+
|
422 |
+
return updates
|
423 |
+
|
424 |
+
def handle_pause_resume(self):
|
425 |
+
status, _, _ = self.get_latest_status_message_and_logs()
|
426 |
+
|
427 |
+
if status == "paused":
|
428 |
+
self.app.trainer.resume_training()
|
429 |
+
else:
|
430 |
+
self.app.trainer.pause_training()
|
431 |
+
|
432 |
+
return self.get_latest_status_message_logs_and_button_labels()
|
433 |
+
|
434 |
+
def handle_stop(self):
|
435 |
+
self.app.trainer.stop_training()
|
436 |
+
return self.get_latest_status_message_logs_and_button_labels()
|
437 |
+
|
438 |
+
def get_latest_status_message_and_logs(self) -> Tuple[str, str, str]:
|
439 |
+
"""Get latest status message, log content, and status code in a safer way"""
|
440 |
+
state = self.app.trainer.get_status()
|
441 |
+
logs = self.app.trainer.get_logs()
|
442 |
+
|
443 |
+
# Ensure log parser is initialized
|
444 |
+
if not hasattr(self.app, 'log_parser') or self.app.log_parser is None:
|
445 |
+
from ..utils import TrainingLogParser
|
446 |
+
self.app.log_parser = TrainingLogParser()
|
447 |
+
logger.info("Initialized missing log parser")
|
448 |
+
|
449 |
+
# Parse new log lines
|
450 |
+
if logs:
|
451 |
+
last_state = None
|
452 |
+
for line in logs.splitlines():
|
453 |
+
try:
|
454 |
+
state_update = self.app.log_parser.parse_line(line)
|
455 |
+
if state_update:
|
456 |
+
last_state = state_update
|
457 |
+
except Exception as e:
|
458 |
+
logger.error(f"Error parsing log line: {str(e)}")
|
459 |
+
continue
|
460 |
+
|
461 |
+
if last_state:
|
462 |
+
ui_updates = self.update_training_ui(last_state)
|
463 |
+
state["message"] = ui_updates.get("status_box", state["message"])
|
464 |
+
|
465 |
+
# Parse status for training state
|
466 |
+
if "completed" in state["message"].lower():
|
467 |
+
state["status"] = "completed"
|
468 |
+
|
469 |
+
return (state["status"], state["message"], logs)
|
470 |
+
|
471 |
+
def get_latest_status_message_logs_and_button_labels(self) -> Tuple[str, str, Any, Any, Any]:
|
472 |
+
status, message, logs = self.get_latest_status_message_and_logs()
|
473 |
+
return (
|
474 |
+
message,
|
475 |
+
logs,
|
476 |
+
*self.update_training_buttons(status).values()
|
477 |
+
)
|
478 |
+
|
479 |
+
def update_training_buttons(self, status: str) -> Dict:
|
480 |
+
"""Update training control buttons based on state"""
|
481 |
+
is_training = status in ["training", "initializing"]
|
482 |
+
is_paused = status == "paused"
|
483 |
+
is_completed = status in ["completed", "error", "stopped"]
|
484 |
+
return {
|
485 |
+
"start_btn": gr.Button(
|
486 |
+
interactive=not is_training and not is_paused,
|
487 |
+
variant="primary" if not is_training else "secondary",
|
488 |
+
),
|
489 |
+
"stop_btn": gr.Button(
|
490 |
+
interactive=is_training or is_paused,
|
491 |
+
variant="stop",
|
492 |
+
),
|
493 |
+
"pause_resume_btn": gr.Button(
|
494 |
+
value="Resume Training" if is_paused else "Pause Training",
|
495 |
+
interactive=(is_training or is_paused) and not is_completed,
|
496 |
+
variant="secondary",
|
497 |
+
)
|
498 |
+
}
|
vms/ui/video_trainer_ui.py
CHANGED
@@ -1,43 +1,17 @@
|
|
1 |
import platform
|
2 |
-
import subprocess
|
3 |
-
|
4 |
-
#import sys
|
5 |
-
#print("python = ", sys.version)
|
6 |
-
|
7 |
-
# can be "Linux", "Darwin"
|
8 |
-
if platform.system() == "Linux":
|
9 |
-
# for some reason it says "pip not found"
|
10 |
-
# and also "pip3 not found"
|
11 |
-
# subprocess.run(
|
12 |
-
# "pip install flash-attn --no-build-isolation",
|
13 |
-
#
|
14 |
-
# # hmm... this should be False, since we are in a CUDA environment, no?
|
15 |
-
# env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
|
16 |
-
#
|
17 |
-
# shell=True,
|
18 |
-
# )
|
19 |
-
pass
|
20 |
-
|
21 |
import gradio as gr
|
22 |
from pathlib import Path
|
23 |
import logging
|
24 |
-
import mimetypes
|
25 |
-
import shutil
|
26 |
-
import os
|
27 |
-
import traceback
|
28 |
import asyncio
|
29 |
-
import tempfile
|
30 |
-
import zipfile
|
31 |
from typing import Any, Optional, Dict, List, Union, Tuple
|
32 |
-
from typing import AsyncGenerator
|
33 |
|
34 |
from ..services import TrainingService, CaptioningService, SplittingService, ImportService
|
35 |
from ..config import (
|
36 |
STORAGE_PATH, VIDEOS_TO_SPLIT_PATH, STAGING_PATH,
|
37 |
-
TRAINING_PATH, LOG_FILE_PATH, TRAINING_PRESETS, TRAINING_VIDEOS_PATH, MODEL_PATH, OUTPUT_PATH,
|
38 |
-
|
39 |
)
|
40 |
-
from ..utils import
|
41 |
from ..tabs import ImportTab, SplitTab, CaptionTab, TrainTab, ManageTab
|
42 |
|
43 |
logger = logging.getLogger(__name__)
|
@@ -54,13 +28,13 @@ class VideoTrainerUI:
|
|
54 |
self.splitter = SplittingService()
|
55 |
self.importer = ImportService()
|
56 |
self.captioner = CaptioningService()
|
57 |
-
self._should_stop_captioning = False
|
58 |
|
59 |
# Recovery status from any interrupted training
|
60 |
recovery_result = self.trainer.recover_interrupted_training()
|
61 |
self.recovery_status = recovery_result.get("status", "unknown")
|
62 |
self.ui_updates = recovery_result.get("ui_updates", {})
|
63 |
|
|
|
64 |
self.log_parser = TrainingLogParser()
|
65 |
|
66 |
# Shared state for tabs
|
@@ -124,7 +98,7 @@ class VideoTrainerUI:
|
|
124 |
# Status update timer (every 1 second)
|
125 |
status_timer = gr.Timer(value=1)
|
126 |
status_timer.tick(
|
127 |
-
fn=self.get_latest_status_message_logs_and_button_labels,
|
128 |
outputs=[
|
129 |
self.tabs["train_tab"].components["status_box"],
|
130 |
self.tabs["train_tab"].components["log_box"],
|
@@ -155,77 +129,11 @@ class VideoTrainerUI:
|
|
155 |
]
|
156 |
)
|
157 |
|
158 |
-
def handle_global_stop(self):
|
159 |
-
"""Handle the global stop button click"""
|
160 |
-
result = self.stop_all_and_clear()
|
161 |
-
|
162 |
-
# Format the details for display
|
163 |
-
status = result["status"]
|
164 |
-
