File size: 14,469 Bytes
7275aef |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 |
# cli/config_wizard.py
import json
import time
from pathlib import Path
from rich.console import Console
from rich.prompt import Prompt, Confirm
from rich.table import Table
from rich.panel import Panel
from rich import print
import sys
import os
import inquirer
# Add the parent directory to the path so we can import from utils
sys.path.insert(0, str(Path(__file__).parent.parent))
from utils.device import get_system_info
from utils.dataset_loader import list_local_datasets
from humigence_datasets.download_datasets import get_available_datasets, ensure_demo_dataset
console = Console()
def show_system_info():
"""Display system information as shown in the screenshots"""
console.print("\n[bold cyan]System Information[/bold cyan]")
info = get_system_info()
table = Table(show_header=False, box=None)
table.add_column(style="cyan", width=15)
table.add_column(style="white")
table.add_row("Platform:", info["Platform"])
table.add_row("Python:", info["Python Version"])
table.add_row("PyTorch:", info["Torch Version"])
table.add_row("CUDA:", info["CUDA Version"] if info["CUDA Available"] else "Not available")
table.add_row("RAM:", info["RAM"])
table.add_row("CPUs:", str(info["CPUs"]))
table.add_row("GPUs detected:", str(info["GPU Count"]))
console.print(table)
def choose_setup_mode():
"""Choose between Basic and Advanced setup"""
console.print("\n[bold red]Fine-tuning Configuration[/bold red]")
questions = [
inquirer.List('setup_mode',
message="? Choose Setup Mode: (Use arrow keys)",
choices=[
"Basic Setup - Essential configuration only",
"Advanced Setup - Full control over all parameters"
],
default="Basic Setup - Essential configuration only")
]
answers = inquirer.prompt(questions)
return "basic" if "Basic" in answers['setup_mode'] else "advanced"
def show_gpu_info():
"""Display available GPU information without prompting for selection"""
info = get_system_info()
if info["GPU Count"] == 0:
console.print("\n[yellow]β οΈ No GPUs detected - CPU training mode[/yellow]")
return {
"gpu_count": 0,
"gpus": []
}
# Display available GPUs
console.print("\n[bold cyan]π₯οΈ Available GPUs:[/bold cyan]")
gpu_table = Table(show_header=True, box=None)
gpu_table.add_column("Index", style="cyan", width=6)
gpu_table.add_column("Name", style="white", width=40)
gpu_table.add_column("VRAM", style="green", width=10)
for i, gpu in enumerate(info["GPUs"]):
gpu_table.add_row(str(i), gpu['name'], gpu['memory'])
console.print(gpu_table)
console.print("[dim]π‘ GPU selection will be made at training launch[/dim]")
return {
"gpu_count": info["GPU Count"],
"gpus": info["GPUs"]
}
def choose_base_model():
"""Choose base model"""
models = [
("Qwen/Qwen2.5-0.5B", "Qwen/Qwen2.5-0.5B (77M) - Local model"),
("microsoft/Phi-2", "microsoft/Phi-2 (839M) - Local model"),
("TinyLlama/TinyLlama-1.1B-Chat-v1.0", "TinyLlama/TinyLlama-1.1B-Chat-v1.0 (369M) - Local model")
]
questions = [
inquirer.List('base_model',
message="Choose base model: (Use arrow keys)",
choices=[display_name for _, display_name in models],
default=models[0][1])
]
answers = inquirer.prompt(questions)
selected_display = answers['base_model']
# Find the actual model name (without the extra text)
for model_name, display_name in models:
if display_name == selected_display:
return model_name
return models[0][0] # fallback
def choose_training_recipe():
"""Choose training recipe"""
recipes = [
"QLoRA (4-bit NF4)",
"QLoRA (4-bit FP4)",
"LoRA (FP16)",
"LoRA (BF16)"
]
questions = [
inquirer.List('training_recipe',
message="? Choose training recipe: (Use arrow keys)",
choices=recipes,
default=recipes[0])
]
answers = inquirer.prompt(questions)
return answers['training_recipe']
# Removed choose_training_mode() - GPU selection now handled in choose_gpu_configuration()
def choose_evaluation_method():
"""Choose evaluation method"""
console.print("\n[bold red]Evaluation Method[/bold red]")
