File size: 1,963 Bytes
d010ca8 |
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 |
# /// script
# dependencies = [
# "trl>=0.12.0",
# "peft>=0.7.0",
# "transformers>=4.36.0",
# "accelerate>=0.24.0",
# "trackio",
# ]
# ///
import trackio
from datasets import load_dataset
from peft import LoraConfig
from trl import SFTTrainer, SFTConfig
# Initialize Trackio for real-time monitoring
trackio.init(
project="qwen-demo-sft",
space_id="evalstate/demo-trackio-dashboard",
config={
"model": "Qwen/Qwen2.5-0.5B",
"dataset": "trl-lib/Capybara",
"examples": 50,
"max_steps": 20,
"note": "Quick demo training"
}
)
# Load dataset (only 50 examples for quick demo)
dataset = load_dataset("trl-lib/Capybara", split="train[:50]")
print(f"✅ Dataset loaded: {len(dataset)} examples")
# Training configuration
config = SFTConfig(
# Hub settings - CRITICAL for saving results
output_dir="qwen-demo-sft",
push_to_hub=True,
hub_model_id="evalstate/qwen-demo-sft",
# Quick training settings
max_steps=20, # Very short for demo
per_device_train_batch_size=2,
gradient_accumulation_steps=2,
learning_rate=2e-5,
# Logging
logging_steps=5,
save_strategy="steps",
save_steps=10,
# Monitoring
report_to="trackio",
)
# LoRA configuration (memory efficient)
peft_config = LoraConfig(
r=16,
lora_alpha=32,
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
target_modules=["q_proj", "v_proj"],
)
# Initialize and train
trainer = SFTTrainer(
model="Qwen/Qwen2.5-0.5B",
train_dataset=dataset,
args=config,
peft_config=peft_config,
)
print("🚀 Starting demo training...")
trainer.train()
print("💾 Pushing to Hub...")
trainer.push_to_hub()
# Finish Trackio tracking
trackio.finish()
print("✅ Demo complete!")
print(f"📦 Model: https://huggingface.co/evalstate/qwen-demo-sft")
print(f"📊 Metrics: https://huggingface.co/spaces/evalstate/demo-trackio-dashboard")
|