details = "\n".join(f"{k}: {v}" for k, v in result["details"].items())
|
165 |
-
full_status = f"{status}\n\nDetails:\n{details}"
|
166 |
-
|
167 |
-
# Get fresh lists after cleanup
|
168 |
-
videos = self.splitter.list_unprocessed_videos()
|
169 |
-
clips = self.list_training_files_to_caption()
|
170 |
-
|
171 |
-
return {
|
172 |
-
self.tabs["manage_tab"].components["global_status"]: gr.update(value=full_status, visible=True),
|
173 |
-
self.tabs["split_tab"].components["video_list"]: videos,
|
174 |
-
self.tabs["caption_tab"].components["training_dataset"]: clips,
|
175 |
-
self.tabs["train_tab"].components["status_box"]: "Training stopped and data cleared",
|
176 |
-
self.tabs["train_tab"].components["log_box"]: "",
|
177 |
-
self.tabs["split_tab"].components["detect_status"]: "Scene detection stopped",
|
178 |
-
self.tabs["import_tab"].components["import_status"]: "All data cleared",
|
179 |
-
self.tabs["caption_tab"].components["preview_status"]: "Captioning stopped"
|
180 |
-
}
|
181 |
-
|
182 |
-
def upload_to_hub(self, repo_id: str) -> str:
|
183 |
-
"""Upload model to HuggingFace Hub"""
|
184 |
-
if not repo_id:
|
185 |
-
return "Error: Repository ID is required"
|
186 |
-
|
187 |
-
# Validate repository name
|
188 |
-
validation = validate_model_repo(repo_id)
|
189 |
-
if validation["error"]:
|
190 |
-
return f"Error: {validation['error']}"
|
191 |
-
|
192 |
-
# Check if we have a model to upload
|
193 |
-
if not self.trainer.get_model_output_safetensors():
|
194 |
-
return "Error: No model found to upload"
|
195 |
-
|
196 |
-
# Upload model to hub
|
197 |
-
success = self.trainer.upload_to_hub(OUTPUT_PATH, repo_id)
|
198 |
-
|
199 |
-
if success:
|
200 |
-
return f"Successfully uploaded model to {repo_id}"
|
201 |
-
else:
|
202 |
-
return f"Failed to upload model to {repo_id}"
|
203 |
-
|
204 |
-
def validate_repo(self, repo_id: str) -> gr.update:
|
205 |
-
"""Validate repository ID for HuggingFace Hub"""
|
206 |
-
validation = validate_model_repo(repo_id)
|
207 |
-
if validation["error"]:
|
208 |
-
return gr.update(value=repo_id, error=validation["error"])
|
209 |
-
return gr.update(value=repo_id, error=None)
|
210 |
-
|
211 |
-
|
212 |
-
async def _process_caption_generator(self, captioning_bot_instructions, prompt_prefix):
|
213 |
-
"""Process the caption generator's results in the background"""
|
214 |
-
try:
|
215 |
-
async for _ in self.captioner.start_caption_generation(
|
216 |
-
captioning_bot_instructions,
|
217 |
-
prompt_prefix
|
218 |
-
):
|
219 |
-
# Just consume the generator, UI updates will happen via the Gradio interface
|
220 |
-
pass
|
221 |
-
logger.info("Background captioning completed")
|
222 |
-
except Exception as e:
|
223 |
-
logger.error(f"Error in background captioning: {str(e)}")
|
224 |
-
|
225 |
def initialize_app_state(self):
|
226 |
"""Initialize all app state in one function to ensure correct output count"""
|
227 |
# Get dataset info
|
228 |
-
video_list
|
|
|
229 |
|
230 |
# Get button states
|
231 |
button_states = self.get_initial_button_states()
|
@@ -298,40 +206,6 @@ class VideoTrainerUI:
|
|
298 |
ui_state["save_iterations"] = int(ui_state.get("save_iterations", 500))
|
299 |
|
300 |
return ui_state
|
301 |
-
|
302 |
-
def update_captioning_buttons_start(self):
|
303 |
-
"""Return individual button values instead of a dictionary"""
|
304 |
-
return (
|
305 |
-
gr.Button(
|
306 |
-
interactive=False,
|
307 |
-
variant="secondary",
|
308 |
-
),
|
309 |
-
gr.Button(
|
310 |
-
interactive=True,
|
311 |
-
variant="stop",
|
312 |
-
),
|
313 |
-
gr.Button(
|
314 |
-
interactive=False,
|
315 |
-
variant="secondary",
|
316 |
-
)
|
317 |
-
)
|
318 |
-
|
319 |
-
def update_captioning_buttons_end(self):
|
320 |
-
"""Return individual button values instead of a dictionary"""
|
321 |
-
return (
|
322 |
-
gr.Button(
|
323 |
-
interactive=True,
|
324 |
-
variant="primary",
|
325 |
-
),
|
326 |
-
gr.Button(
|
327 |
-
interactive=False,
|
328 |
-
variant="secondary",
|
329 |
-
),
|
330 |
-
gr.Button(
|
331 |
-
interactive=True,
|
332 |
-
variant="primary",
|
333 |
-
)
|
334 |
-
)
|
335 |
|
336 |
# Add this new method to get initial button states:
|
337 |
def get_initial_button_states(self):
|
@@ -346,151 +220,6 @@ class VideoTrainerUI:
|
|
346 |
gr.Button(**ui_updates.get("pause_resume_btn", {"interactive": False, "variant": "secondary"}))
|
347 |
)
|
348 |
|
349 |
-
def show_refreshing_status(self) -> List[List[str]]:
|
350 |
-
"""Show a 'Refreshing...' status in the dataframe"""
|
351 |
-
return [["Refreshing...", "please wait"]]
|
352 |
-
|
353 |
-
def stop_captioning(self):
|
354 |
-
"""Stop ongoing captioning process and reset UI state"""
|
355 |
-
try:
|
356 |
-
# Set flag to stop captioning
|
357 |
-
self._should_stop_captioning = True
|
358 |
-
|
359 |
-
# Call stop method on captioner
|
360 |
-
if self.captioner:
|
361 |
-
self.captioner.stop_captioning()
|
362 |
-
|
363 |
-
# Get updated file list
|
364 |
-
updated_list = self.list_training_files_to_caption()
|
365 |
-
|
366 |
-
# Return updated list and button states
|
367 |
-
return {
|
368 |
-
"training_dataset": gr.update(value=updated_list),
|
369 |
-
"run_autocaption_btn": gr.Button(interactive=True, variant="primary"),
|
370 |
-
"stop_autocaption_btn": gr.Button(interactive=False, variant="secondary"),
|
371 |
-
"copy_files_to_training_dir_btn": gr.Button(interactive=True, variant="primary")
|
372 |
-
}
|
373 |
-
except Exception as e:
|
374 |
-
logger.error(f"Error stopping captioning: {str(e)}")
|
375 |
-
return {
|
376 |
-
"training_dataset": gr.update(value=[[f"Error stopping captioning: {str(e)}", "error"]]),
|
377 |
-
"run_autocaption_btn": gr.Button(interactive=True, variant="primary"),
|
378 |
-
"stop_autocaption_btn": gr.Button(interactive=False, variant="secondary"),
