File size: 6,258 Bytes
d4a00b2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/usr/bin/env python3
"""
Script para configurar reposit贸rios e pipeline de treinamento no HuggingFace.
Cria:
- Reposit贸rio de dataset
- Reposit贸rio de modelo (LoRA)
- Espa莽o de treinamento (se necess谩rio)

Autentica莽茫o: defina HF_TOKEN no ambiente ou use `huggingface-cli login`.
Nunca commite tokens no reposit贸rio.
"""

import json
import os
import sys
from pathlib import Path

from huggingface_hub import HfApi, create_repo, login, whoami

HF_USER = os.environ.get("HF_USER", "beAnalytic")
DATASET_REPO = os.environ.get("DATASET_REPO", f"{HF_USER}/eda-training-dataset")
MODEL_REPO = os.environ.get("OUTPUT_REPO", f"{HF_USER}/eda-llm-qwen2.5-lora")
TRAINING_SPACE = os.environ.get("TRAINING_SPACE", f"{HF_USER}/Training")


def _token() -> str | None:
    return os.environ.get("HF_TOKEN")


def print_step(message: str, status: str = "info"):
    """Print formatado para diferentes tipos de mensagens"""
    colors = {
        "info": "\033[0;34m",
        "success": "\033[0;32m",
        "warning": "\033[1;33m",
        "error": "\033[0;31m",
    }
    symbols = {
        "info": "i",
        "success": "ok",
        "warning": "!",
        "error": "x",
    }
    reset = "\033[0m"
    print(f"{colors.get(status, colors['info'])}[{symbols.get(status, 'i')}] {message}{reset}")


def main():
    print_step("Configurando HuggingFace - BeAnalytic", "info")
    print()
    
    hf_token = _token()
    if not hf_token:
        print_step(
            "HF_TOKEN nao definido. Exporte HF_TOKEN ou execute: huggingface-cli login",
            "error",
        )
        sys.exit(1)

    print_step("Autenticando no HuggingFace...", "info")
    try:
        login(token=hf_token, add_to_git_credential=False)
        user_info = whoami()
        print_step(f"Autenticado como: {user_info['name']}", "success")
        if user_info["name"] != HF_USER:
            print_step(
                f"Aviso: usuario esperado '{HF_USER}' mas autenticado como '{user_info['name']}'",
                "warning",
            )
    except Exception as e:
        print_step(f"Erro ao autenticar: {e}", "error")
        sys.exit(1)
    
    api = HfApi(token=hf_token)
    
    print()
    print_step(f"Criando repositorio de dataset: {DATASET_REPO}", "info")
    try:
        create_repo(
            repo_id=DATASET_REPO,
            repo_type="dataset",
            token=hf_token,
            exist_ok=True,
            private=False,
        )
        print_step(f"Dataset: {DATASET_REPO}", "success")
    except Exception as e:
        print_step(f"Erro ao criar dataset: {e}", "error")
        if "already exists" not in str(e).lower():
            sys.exit(1)
        print_step("Repositorio ja existe, continuando...", "warning")
    
    print()
    print_step(f"Criando repositorio de modelo (LoRA): {MODEL_REPO}", "info")
    try:
        create_repo(
            repo_id=MODEL_REPO,
            repo_type="model",
            token=hf_token,
            exist_ok=True,
            private=False,
        )
        print_step(f"Modelo: {MODEL_REPO}", "success")
    except Exception as e:
        print_step(f"Erro ao criar modelo: {e}", "error")
        if "already exists" not in str(e).lower():
            sys.exit(1)
        print_step("Repositorio ja existe, continuando...", "warning")
    
    print()
    print_step(f"Criando espaco de treinamento: {TRAINING_SPACE}", "info")
    try:
        create_repo(
            repo_id=TRAINING_SPACE,
            repo_type="space",
            token=hf_token,
            exist_ok=True,
            private=False,
            space_sdk="docker",
        )
        print_step(f"Space: {TRAINING_SPACE}", "success")
    except Exception as e:
        print_step(f"Erro ao criar space: {e}", "error")
        if "already exists" not in str(e).lower():
            print_step("Continuando sem space...", "warning")
        else:
            print_step("Space ja existe, continuando...", "warning")
    
    print()
    print_step("Atualizando arquivos de configuracao locais...", "info")
    
    script_dir = Path(__file__).parent
    
    config_file = script_dir.parent / "training_config.json"
    if config_file.exists():
        with open(config_file, encoding="utf-8") as f:
            config = json.load(f)
        
        config["dataset"] = DATASET_REPO
        config["output_dir"] = MODEL_REPO
        
        with open(config_file, "w", encoding="utf-8") as f:
            json.dump(config, f, indent=2)
        print_step(f"Atualizado: {config_file.name}", "success")
    
    dockerfile = script_dir.parent / "Dockerfile"
    if dockerfile.exists():
        content = dockerfile.read_text(encoding="utf-8")
        content = content.replace("DATASET_REPO=amarorn/eda-training-dataset", f"DATASET_REPO={DATASET_REPO}")
        content = content.replace("OUTPUT_REPO=amarorn/eda-llm-model", f"OUTPUT_REPO={MODEL_REPO}")
        dockerfile.write_text(content, encoding="utf-8")
        print_step(f"Atualizado: {dockerfile.name}", "success")
    
    env_example = script_dir / ".env.example"
    env_content = f"""# Copie para .env e preencha (nao commite .env)
# HF_TOKEN=obtenha em https://huggingface.co/settings/tokens
HF_USER={HF_USER}
DATASET_REPO={DATASET_REPO}
OUTPUT_REPO={MODEL_REPO}
MODEL_NAME=Qwen/Qwen2.5-1.5B-Instruct

NUM_EPOCHS=3
BATCH_SIZE=4
LEARNING_RATE=3e-05
"""
    env_example.write_text(env_content, encoding="utf-8")
    print_step(f"Atualizado: {env_example.name}", "success")
    
    print()
    print_step("=" * 60, "info")
    print_step("Configuracao concluida.", "success")
    print_step("=" * 60, "info")
    print()
    print("Repositorios:")
    print(f"   - Dataset:  https://huggingface.co/datasets/{DATASET_REPO}")
    print(f"   - Modelo:   https://huggingface.co/{MODEL_REPO}")
    print(f"   - Training: https://huggingface.co/spaces/{TRAINING_SPACE}")
    print()
    print("Registro auditavel: ml/configs/huggingface_training_config/hf_registry.json")
    print()
    print("Proximos passos:")
    print("   1. Upload do dataset para o repositorio de dataset")
    print("   2. No Space, secrets: HF_TOKEN, DATASET_REPO, OUTPUT_REPO, MODEL_NAME")
    print("   3. Executar treinamento")
    print()


if __name__ == "__main__":
    main()