File size: 6,803 Bytes
1f3f75e
 
65f8095
 
 
 
1f3f75e
65f8095
 
 
 
1f3f75e
65f8095
1f3f75e
 
 
 
 
65f8095
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1f3f75e
65f8095
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1f3f75e
 
 
 
 
 
 
 
65f8095
 
 
1f3f75e
65f8095
 
1f3f75e
65f8095
1f3f75e
65f8095
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
import os
import uuid
import yaml
import json
import shutil
import torch
from pathlib import Path
from PIL import Image
from fastapi import FastAPI
from fastapi.responses import JSONResponse
from huggingface_hub import hf_hub_download, whoami

# ========== CONFIGURATION ==========
REPO_ID = "rahul7star/ohamlab"
FOLDER_IN_REPO = "filter-demo/upload_20250708_041329_9c5c81"
CONCEPT_SENTENCE = "ohamlab style"
LORA_NAME = "ohami_filter_autorun"

# ========== FASTAPI APP ==========
app = FastAPI()

# ========== HELPERS ==========
def create_dataset(images, *captions):
    destination_folder = f"datasets_{uuid.uuid4()}"
    os.makedirs(destination_folder, exist_ok=True)

    jsonl_file_path = os.path.join(destination_folder, "metadata.jsonl")
    with open(jsonl_file_path, "a") as jsonl_file:
        for index, image in enumerate(images):
            new_image_path = shutil.copy(str(image), destination_folder)
            caption = captions[index]
            file_name = os.path.basename(new_image_path)
            data = {"file_name": file_name, "prompt": caption}
            jsonl_file.write(json.dumps(data) + "\n")

    return destination_folder

def recursive_update(d, u):
    for k, v in u.items():
        if isinstance(v, dict) and v:
            d[k] = recursive_update(d.get(k, {}), v)
        else:
            d[k] = v
    return d

def start_training(
    lora_name,
    concept_sentence,
    steps,
    lr,
    rank,
    model_to_train,
    low_vram,
    dataset_folder,
    sample_1,
    sample_2,
    sample_3,
    use_more_advanced_options,
    more_advanced_options,
):
    try:
        user = whoami()
        username = user.get("name", "anonymous")
        push_to_hub = True
    except:
        username = "anonymous"
        push_to_hub = False

    slugged_lora_name = lora_name.replace(" ", "_").lower()

    # Load base config
    config = {
        "config": {
            "name": slugged_lora_name,
            "process": [
                {
                    "model": {
                        "low_vram": low_vram,
                        "is_flux": True,
                        "quantize": True,
                        "name_or_path": "black-forest-labs/FLUX.1-dev"
                    },
                    "network": {
                        "linear": rank,
                        "linear_alpha": rank,
                        "type": "lora"
                    },
                    "train": {
                        "steps": steps,
                        "lr": lr,
                        "skip_first_sample": True,
                        "batch_size": 1,
                        "dtype": "bf16",
                        "gradient_accumulation_steps": 1,
                        "gradient_checkpointing": True,
                        "noise_scheduler": "flowmatch",
                        "optimizer": "adamw8bit",
                        "ema_config": {
                            "use_ema": True,
                            "ema_decay": 0.99
                        }
                    },
                    "datasets": [
                        {"folder_path": dataset_folder}
                    ],
                    "save": {
                        "dtype": "float16",
                        "save_every": 10000,
                        "push_to_hub": push_to_hub,
                        "hf_repo_id": f"{username}/{slugged_lora_name}",
                        "hf_private": True,
                        "max_step_saves_to_keep": 4
                    },
                    "sample": {
                        "guidance_scale": 3.5,
                        "sample_every": steps,
                        "sample_steps": 28,
                        "width": 1024,
                        "height": 1024,
                        "walk_seed": True,
                        "seed": 42,
                        "sampler": "flowmatch",
                        "prompts": [p for p in [sample_1, sample_2, sample_3] if p]
                    },
                    "trigger_word": concept_sentence
                }
            ]
        }
    }

    # Apply advanced YAML overrides if any
    if use_more_advanced_options and more_advanced_options:
        advanced_config = yaml.safe_load(more_advanced_options)
        config["config"]["process"][0] = recursive_update(config["config"]["process"][0], advanced_config)

    # Save YAML config
    os.makedirs("tmp_configs", exist_ok=True)
    config_path = f"tmp_configs/{uuid.uuid4()}_{slugged_lora_name}.yaml"
    with open(config_path, "w") as f:
        yaml.dump(config, f)

    # Simulate training
    print(f"[INFO] Starting training with config: {config_path}")
    print(json.dumps(config, indent=2))
    return f"Training started successfully with config: {config_path}"

# ========== MAIN ENDPOINT ==========
@app.post("/train-from-hf")
def auto_run_lora_from_repo():
    try:
        local_dir = Path(f"/tmp/{LORA_NAME}-{uuid.uuid4()}")
        os.makedirs(local_dir, exist_ok=True)

        hf_hub_download(
            repo_id=REPO_ID,
            repo_type="dataset",
            subfolder=FOLDER_IN_REPO,
            local_dir=local_dir,
            local_dir_use_symlinks=False,
            force_download=False,
            etag_timeout=10,
            allow_patterns=["*.jpg", "*.png", "*.jpeg"],
        )

        image_dir = local_dir / FOLDER_IN_REPO
        image_paths = list(image_dir.rglob("*.jpg")) + list(image_dir.rglob("*.jpeg")) + list(image_dir.rglob("*.png"))

        if not image_paths:
            return JSONResponse(status_code=400, content={"error": "No images found in the HF repo folder."})

        captions = [
            f"Autogenerated caption for {img.stem} in the {CONCEPT_SENTENCE} [trigger]" for img in image_paths
        ]

        dataset_path = create_dataset(image_paths, *captions)

        result = start_training(
            lora_name=LORA_NAME,
            concept_sentence=CONCEPT_SENTENCE,
            steps=1000,
            lr=4e-4,
            rank=16,
            model_to_train="dev",
            low_vram=True,
            dataset_folder=dataset_path,
            sample_1=f"A stylized portrait using {CONCEPT_SENTENCE}",
            sample_2=f"A cat in the {CONCEPT_SENTENCE}",
            sample_3=f"A selfie processed in {CONCEPT_SENTENCE}",
            use_more_advanced_options=True,
            more_advanced_options="""
training:
  seed: 42
  precision: bf16
  batch_size: 2
augmentation:
  flip: true
  color_jitter: true
"""
        )

        return {"message": result}

    except Exception as e:
        return JSONResponse(status_code=500, content={"error": str(e)})

# ========== FASTAPI RUNNER ==========
if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)