File size: 4,191 Bytes
b6a400a | 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 | #!/usr/bin/env python3
# /// script
# dependencies = [
# "trl>=0.12.0",
# "peft>=0.7.0",
# "transformers>=4.36.0",
# "accelerate>=0.24.0",
# "trackio",
# ]
# ///
from datasets import load_dataset
from peft import LoraConfig
from trl import SFTTrainer, SFTConfig
import trackio
import os
print("🚀 Medium-Scale SFT Training with Trackio")
print("=" * 60)
# Initialize Trackio with Space sync
print("\n📊 Initializing Trackio...")
trackio.init(
project="medium-sft-training",
space_id="evalstate/trl-trackio-dashboard",
config={
"model": "Qwen/Qwen2.5-0.5B",
"dataset": "trl-lib/Capybara",
"dataset_size": 1000,
"num_epochs": 3,
"learning_rate": 2e-5,
"batch_size": 4,
"gradient_accumulation": 4,
"lora_r": 16,
"lora_alpha": 32,
"hardware": "a10g-large",
}
)
print("✅ Trackio initialized!")
print("📈 Dashboard: https://huggingface.co/spaces/evalstate/trl-trackio-dashboard")
# Load dataset - 1000 examples
print("\n📊 Loading dataset...")
dataset = load_dataset("trl-lib/Capybara", split="train[:1000]")
print(f"✅ Dataset loaded: {len(dataset)} examples")
# Get username
username = os.environ.get("HF_USERNAME", "evalstate")
# Training configuration - production settings
print("\n⚙️ Configuring training...")
config = SFTConfig(
# Output and Hub settings
output_dir="qwen-capybara-medium",
push_to_hub=True,
hub_model_id=f"{username}/qwen-capybara-medium",
hub_strategy="every_save", # Push all checkpoints
# Training parameters - 3 epochs on 1K examples
num_train_epochs=3,
per_device_train_batch_size=4,
gradient_accumulation_steps=4, # Effective batch size = 16
# Learning rate and schedule
learning_rate=2e-5,
warmup_ratio=0.1,
lr_scheduler_type="cosine",
# Logging and checkpointing
logging_steps=10, # Log every 10 steps
save_strategy="steps",
save_steps=50, # Save every 50 steps
save_total_limit=3, # Keep only 3 latest checkpoints
# Evaluation
eval_strategy="steps",
eval_steps=50,
# Optimization
bf16=True, # Use bfloat16 for A10G
gradient_checkpointing=True, # Save memory
# Trackio monitoring
report_to="trackio",
)
# LoRA configuration - larger than demo
print("🔧 Setting up LoRA (r=16)...")
peft_config = LoraConfig(
r=16,
lora_alpha=32,
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
target_modules=["q_proj", "k_proj", "v_proj", "o_proj"], # More modules
)
# Create eval split
print("\n🔀 Creating train/eval split...")
dataset_split = dataset.train_test_split(test_size=0.1, seed=42)
train_dataset = dataset_split["train"]
eval_dataset = dataset_split["test"]
print(f" Train: {len(train_dataset)} examples")
print(f" Eval: {len(eval_dataset)} examples")
# Initialize trainer
print("\n🎯 Initializing trainer...")
trainer = SFTTrainer(
model="Qwen/Qwen2.5-0.5B",
train_dataset=train_dataset,
eval_dataset=eval_dataset,
args=config,
peft_config=peft_config,
)
# Calculate training info
total_steps = (len(train_dataset) // (4 * 4)) * 3 # samples / (batch * grad_accum) * epochs
print(f"\n📊 Training Info:")
print(f" Total steps: ~{total_steps}")
print(f" Epochs: 3")
print(f" Effective batch size: 16")
print(f" Expected time: ~45-60 minutes")
print(f" Checkpoints saved every 50 steps")
# Train!
print("\n🏃 Starting training...")
print("📈 Watch live metrics: https://huggingface.co/spaces/evalstate/trl-trackio-dashboard")
print("-" * 60)
trainer.train()
# Save to Hub
print("\n💾 Pushing final model to Hub...")
trainer.push_to_hub()
# Finish Trackio
print("\n📊 Finalizing Trackio metrics...")
trackio.finish()
print("\n" + "=" * 60)
print("✅ Training complete!")
print(f"📦 Model: https://huggingface.co/{username}/qwen-capybara-medium")
print(f"📊 Metrics: https://huggingface.co/spaces/evalstate/trl-trackio-dashboard")
print(f"💡 Try the model with:")
print(f' from transformers import pipeline')
print(f' generator = pipeline("text-generation", model="{username}/qwen-capybara-medium")')
print("=" * 60)
|