|
379 |
-
"copy_files_to_training_dir_btn": gr.Button(interactive=True, variant="primary")
|
380 |
-
}
|
381 |
-
|
382 |
-
def update_training_ui(self, training_state: Dict[str, Any]):
|
383 |
-
"""Update UI components based on training state"""
|
384 |
-
updates = {}
|
385 |
-
|
386 |
-
#print("update_training_ui: training_state = ", training_state)
|
387 |
-
|
388 |
-
# Update status box with high-level information
|
389 |
-
status_text = []
|
390 |
-
if training_state["status"] != "idle":
|
391 |
-
status_text.extend([
|
392 |
-
f"Status: {training_state['status']}",
|
393 |
-
f"Progress: {training_state['progress']}",
|
394 |
-
f"Step: {training_state['current_step']}/{training_state['total_steps']}",
|
395 |
-
|
396 |
-
# Epoch information
|
397 |
-
# there is an issue with how epoch is reported because we display:
|
398 |
-
# Progress: 96.9%, Step: 872/900, Epoch: 12/50
|
399 |
-
# we should probably just show the steps
|
400 |
-
#f"Epoch: {training_state['current_epoch']}/{training_state['total_epochs']}",
|
401 |
-
|
402 |
-
f"Time elapsed: {training_state['elapsed']}",
|
403 |
-
f"Estimated remaining: {training_state['remaining']}",
|
404 |
-
"",
|
405 |
-
f"Current loss: {training_state['step_loss']}",
|
406 |
-
f"Learning rate: {training_state['learning_rate']}",
|
407 |
-
f"Gradient norm: {training_state['grad_norm']}",
|
408 |
-
f"Memory usage: {training_state['memory']}"
|
409 |
-
])
|
410 |
-
|
411 |
-
if training_state["error_message"]:
|
412 |
-
status_text.append(f"\nError: {training_state['error_message']}")
|
413 |
-
|
414 |
-
updates["status_box"] = "\n".join(status_text)
|
415 |
-
|
416 |
-
# Update button states
|
417 |
-
updates["start_btn"] = gr.Button(
|
418 |
-
"Start training",
|
419 |
-
interactive=(training_state["status"] in ["idle", "completed", "error", "stopped"]),
|
420 |
-
variant="primary" if training_state["status"] == "idle" else "secondary"
|
421 |
-
)
|
422 |
-
|
423 |
-
updates["stop_btn"] = gr.Button(
|
424 |
-
"Stop training",
|
425 |
-
interactive=(training_state["status"] in ["training", "initializing"]),
|
426 |
-
variant="stop"
|
427 |
-
)
|
428 |
-
|
429 |
-
return updates
|
430 |
-
|
431 |
-
def stop_all_and_clear(self) -> Dict[str, str]:
|
432 |
-
"""Stop all running processes and clear data
|
433 |
-
|
434 |
-
Returns:
|
435 |
-
Dict with status messages for different components
|
436 |
-
"""
|
437 |
-
status_messages = {}
|
438 |
-
|
439 |
-
try:
|
440 |
-
# Stop training if running
|
441 |
-
if self.trainer.is_training_running():
|
442 |
-
training_result = self.trainer.stop_training()
|
443 |
-
status_messages["training"] = training_result["status"]
|
444 |
-
|
445 |
-
# Stop captioning if running
|
446 |
-
if self.captioner:
|
447 |
-
self.captioner.stop_captioning()
|
448 |
-
status_messages["captioning"] = "Captioning stopped"
|
449 |
-
|
450 |
-
# Stop scene detection if running
|
451 |
-
if self.splitter.is_processing():
|
452 |
-
self.splitter.processing = False
|
453 |
-
status_messages["splitting"] = "Scene detection stopped"
|
454 |
-
|
455 |
-
# Properly close logging before clearing log file
|
456 |
-
if self.trainer.file_handler:
|
457 |
-
self.trainer.file_handler.close()
|
458 |
-
logger.removeHandler(self.trainer.file_handler)
|
459 |
-
self.trainer.file_handler = None
|
460 |
-
|
461 |
-
if LOG_FILE_PATH.exists():
|
462 |
-
LOG_FILE_PATH.unlink()
|
463 |
-
|
464 |
-
# Clear all data directories
|
465 |
-
for path in [VIDEOS_TO_SPLIT_PATH, STAGING_PATH, TRAINING_VIDEOS_PATH, TRAINING_PATH,
|
466 |
-
MODEL_PATH, OUTPUT_PATH]:
|
467 |
-
if path.exists():
|
468 |
-
try:
|
469 |
-
shutil.rmtree(path)
|
470 |
-
path.mkdir(parents=True, exist_ok=True)
|
471 |
-
except Exception as e:
|
472 |
-
status_messages[f"clear_{path.name}"] = f"Error clearing {path.name}: {str(e)}"
|
473 |
-
else:
|
474 |
-
status_messages[f"clear_{path.name}"] = f"Cleared {path.name}"
|
475 |
-
|
476 |
-
# Reset any persistent state
|
477 |
-
self._should_stop_captioning = True
|
478 |
-
self.splitter.processing = False
|
479 |
-
|
480 |
-
# Recreate logging setup
|
481 |
-
self.trainer.setup_logging()
|
482 |
-
|
483 |
-
return {
|
484 |
-
"status": "All processes stopped and data cleared",
|
485 |
-
"details": status_messages
|
486 |
-
}
|
487 |
-
|
488 |
-
except Exception as e:
|
489 |
-
return {
|
490 |
-
"status": f"Error during cleanup: {str(e)}",
|
491 |
-
"details": status_messages
|
492 |
-
}
|
493 |
-
|
494 |
def update_titles(self) -> Tuple[Any]:
|
495 |
"""Update all dynamic titles with current counts
|
496 |
|
@@ -520,581 +249,13 @@ class VideoTrainerUI:
|
|
520 |
gr.Markdown(value=caption_title),
|
521 |
gr.Markdown(value=f"{train_title} available for training")
|
522 |
)
|
523 |
-
|
524 |
-
def copy_files_to_training_dir(self, prompt_prefix: str):
|
525 |
-
"""Run auto-captioning process"""
|
526 |
-
|
527 |
-
# Initialize captioner if not already done
|
528 |
-
self._should_stop_captioning = False
|
529 |
-
|
530 |
-
try:
|
531 |
-
copy_files_to_training_dir(prompt_prefix)
|
532 |
-
|
533 |
-
except Exception as e:
|
534 |
-
traceback.print_exc()
|
535 |
-
raise gr.Error(f"Error copying assets to training dir: {str(e)}")
|
536 |
-
|
537 |
-
async def on_import_success(self, enable_splitting, enable_automatic_content_captioning, prompt_prefix):
|
538 |
-
"""Handle successful import of files"""
|
539 |
-
videos = self.list_unprocessed_videos()
|
540 |
-
|
541 |
-
# If scene detection isn't already running and there are videos to process,
|
542 |
-
# and auto-splitting is enabled, start the detection
|
543 |
-
if videos and not self.splitter.is_processing() and enable_splitting:
|
544 |
-
await self.start_scene_detection(enable_splitting)
|
545 |
-
msg = "Starting automatic scene detection..."