console.print("[dim]Choose how to evaluate your model after training:[/dim]")
questions = [
inquirer.List('evaluation_method',
message="? Choose Evaluation Method: (Use arrow keys)",
choices=[
("Standard Evaluation - Fast, may have device issues", "standard"),
("Atomic Evaluation - True process isolation, guaranteed success π", "atomic")
])
]
answers = inquirer.prompt(questions)
return answers['evaluation_method']
def choose_dataset():
"""Choose dataset from available local datasets"""
# Get available datasets by scanning ~/humigence_data/
available_datasets = list_local_datasets()
if not available_datasets:
console.print("[yellow]β οΈ No datasets found in ~/humigence_data[/yellow]")
console.print("[cyan]You can create a dataset by placing a .jsonl file in ~/humigence_data/[/cyan]")
console.print("[cyan]Each line should be a JSON object with 'instruction' and 'output' fields.[/cyan]")
console.print("\n[blue]Example format:[/blue]")
console.print('{"instruction": "What is machine learning?", "output": "Machine learning is..."}')
# Still offer custom path and dummy data options
console.print("\n[yellow]Available fallback options:[/yellow]")
choices = ["Enter custom path", "Use dummy data for testing"]
questions = [
inquirer.List('dataset',
message="Choose fallback option: (Use arrow keys)",
choices=choices,
default=choices[0])
]
answers = inquirer.prompt(questions)
selected_display = answers['dataset']
if selected_display == "Enter custom path":
custom_path = Prompt.ask("Enter custom dataset path")
return custom_path
else: # Use dummy data
# Create a minimal dummy dataset
dummy_data = [
{"instruction": "Hello", "output": "Hi there!"},
{"instruction": "How are you?", "output": "I'm doing well, thank you!"}
]
dummy_path = Path.home() / "humigence_data" / "dummy.jsonl"
dummy_path.parent.mkdir(parents=True, exist_ok=True)
import json
with open(dummy_path, 'w') as f:
for item in dummy_data:
f.write(json.dumps(item) + '\n')
console.print(f"[green]β
Created dummy dataset at {dummy_path}[/green]")
return str(dummy_path)
# Create choices for inquirer with sample counts
choices = []
dataset_map = {}
for name, path, count in available_datasets:
if count == "?":
display_name = f"{name} (unknown samples)"
else:
# Format count with commas for readability
formatted_count = f"{count:,}" if isinstance(count, int) else str(count)
display_name = f"{name} ({formatted_count} samples)"
choices.append(display_name)
dataset_map[display_name] = path
# Add custom path option
choices.append("Enter custom path")
dataset_map["Enter custom path"] = "custom"
questions = [
inquirer.List('dataset',
message="Choose dataset: (Use arrow keys)",
choices=choices,
default=choices[0])
]
answers = inquirer.prompt(questions)
selected_display = answers['dataset']
if selected_display == "Enter custom path":
custom_path = Prompt.ask("Enter custom dataset path")
return custom_path
else:
return dataset_map[selected_display]
def show_config_summary(config):
"""Show configuration summary and ask for confirmation"""
console.print("\n[bold cyan]Configuration Summary[/bold cyan]")
# Display key configuration parameters
table = Table(show_header=False, box=None)
table.add_column(style="cyan", width=25)
table.add_column(style="white")
table.add_row("use_flash_attn:", "True")
table.add_row("lora:", "True")
table.add_row("qlora:", "True" if "QLoRA" in config["training_recipe"] else "False")
table.add_row("quant_bits:", "None")
table.add_row("lora_r:", "16")
table.add_row("lora_alpha:", "32")
table.add_row("lora_dropout:", "0.05")
table.add_row("per_device_train_batch_size:", "2")
table.add_row("gradient_accumulation_steps:", "4")
table.add_row("learning_rate:", "0.0002")
table.add_row("num_train_epochs:", "1.0")
table.add_row("fp16:", "True")
table.add_row("bf16:", "False")
gpu_count = config.get("gpu_info", {}).get("gpu_count", 0)
if gpu_count == 0:
table.add_row("gpu_mode:", "CPU Training (No GPUs detected)")
