-
Notifications
You must be signed in to change notification settings - Fork 2
/
eden_trainer.py
179 lines (147 loc) · 5.82 KB
/
eden_trainer.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
from datetime import datetime
start_time = datetime.utcnow()
import logging
import sys
import argparse
import eden_utils
from main import *
from bson import ObjectId
from eden_utils import tasks_collection
logging.basicConfig(
level=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s',
stream=sys.stdout
)
def main():
parser = argparse.ArgumentParser(description='Training script for flux network.')
parser.add_argument('--task_id', help="Eden task ID")
parser.add_argument('--env', type=str, default="STAGE", choices=["STAGE", "PROD"], help='Environment')
parser.add_argument('--config', type=str, default="template/train_config.json", help='Path to the training config file (JSON).')
args = parser.parse_args()
# Get task
task = tasks_collection.find_one({"_id": ObjectId(args.task_id)})
if not task:
raise ValueError(f"Task {args.task_id} not found!")
try:
wait_time = (start_time - task["createdAt"]).total_seconds()
# Mark task status running
tasks_collection.update_one(
{"_id": ObjectId(args.task_id)},
{"$set": {
"status": "running",
"performance": {
"waitTime": wait_time,
},
"updatedAt": datetime.utcnow(),
}}
)
# Get task args
task_args = task["args"]
print("task_args", task_args)
# Override args
with open(args.config, 'r') as f:
config_json = json.load(f)
config_json["lora_rank"] = str(task_args["lora_rank"])
config_json["learning_rate"] = str(task_args["learning_rate"])
config_json["seed"] = str(task_args.get("seed", random.randint(0, 2147483648)))
config_json["max_train_steps"] = str(task_args["max_train_steps"])
config_json["caption_prefix"] = task_args.get("caption_prefix", config_json.get("caption_prefix", "TOK"))
with open(args.config, 'w') as f:
json.dump(config_json, f, indent=2)
# Load the training config from the provided file
config = construct_config(args.config)
print(" ========= Config !!! ========== ")
print(config)
print(" ========================== ")
# Download the dataset from the URL provided
lora_training_urls = task_args["lora_training_urls"]
download_dataset(config["dataset_path"], lora_training_urls)
# Use GPT4v to check if the dataset is a face or style
config["mode"] = "style"
try:
if eden_utils.check_if_face(config["dataset_path"]):
config["mode"] = "face"
except Exception as e:
print("GPT error, assuming mode=style. Error: ", e)
# Preprocess the dataset if required
if config.get("prep_dataset"):
prep_dataset(config["dataset_path"], hard_prep=True)
# Perform dataset captioning if enabled in the config
if config.get("caption_mode"):
florence_caption_dataset(config["dataset_path"], caption_mode=config["caption_mode"])
# Step 5: Construct and run the training command
cmd = construct_train_command(config)
print(" ========= Config 2 !!! ========== ")
print(config)
print(" ========================== ")
run_job(cmd, config)
# save the result
output_dir = config["output_dir"]
output_name = config["output_name"]
# upload to eden
file_url, _ = eden_utils.upload_file(
f"{output_dir}/{output_name}.safetensors",
env=args.env
)
print("file_url", file_url)
# make thumbnail and slug
sample_dir = os.path.join(output_dir, "sample")
thumbnail_url = eden_utils.create_thumbnail(sample_dir, env=args.env)
# slug = eden_utils.make_slug(task)
# save model
model_id = eden_utils.models_collection.insert_one({
"args": task_args,
"checkpoint": file_url,
"base_model": "flux-dev",
"name": task_args["name"],
"public": False,
"task": task["_id"],
"thumbnail": thumbnail_url,
# "slug": slug,
"user": task["user"],
"createdAt": datetime.utcnow(),
"updatedAt": datetime.utcnow(),
}).inserted_id
print("saved model_id", model_id)
finish_time = datetime.utcnow()
run_time = (finish_time - start_time).total_seconds()
# Mark task status completed
tasks_collection.update_one(
{"_id": ObjectId(args.task_id)},
{"$set": {
"status": "completed",
"performance": {
"waitTime": wait_time,
"runTime": run_time,
},
"result": [{
"filename": file_url.split("/")[-1],
"metadata": config,
"mediaAttributes": {
"mimeType": "application/zip"
},
"thumbnail": thumbnail_url,
"model": model_id
}],
"updatedAt": datetime.utcnow(),
}}
)
except Exception as e:
logging.error(f"Error: {e}")
print("Error: ", e)
finish_time = datetime.utcnow()
run_time = (finish_time - start_time).total_seconds()
tasks_collection.update_one(
{"_id": ObjectId(args.task_id)},
{"$set": {
"status": "failed",
"error": str(e),
"performance": {
"waitTime": wait_time,
"runTime": run_time,
},
"updatedAt": datetime.utcnow(),
}}
)
if __name__ == "__main__":
main()