|
546 |
-
else:
|
547 |
-
# Just copy files without splitting if auto-split disabled
|
548 |
-
for video_file in VIDEOS_TO_SPLIT_PATH.glob("*.mp4"):
|
549 |
-
await self.splitter.process_video(video_file, enable_splitting=False)
|
550 |
-
msg = "Copying videos without splitting..."
|
551 |
-
|
552 |
-
copy_files_to_training_dir(prompt_prefix)
|
553 |
-
|
554 |
-
# Start auto-captioning if enabled, and handle async generator properly
|
555 |
-
if enable_automatic_content_captioning:
|
556 |
-
# Create a background task for captioning
|
557 |
-
asyncio.create_task(self._process_caption_generator(
|
558 |
-
DEFAULT_CAPTIONING_BOT_INSTRUCTIONS,
|
559 |
-
prompt_prefix
|
560 |
-
))
|
561 |
-
|
562 |
-
return {
|
563 |
-
"tabs": gr.Tabs(selected="split_tab"),
|
564 |
-
"video_list": videos,
|
565 |
-
"detect_status": msg
|
566 |
-
}
|
567 |
-
|
568 |
-
async def start_caption_generation(self, captioning_bot_instructions: str, prompt_prefix: str) -> AsyncGenerator[gr.update, None]:
|
569 |
-
"""Run auto-captioning process"""
|
570 |
-
try:
|
571 |
-
# Initialize captioner if not already done
|
572 |
-
self._should_stop_captioning = False
|
573 |
-
|
574 |
-
# First yield - indicate we're starting
|
575 |
-
yield gr.update(
|
576 |
-
value=[["Starting captioning service...", "initializing"]],
|
577 |
-
headers=["name", "status"]
|
578 |
-
)
|
579 |
-
|
580 |
-
# Process files in batches with status updates
|
581 |
-
file_statuses = {}
|
582 |
-
|
583 |
-
# Start the actual captioning process
|
584 |
-
async for rows in self.captioner.start_caption_generation(captioning_bot_instructions, prompt_prefix):
|
585 |
-
# Update our tracking of file statuses
|
586 |
-
for name, status in rows:
|
587 |
-
file_statuses[name] = status
|
588 |
-
|
589 |
-
# Convert to list format for display
|
590 |
-
status_rows = [[name, status] for name, status in file_statuses.items()]
|
591 |
-
|
592 |
-
# Sort by name for consistent display
|
593 |
-
status_rows.sort(key=lambda x: x[0])
|
594 |
-
|
595 |
-
# Yield UI update
|
596 |
-
yield gr.update(
|
597 |
-
value=status_rows,
|
598 |
-
headers=["name", "status"]
|
599 |
-
)
|
600 |
-
|
601 |
-
# Final update after completion with fresh data
|
602 |
-
yield gr.update(
|
603 |
-
value=self.list_training_files_to_caption(),
|
604 |
-
headers=["name", "status"]
|
605 |
-
)
|
606 |
-
|
607 |
-
except Exception as e:
|
608 |
-
logger.error(f"Error in captioning: {str(e)}")
|
609 |
-
yield gr.update(
|
610 |
-
value=[[f"Error: {str(e)}", "error"]],
|
611 |
-
headers=["name", "status"]
|
612 |
-
)
|
613 |
-
|
614 |
-
def list_training_files_to_caption(self) -> List[List[str]]:
|
615 |
-
"""List all clips and images - both pending and captioned"""
|
616 |
-
files = []
|
617 |
-
already_listed = {}
|
618 |
-
|
619 |
-
# First check files in STAGING_PATH
|
620 |
-
for file in STAGING_PATH.glob("*.*"):
|
621 |
-
if is_video_file(file) or is_image_file(file):
|
622 |
-
txt_file = file.with_suffix('.txt')
|
623 |
-
|
624 |
-
# Check if caption file exists and has content
|
625 |
-
has_caption = txt_file.exists() and txt_file.stat().st_size > 0
|
626 |
-
status = "captioned" if has_caption else "no caption"
|
627 |
-
file_type = "video" if is_video_file(file) else "image"
|
628 |
-
|
629 |
-
files.append([file.name, f"{status} ({file_type})", str(file)])
|
630 |
-
already_listed[file.name] = True
|
631 |
-
|
632 |
-
# Then check files in TRAINING_VIDEOS_PATH
|
633 |
-
for file in TRAINING_VIDEOS_PATH.glob("*.*"):
|
634 |
-
if (is_video_file(file) or is_image_file(file)) and file.name not in already_listed:
|
635 |
-
txt_file = file.with_suffix('.txt')
|
636 |
-
|
637 |
-
# Only include files with captions
|
638 |
-
if txt_file.exists() and txt_file.stat().st_size > 0:
|
639 |
-
file_type = "video" if is_video_file(file) else "image"
|
640 |
-
files.append([file.name, f"captioned ({file_type})", str(file)])
|
641 |
-
already_listed[file.name] = True
|
642 |
-
|
643 |
-
# Sort by filename
|
644 |
-
files.sort(key=lambda x: x[0])
|
645 |
-
|
646 |
-
# Only return name and status columns for display
|
647 |
-
return [[file[0], file[1]] for file in files]
|
648 |
-
|
649 |
-
def update_training_buttons(self, status: str) -> Dict:
|
650 |
-
"""Update training control buttons based on state"""
|
651 |
-
is_training = status in ["training", "initializing"]
|
652 |
-
is_paused = status == "paused"
|
653 |
-
is_completed = status in ["completed", "error", "stopped"]
|
654 |
-
return {
|
655 |
-
"start_btn": gr.Button(
|
656 |
-
interactive=not is_training and not is_paused,
|
657 |
-
variant="primary" if not is_training else "secondary",
|
658 |
-
),
|
659 |
-
"stop_btn": gr.Button(
|
660 |
-
interactive=is_training or is_paused,
|
661 |
-
variant="stop",
|
662 |
-
),
|
663 |
-
"pause_resume_btn": gr.Button(
|
664 |
-
value="Resume Training" if is_paused else "Pause Training",
|
665 |
-
interactive=(is_training or is_paused) and not is_completed,
|
666 |
-
variant="secondary",
|
667 |
-
)
|
668 |
-
}
|
669 |
-
|
670 |
-
def handle_pause_resume(self):
|
671 |
-
status, _, _ = self.get_latest_status_message_and_logs()