table.add_row("gpu_ids:", "N/A")
elif gpu_count == 1:
gpu_name = config.get("gpu_info", {}).get("gpus", [{}])[0].get("name", "Unknown")
table.add_row("gpu_mode:", f"Single GPU (will auto-select GPU 0: {gpu_name})")
table.add_row("gpu_ids:", "Will be [0]")
else:
table.add_row("gpu_mode:", f"Multi-GPU or Single GPU (will prompt from {gpu_count} available GPUs)")
table.add_row("gpu_ids:", "Will be selected at training launch")
table.add_row("dataset_path:", config["dataset_path"])
table.add_row("data_schema:", "instruction_output")
table.add_row("demo_mode:", str(config.get("demo_mode", False)))
table.add_row("max_samples:", str(config.get("max_samples", "None (full dataset)")))
table.add_row("train_val_test_split:", str(config.get("train_val_test_split", [0.8, 0.1, 0.1])))
table.add_row("split_seed:", str(config.get("split_seed", 42)))
table.add_row("eval_single_gpu:", str(config.get("eval_single_gpu", True)))
table.add_row("eval_gpu_index:", str(config.get("eval_gpu_index", 0)))
table.add_row("eval_batch_size:", str(config.get("eval_batch_size", 8)))
table.add_row("num_workers:", str(config.get("num_workers", 4)))
table.add_row("pin_memory:", str(config.get("pin_memory", True)))
table.add_row("max_seq_len:", "1024")
table.add_row("processed_dir:", "data/processed")
console.print(table)
console.print(f"\n[bold green]β Configuration saved to: ./runs/humigence/config.snapshot.json[/bold green]")
def collect_training_config():
"""Main function to collect all training configuration interactively"""
try:
# Step 1: Setup Mode
setup_mode = choose_setup_mode()
# Step 2: System Information
show_system_info()
# Step 3: GPU Information (no selection yet)
gpu_info = show_gpu_info()
# Step 4: Base Model Selection
base_model = choose_base_model()
# Step 5: Training Recipe
training_recipe = choose_training_recipe()
# Step 6: Dataset Selection
dataset_path = choose_dataset()
# Build configuration with new invariants
config = {
"setup_mode": setup_mode,
"gpu_info": gpu_info, # Store GPU info for later use
"base_model": base_model,
"training_recipe": training_recipe,
"dataset_path": dataset_path,
# A) Config & UX invariants
"demo_mode": False, # No silent limits unless demo mode
"max_samples": None, # null => use 100% of dataset
"train_val_test_split": [0.8, 0.1, 0.1], # deterministic split ratios
"split_seed": 42, # deterministic splitting
"eval_single_gpu": True, # hard-isolate eval to single GPU
"eval_gpu_index": 0, # which GPU to use for evaluation
"eval_batch_size": 8, # evaluation batch size
"num_workers": 4, # DataLoader workers
"pin_memory": True, # DataLoader pin_memory
# Legacy compatibility
"split_ratios": [0.8, 0.1, 0.1], # alias for train_val_test_split
"min_train_samples": 1000,
"min_val_samples": 100,
"min_test_samples": 100,
"min_tokens_per_sample": 50,
"random_seed": 42, # alias for split_seed
"learning_rate": "2e-4",
"num_train_epochs": "1",
"gradient_accumulation_steps": "4",
"logging_steps": "10",
"save_steps": "100",
"timestamp": time.strftime("%Y-%m-%dT%H:%M:%S")
}
# Step 7: Configuration Summary
show_config_summary(config)
# Step 8: Confirmation
questions = [
inquirer.Confirm('proceed',
message="? Proceed with training? (Y/n)",
default=True)
]
answers = inquirer.prompt(questions)
proceed = answers['proceed']
if not proceed:
console.print("[bold red]β Training aborted β configuration not confirmed.[/bold red]")
return None
# Save configuration with migration and validation
from config_migration import save_config_snapshot
config_path = "runs/humigence/config.snapshot.json"
validated_config = save_config_snapshot(config, config_path)
console.print(f"[bold green]β
Configuration saved to {config_path}[/bold green]")
return str(config_path)
except KeyboardInterrupt:
console.print("\n[bold red]β Training aborted β configuration not confirmed.[/bold red]")
return None
except Exception as e:
console.print(f"[bold red]β Error during configuration: {e}[/bold red]")
return None
|