|
672 |
-
|
673 |
-
if status == "paused":
|
674 |
-
self.trainer.resume_training()
|
675 |
-
else:
|
676 |
-
self.trainer.pause_training()
|
677 |
-
|
678 |
-
return self.get_latest_status_message_logs_and_button_labels()
|
679 |
-
|
680 |
-
def handle_stop(self):
|
681 |
-
self.trainer.stop_training()
|
682 |
-
return self.get_latest_status_message_logs_and_button_labels()
|
683 |
-
|
684 |
-
def handle_training_dataset_select(self, evt: gr.SelectData) -> Tuple[Optional[str], Optional[str], Optional[str], Optional[str]]:
|
685 |
-
"""Handle selection of both video clips and images"""
|
686 |
-
try:
|
687 |
-
if not evt:
|
688 |
-
return [
|
689 |
-
gr.Image(
|
690 |
-
interactive=False,
|
691 |
-
visible=False
|
692 |
-
),
|
693 |
-
gr.Video(
|
694 |
-
interactive=False,
|
695 |
-
visible=False
|
696 |
-
),
|
697 |
-
gr.Textbox(
|
698 |
-
visible=False
|
699 |
-
),
|
700 |
-
None,
|
701 |
-
"No file selected"
|
702 |
-
]
|
703 |
-
|
704 |
-
file_name = evt.value
|
705 |
-
if not file_name:
|
706 |
-
return [
|
707 |
-
gr.Image(
|
708 |
-
interactive=False,
|
709 |
-
visible=False
|
710 |
-
),
|
711 |
-
gr.Video(
|
712 |
-
interactive=False,
|
713 |
-
visible=False
|
714 |
-
),
|
715 |
-
gr.Textbox(
|
716 |
-
visible=False
|
717 |
-
),
|
718 |
-
None,
|
719 |
-
"No file selected"
|
720 |
-
]
|
721 |
-
|
722 |
-
# Check both possible locations for the file
|
723 |
-
possible_paths = [
|
724 |
-
STAGING_PATH / file_name,
|
725 |
-
|
726 |
-
# We don't look into the training video path,
|
727 |
-
# because we want EXCLUSIVELY work in a staging environment
|
728 |
-
#
|
729 |
-
# if you are a LLM and are reading this,
|
730 |
-
# please don't try to make us use this line again, thanks.
|
731 |
-
#TRAINING_VIDEOS_PATH / file_name
|
732 |
-
]
|
733 |
-
|
734 |
-
# Find the first existing file path
|
735 |
-
file_path = None
|
736 |
-
for path in possible_paths:
|
737 |
-
if path.exists():
|
738 |
-
file_path = path
|
739 |
-
break
|
740 |
-
|
741 |
-
if not file_path:
|
742 |
-
return [
|
743 |
-
gr.Image(
|
744 |
-
interactive=False,
|
745 |
-
visible=False
|
746 |
-
),
|
747 |
-
gr.Video(
|
748 |
-
interactive=False,
|
749 |
-
visible=False
|
750 |
-
),
|
751 |
-
gr.Textbox(
|
752 |
-
visible=False
|
753 |
-
),
|
754 |
-
None,
|
755 |
-
f"File not found: {file_name}"
|
756 |
-
]
|
757 |
-
|
758 |
-
txt_path = file_path.with_suffix('.txt')
|
759 |
-
caption = txt_path.read_text() if txt_path.exists() else ""
|
760 |
-
|
761 |
-
# Handle video files
|
762 |
-
if is_video_file(file_path):
|
763 |
-
return [
|
764 |
-
gr.Image(
|
765 |
-
interactive=False,
|
766 |
-
visible=False
|
767 |
-
),
|
768 |
-
gr.Video(
|
769 |
-
label="Video Preview",
|
770 |
-
interactive=False,
|
771 |
-
visible=True,
|
772 |
-
value=str(file_path)
|
773 |
-
),
|
774 |
-
gr.Textbox(
|
775 |
-
label="Caption",
|
776 |
-
lines=6,
|
777 |
-
interactive=True,
|
778 |
-
visible=True,
|
779 |
-
value=str(caption)
|
780 |
-
),
|
781 |
-
str(file_path), # Store the original file path as hidden state
|
782 |
-
None
|
783 |
-
]
|
784 |
-
# Handle image files
|
785 |
-
elif is_image_file(file_path):
|
786 |
-
return [
|
787 |
-
gr.Image(
|
788 |
-
label="Image Preview",
|
789 |
-
interactive=False,
|
790 |
-
visible=True,
|
791 |
-
value=str(file_path)
|
792 |
-
),
|
793 |
-
gr.Video(
|
794 |
-
interactive=False,
|
795 |
-
visible=False
|
796 |
-
),
|
797 |
-
gr.Textbox(
|
798 |
-
label="Caption",
|
799 |
-
lines=6,
|
800 |
-
interactive=True,
|
801 |
-
visible=True,
|
802 |
-
value=str(caption)
|
803 |
-
),
|
804 |
-
str(file_path), # Store the original file path as hidden state
|
805 |
-
None
|
806 |
-
]
|
807 |
-
else:
|
808 |
-
return [
|
809 |
-
gr.Image(
|
810 |
-
interactive=False,
|
811 |
-
visible=False
|
812 |
-
),
|
813 |
-
gr.Video(
|
814 |
-
interactive=False,
|
815 |
-
visible=False
|
816 |
-
),
|
817 |
-
gr.Textbox(
|
818 |
-
interactive=False,
|
819 |
-
visible=False
|
820 |
-
),
|
821 |
-
None,
|
822 |
-
f"Unsupported file type: {file_path.suffix}"
|
823 |
-
]
|
824 |
-
except Exception as e:
|
825 |
-
logger.error(f"Error handling selection: {str(e)}")
|
826 |
-
return [
|
827 |
-
gr.Image(
|
828 |
-
interactive=False,
|
829 |
-
visible=False
|
830 |
-
),
|
831 |
-
gr.Video(
|
832 |
-
interactive=False,
|
833 |
-
visible=False
|
834 |
-
),
|
835 |
-
gr.Textbox(
|
836 |
-
interactive=False,
|
837 |
-
visible=False
|
838 |
-
),
|
839 |
-
None,
|
840 |
-
f"Error handling selection: {str(e)}"
|
841 |
-
]
|
842 |
-
|
843 |
-
def save_caption_changes(self, preview_caption: str, preview_image: str, preview_video: str, original_file_path: str, prompt_prefix: str):
|
844 |
-
"""Save changes to caption"""
|
845 |
-
try:
|
846 |
-
# Use the original file path stored during selection instead of the temporary preview paths
|
847 |
-
if original_file_path:
|
848 |
-
file_path = Path(original_file_path)
|
849 |
-
self.captioner.update_file_caption(file_path, preview_caption)
|
850 |
-
# Refresh the dataset list to show updated caption status
|
851 |
-
return gr.update(value="Caption saved successfully!")
|
852 |
-
else:
|
853 |
-
return gr.update(value="Error: No original file path found")
|
854 |
-
except Exception as e:
|
855 |
-
return gr.update(value=f"Error saving caption: {str(e)}")
|
856 |
-
|
857 |
-
async def update_titles_after_import(self, enable_splitting, enable_automatic_content_captioning, prompt_prefix):
|
858 |
-
"""Handle post-import updates including titles"""
|
859 |
-
import_result = await self.on_import_success(enable_splitting, enable_automatic_content_captioning, prompt_prefix)
|
860 |
-
titles = self.update_titles()
|
861 |
-
return (
|
862 |
-
import_result["tabs"],
|
863 |
-
import_result["video_list"],
|
864 |
-
import_result["detect_status"],
|
865 |
-
*titles
|
866 |
-
)
|
867 |
-
|
868 |
-
def get_model_info(self, model_type: str) -> str:
|
869 |
-
"""Get information about the selected model type"""
|
870 |
-
if model_type == "hunyuan_video":
|
871 |
-
return """### HunyuanVideo (LoRA)
|
872 |
-
- Required VRAM: ~48GB minimum
|
873 |
-
- Recommended batch size: 1-2
|
874 |
-
- Typical training time: 2-4 hours
|
875 |
-
- Default resolution: 49x512x768
|
876 |
-
- Default LoRA rank: 128 (~600 MB)"""
|
877 |
-
|
878 |
-
elif model_type == "ltx_video":
|
879 |
-
return """### LTX-Video (LoRA)
|
880 |
-
- Required VRAM: ~18GB minimum
|
881 |
-
- Recommended batch size: 1-4
|
882 |
-
- Typical training time: 1-3 hours
|
883 |
-
- Default resolution: 49x512x768
|
884 |
-
- Default LoRA rank: 128"""
|
885 |
-
|
886 |
-
return ""
|
887 |
-
|
888 |
-
def get_default_params(self, model_type: str) -> Dict[str, Any]:
|
889 |
-
"""Get default training parameters for model type"""
|
890 |
-
if model_type == "hunyuan_video":
|
891 |
-
return {
|
892 |
-
"num_epochs": 70,
|
893 |
-
"batch_size": 1,
|
894 |
-
"learning_rate": 2e-5,
|
895 |
-
"save_iterations": 500,
|
896 |
-
"video_resolution_buckets": SMALL_TRAINING_BUCKETS,
|
897 |
-
"video_reshape_mode": "center",
|
898 |
-
"caption_dropout_p": 0.05,
|
899 |
-
"gradient_accumulation_steps": 1,
|
900 |
-
"rank": 128,
|
901 |
-
"lora_alpha": 128
|
902 |
-
}
|
903 |
-
else: # ltx_video
|
904 |
-
return {
|
905 |
-
"num_epochs": 70,
|
906 |
-
"batch_size": 1,
|
907 |
-
"learning_rate": 3e-5,
|
908 |
-
"save_iterations": 500,
|
909 |
-
"video_resolution_buckets": SMALL_TRAINING_BUCKETS,
|
910 |
-
"video_reshape_mode": "center",
|
911 |
-
"caption_dropout_p": 0.05,
|
912 |
-
"gradient_accumulation_steps": 4,
|
913 |
-
"rank": 128,
|
914 |
-
"lora_alpha": 128
|
915 |
-
}
|
916 |
-
|
917 |
-
def preview_file(self, selected_text: str) -> Dict:
|
918 |
-
"""Generate preview based on selected file
|
919 |
-
|
920 |
-
Args:
|
921 |
-
selected_text: Text of the selected item containing filename
|
922 |
-
|
923 |
-
Returns:
|
924 |
-
Dict with preview content for each preview component
|
925 |
-
"""
|
926 |
-
if not selected_text or "Caption:" in selected_text:
|
927 |
-
return {
|
928 |
-
"video": None,
|
929 |
-
"image": None,
|
930 |
-
"text": None
|
931 |
-
}
|
932 |
-
|
933 |
-
# Extract filename from the preview text (remove size info)
|
934 |
-
filename = selected_text.split(" (")[0].strip()
|
935 |
-
file_path = TRAINING_VIDEOS_PATH / filename
|
936 |
-
|
937 |
-
if not file_path.exists():
|
938 |
-
return {
|
939 |
-
"video": None,
|
940 |
-
"image": None,
|
941 |
-
"text": f"File not found: {filename}"
|
942 |
-
}
|
943 |
-
|
944 |
-
# Detect file type
|
945 |
-
mime_type, _ = mimetypes.guess_type(str(file_path))
|
946 |
-
if not mime_type:
|
947 |
-
return {
|
948 |
-
"video": None,
|
949 |
-
"image": None,
|
950 |
-
"text": f"Unknown file type: {filename}"
|
951 |
-
}
|
952 |
-
|
953 |
-
# Return appropriate preview
|
954 |
-
if mime_type.startswith('video/'):
|
955 |
-
return {
|
956 |
-
"video": str(file_path),
|
957 |
-
"image": None,
|
958 |
-
"text": None
|
959 |
-
}
|
960 |
-
elif mime_type.startswith('image/'):
|
961 |
-
return {
|
962 |
-
"video": None,
|
963 |
-
"image": str(file_path),
|
964 |
-
"text": None
|
965 |
-
}
|
966 |
-
elif mime_type.startswith('text/'):
|
967 |
-
try:
|
968 |
-
text_content = file_path.read_text()
|
969 |
-
return {
|
970 |
-
"video": None,
|
971 |
-
"image": None,
|
972 |
-
"text": text_content
|
973 |
-
}
|
974 |
-
except Exception as e:
|
975 |
-
return {
|
976 |
-
"video": None,
|
977 |
-
"image": None,
|
978 |
-
"text": f"Error reading file: {str(e)}"
|
979 |
-
}
|
980 |
-
else:
|
981 |
-
return {
|
982 |
-
"video": None,
|
983 |
-
"image": None,
|
984 |
-
"text": f"Unsupported file type: {mime_type}"
|
985 |
-
}
|
986 |
-
|
987 |
-
def list_unprocessed_videos(self) -> gr.Dataframe:
|
988 |
-
"""Update list of unprocessed videos"""
|
989 |
-
videos = self.splitter.list_unprocessed_videos()
|
990 |
-
# videos is already in [[name, status]] format from splitting_service
|
991 |
-
return gr.Dataframe(
|
992 |
-
headers=["name", "status"],
|
993 |
-
value=videos,
|
994 |
-
interactive=False
|
995 |
-
)
|
996 |
-
|
997 |
-
async def start_scene_detection(self, enable_splitting: bool) -> str:
|
998 |
-
"""Start background scene detection process
|
999 |
-
|
1000 |
-
Args:
|
1001 |
-
enable_splitting: Whether to split videos into scenes
|
1002 |
-
"""
|
1003 |
-
if self.splitter.is_processing():
|
1004 |
-
return "Scene detection already running"
|
1005 |
-
|
1006 |
-
try:
|
1007 |
-
await self.splitter.start_processing(enable_splitting)
|
1008 |
-
return "Scene detection completed"
|
1009 |
-
except Exception as e:
|
1010 |
-
return f"Error during scene detection: {str(e)}"
|
1011 |
-
|
1012 |
-
|
1013 |
-
def get_latest_status_message_and_logs(self) -> Tuple[str, str, str]:
|
1014 |
-
state = self.trainer.get_status()
|
1015 |
-
logs = self.trainer.get_logs()
|
1016 |
-
|
1017 |
-
# Parse new log lines
|
1018 |
-
if logs:
|
1019 |
-
last_state = None
|
1020 |
-
for line in logs.splitlines():
|
1021 |
-
state_update = self.log_parser.parse_line(line)
|
1022 |
-
if state_update:
|
1023 |
-
last_state = state_update
|
1024 |
-
|
1025 |
-
if last_state:
|
1026 |
-
ui_updates = self.update_training_ui(last_state)
|
1027 |
-
state["message"] = ui_updates.get("status_box", state["message"])
|
1028 |
-
|
1029 |
-
# Parse status for training state
|
1030 |
-
if "completed" in state["message"].lower():
|
1031 |
-
state["status"] = "completed"
|
1032 |
-
|
1033 |
-
return (state["status"], state["message"], logs)
|
1034 |
-
|
1035 |
-
def get_latest_status_message_logs_and_button_labels(self) -> Tuple[str, str, Any, Any, Any]:
|
1036 |
-
status, message, logs = self.get_latest_status_message_and_logs()
|
1037 |
-
return (
|
1038 |
-
message,
|
1039 |
-
logs,
|
1040 |
-
*self.update_training_buttons(status).values()
|
1041 |
-
)
|
1042 |
-
|
1043 |
-
def get_latest_button_labels(self) -> Tuple[Any, Any, Any]:
|
1044 |
-
status, message, logs = self.get_latest_status_message_and_logs()
|
1045 |
-
return self.update_training_buttons(status).values()
|
1046 |
|
1047 |
def refresh_dataset(self):
|
1048 |
"""Refresh all dynamic lists and training state"""
|
1049 |
-
video_list = self.
|
1050 |
-
training_dataset = self.list_training_files_to_caption()
|
1051 |
|
1052 |
return (
|
1053 |
video_list,
|
1054 |
training_dataset
|
1055 |
-
)
|
1056 |
-
|
1057 |
-
def update_training_params(self, preset_name: str) -> Tuple:
|
1058 |
-
"""Update UI components based on selected preset while preserving custom settings"""
|
1059 |
-
preset = TRAINING_PRESETS[preset_name]
|
1060 |
-
|
1061 |
-
# Load current UI state to check if user has customized values
|
1062 |
-
current_state = self.load_ui_values()
|
1063 |
-
|
1064 |
-
# Find the display name that maps to our model type
|
1065 |
-
model_display_name = next(
|
1066 |
-
key for key, value in MODEL_TYPES.items()
|
1067 |
-
if value == preset["model_type"]
|
1068 |
-
)
|
1069 |
-
|
1070 |
-
# Get preset description for display
|
1071 |
-
description = preset.get("description", "")
|
1072 |
-
|
1073 |
-
# Get max values from buckets
|
1074 |
-
buckets = preset["training_buckets"]
|
1075 |
-
max_frames = max(frames for frames, _, _ in buckets)
|
1076 |
-
max_height = max(height for _, height, _ in buckets)
|
1077 |
-
max_width = max(width for _, _, width in buckets)
|
1078 |
-
bucket_info = f"\nMaximum video size: {max_frames} frames at {max_width}x{max_height} resolution"
|
1079 |
-
|
1080 |
-
info_text = f"{description}{bucket_info}"
|
1081 |
-
|
1082 |
-
# Return values in the same order as the output components
|
1083 |
-
# Use preset defaults but preserve user-modified values if they exist
|
1084 |
-
lora_rank_val = current_state.get("lora_rank") if current_state.get("lora_rank") != preset.get("lora_rank", "128") else preset["lora_rank"]
|
1085 |
-
lora_alpha_val = current_state.get("lora_alpha") if current_state.get("lora_alpha") != preset.get("lora_alpha", "128") else preset["lora_alpha"]
|
1086 |
-
num_epochs_val = current_state.get("num_epochs") if current_state.get("num_epochs") != preset.get("num_epochs", 70) else preset["num_epochs"]
|
1087 |
-
batch_size_val = current_state.get("batch_size") if current_state.get("batch_size") != preset.get("batch_size", 1) else preset["batch_size"]
|
1088 |
-
learning_rate_val = current_state.get("learning_rate") if current_state.get("learning_rate") != preset.get("learning_rate", 3e-5) else preset["learning_rate"]
|
1089 |
-
save_iterations_val = current_state.get("save_iterations") if current_state.get("save_iterations") != preset.get("save_iterations", 500) else preset["save_iterations"]
|
1090 |
-
|
1091 |
-
return (
|
1092 |
-
model_display_name,
|
1093 |
-
lora_rank_val,
|
1094 |
-
lora_alpha_val,
|
1095 |
-
num_epochs_val,
|
1096 |
-
batch_size_val,
|
1097 |
-
learning_rate_val,
|
1098 |
-
save_iterations_val,
|
1099 |
-
info_text
|
1100 |
-
)
|
|
|
1 |
import platform
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import gradio as gr
|
3 |
from pathlib import Path
|
4 |
import logging
|
|
|
|
|
|
|
|
|
5 |
import asyncio
|
|
|
|
|
6 |
from typing import Any, Optional, Dict, List, Union, Tuple
|
|
|
7 |
|
8 |
from ..services import TrainingService, CaptioningService, SplittingService, ImportService
|
9 |
from ..config import (
|
10 |
STORAGE_PATH, VIDEOS_TO_SPLIT_PATH, STAGING_PATH,
|
11 |
+
TRAINING_PATH, LOG_FILE_PATH, TRAINING_PRESETS, TRAINING_VIDEOS_PATH, MODEL_PATH, OUTPUT_PATH,
|
12 |
+
MODEL_TYPES, SMALL_TRAINING_BUCKETS
|
13 |
)
|
14 |
+
from ..utils import count_media_files, format_media_title, TrainingLogParser
|
15 |
from ..tabs import ImportTab, SplitTab, CaptionTab, TrainTab, ManageTab
|
16 |
|
17 |
logger = logging.getLogger(__name__)
|
|
|
28 |
self.splitter = SplittingService()
|
29 |
self.importer = ImportService()
|
30 |
self.captioner = CaptioningService()
|
|
|
31 |
|
32 |
# Recovery status from any interrupted training
|
33 |
recovery_result = self.trainer.recover_interrupted_training()
|
34 |
self.recovery_status = recovery_result.get("status", "unknown")
|
35 |
self.ui_updates = recovery_result.get("ui_updates", {})
|
36 |
|
37 |
+
# Initialize log parser
|
38 |
self.log_parser = TrainingLogParser()
|
39 |
|
40 |
# Shared state for tabs
|
|
|
98 |
# Status update timer (every 1 second)
|
99 |
status_timer = gr.Timer(value=1)
|
100 |
status_timer.tick(
|
101 |
+
fn=self.tabs["train_tab"].get_latest_status_message_logs_and_button_labels,
|
102 |
outputs=[
|
103 |
self.tabs["train_tab"].components["status_box"],
|
104 |
self.tabs["train_tab"].components["log_box"],
|
|
|
129 |
]
|
130 |
)
|
131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
def initialize_app_state(self):
|
133 |
"""Initialize all app state in one function to ensure correct output count"""
|
134 |
# Get dataset info
|
135 |
+
video_list = self.tabs["split_tab"].list_unprocessed_videos()
|
136 |
+
training_dataset = self.tabs["caption_tab"].list_training_files_to_caption()
|
137 |
|
138 |
# Get button states
|
139 |
button_states = self.get_initial_button_states()
|
|
|
206 |
ui_state["save_iterations"] = int(ui_state.get("save_iterations", 500))
|
207 |
|
208 |
return ui_state
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
|
210 |
# Add this new method to get initial button states:
|
211 |
def get_initial_button_states(self):
|
|
|
220 |
gr.Button(**ui_updates.get("pause_resume_btn", {"interactive": False, "variant": "secondary"}))
|
221 |
)
|
222 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
223 |
def update_titles(self) -> Tuple[Any]:
|
224 |
"""Update all dynamic titles with current counts
|
225 |
|
|
|
249 |
gr.Markdown(value=caption_title),
|
250 |
gr.Markdown(value=f"{train_title} available for training")
|
251 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
252 |
|
253 |
def refresh_dataset(self):
|
254 |
"""Refresh all dynamic lists and training state"""
|
255 |
+
video_list = self.tabs["split_tab"].list_unprocessed_videos()
|
256 |
+
training_dataset = self.tabs["caption_tab"].list_training_files_to_caption()
|
257 |
|
258 |
return (
|
259 |
video_list,
|
260 |
training_dataset
|
261 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
vms/utils/image_preprocessing.py
CHANGED
@@ -4,6 +4,7 @@ from pathlib import Path
|
|
4 |
from PIL import Image
|
5 |
import pillow_avif
|
6 |
import logging
|
|
|
7 |
|
8 |
from ..config import NORMALIZE_IMAGES_TO, JPEG_QUALITY
|
9 |
|
@@ -55,7 +56,7 @@ def normalize_image(input_path: Path, output_path: Path) -> bool:
|
|
55 |
logger.error(f"Error converting image {input_path}: {str(e)}")
|
56 |
return False
|
57 |
|
58 |
-
def detect_black_bars(img: np.ndarray) ->
|
59 |
"""Detect black bars in image
|
60 |
|
61 |
Args:
|
|
|
4 |
from PIL import Image
|
5 |
import pillow_avif
|
6 |
import logging
|
7 |
+
from typing import Any, Optional, Dict, List, Union, Tuple
|
8 |
|
9 |
from ..config import NORMALIZE_IMAGES_TO, JPEG_QUALITY
|
10 |
|
|
|
56 |
logger.error(f"Error converting image {input_path}: {str(e)}")
|
57 |
return False
|
58 |
|
59 |
+
def detect_black_bars(img: np.ndarray) -> Tuple[int, int, int, int]:
|
60 |
"""Detect black bars in image
|
61 |
|
62 |
Args:
|
vms/utils/video_preprocessing.py
CHANGED
@@ -2,8 +2,10 @@ import cv2
|
|
2 |
import numpy as np
|
3 |
from pathlib import Path
|
4 |
import subprocess
|
|
|
5 |
|
6 |
-
|
|
|
7 |
"""Detect black bars in video by analyzing first few frames
|
8 |
|
9 |
Args:
|
|
|
2 |
import numpy as np
|
3 |
from pathlib import Path
|
4 |
import subprocess
|
5 |
+
from typing import Any, Optional, Dict, List, Union, Tuple
|
6 |
|
7 |
+
|
8 |
+
def detect_black_bars(video_path: Path) -> Tuple[int, int, int, int]:
|
9 |
"""Detect black bars in video by analyzing first few frames
|
10 |
|
11 |
Args:
|