SmolFactory / scripts /dataset_tonic /setup_hf_dataset.py
Tonic's picture
adds single token logic read/write , adds gpt-oss demo space , adds spaces refactor , adds new version of track tonic , adds logic in launch.sh
75bcdb3
#!/usr/bin/env python3
"""
Setup script for Hugging Face Dataset repository for Trackio experiments
"""
import os
import sys
import json
import time
from datetime import datetime
from pathlib import Path
from datasets import Dataset
from typing import Optional, Dict, Any
from huggingface_hub import HfApi, create_repo
import subprocess
def get_username_from_token(token: str) -> Optional[str]:
"""
Get username from HF token using the API.
Args:
token (str): Hugging Face token
Returns:
Optional[str]: Username if successful, None otherwise
"""
try:
# Create API client with token directly
api = HfApi(token=token)
# Get user info
user_info = api.whoami()
username = user_info.get("name", user_info.get("username"))
return username
except Exception as e:
print(f"❌ Error getting username from token: {e}")
return None
def create_dataset_repository(username: str, dataset_name: str = "trackio-experiments", token: str = None) -> str:
"""
Create a dataset repository on Hugging Face.
Args:
username (str): HF username
dataset_name (str): Name for the dataset repository
token (str): HF token for authentication
Returns:
str: Full repository name (username/dataset_name)
"""
repo_id = f"{username}/{dataset_name}"
try:
# Create the dataset repository
create_repo(
repo_id=repo_id,
repo_type="dataset",
token=token,
exist_ok=True,
private=False # Public dataset for easier sharing
)
print(f"✅ Successfully created dataset repository: {repo_id}")
return repo_id
except Exception as e:
if "already exists" in str(e).lower():
print(f"ℹ️ Dataset repository already exists: {repo_id}")
return repo_id
else:
print(f"❌ Error creating dataset repository: {e}")
return None
def setup_trackio_dataset(dataset_name: str = None, token: str = None) -> bool:
"""
Set up Trackio dataset repository automatically.
Args:
dataset_name (str): Optional custom dataset name (default: trackio-experiments)
token (str): HF token for authentication
Returns:
bool: True if successful, False otherwise
"""
print("🚀 Setting up Trackio Dataset Repository")
print("=" * 50)
# Get token from parameter, environment, or command line
if not token:
token = os.environ.get('HUGGING_FACE_HUB_TOKEN') or os.environ.get('HF_TOKEN')
# If no token in environment, try command line argument
if not token and len(sys.argv) > 1:
token = sys.argv[1]
if not token:
print("❌ No HF token found. Please set HUGGING_FACE_HUB_TOKEN environment variable or provide as argument.")
return False
# Get username from token
print("🔍 Getting username from token...")
username = get_username_from_token(token)
if not username:
print("❌ Could not determine username from token. Please check your token.")
return False
print(f"✅ Authenticated as: {username}")
# Use provided dataset name or default
if not dataset_name:
dataset_name = "trackio-experiments"
# Create dataset repository
print(f"🔧 Creating dataset repository: {username}/{dataset_name}")
repo_id = create_dataset_repository(username, dataset_name, token)
if not repo_id:
print("❌ Failed to create dataset repository")
return False
# Set environment variable for other scripts
os.environ['TRACKIO_DATASET_REPO'] = repo_id
print(f"✅ Set TRACKIO_DATASET_REPO={repo_id}")
# Add initial experiment data
print("📊 Adding initial experiment data...")
if add_initial_experiment_data(repo_id, token):
print("✅ Successfully added initial experiment data")
else:
print("⚠️ Could not add initial experiment data (this is optional)")
# Add dataset README
print("📝 Adding dataset README...")
if add_dataset_readme(repo_id, token):
print("✅ Successfully added dataset README")
else:
print("⚠️ Could not add dataset README (this is optional)")
print(f"\n🎉 Dataset setup complete!")
print(f"📊 Dataset URL: https://huggingface.co/datasets/{repo_id}")
print(f"🔧 Repository ID: {repo_id}")
return True
def add_initial_experiment_data(repo_id: str, token: str = None) -> bool:
"""
Add initial experiment data to the dataset using data preservation.
Args:
repo_id (str): Dataset repository ID
token (str): HF token for authentication
Returns:
bool: True if successful, False otherwise
"""
try:
# Get token from parameter or environment
if not token:
token = os.environ.get('HUGGING_FACE_HUB_TOKEN') or os.environ.get('HF_TOKEN')
if not token:
print("⚠️ No token available for uploading data")
return False
# Import dataset manager
import sys
sys.path.insert(0, os.path.join(os.path.dirname(__file__), '..', '..', 'src'))
from dataset_utils import TrackioDatasetManager
# Initialize dataset manager
dataset_manager = TrackioDatasetManager(repo_id, token)
# Check if dataset already has data
existing_experiments = dataset_manager.load_existing_experiments()
if existing_experiments:
print(f"ℹ️ Dataset already contains {len(existing_experiments)} experiments, preserving existing data")
# Initial experiment data
initial_experiment = {
'experiment_id': f'exp_demo_{datetime.now().strftime("%Y%m%d_%H%M%S")}',
'name': 'smollm3-finetune-demo',
'description': 'SmolLM3 fine-tuning experiment demo with comprehensive metrics tracking',
'created_at': datetime.now().isoformat(),
'status': 'completed',
'metrics': json.dumps([
{
'timestamp': datetime.now().isoformat(),
'step': 100,
'metrics': {
'loss': 1.15,
'grad_norm': 10.5,
'learning_rate': 5e-6,
'num_tokens': 1000000.0,
'mean_token_accuracy': 0.76,
'epoch': 0.1,
'total_tokens': 1000000.0,
'throughput': 2000000.0,
'step_time': 0.5,
'batch_size': 2,
'seq_len': 4096,
'token_acc': 0.76,
'gpu_memory_allocated': 15.2,
'gpu_memory_reserved': 70.1,
'gpu_utilization': 85.2,
'cpu_percent': 2.7,
'memory_percent': 10.1
}
}
]),
'parameters': json.dumps({
'model_name': 'HuggingFaceTB/SmolLM3-3B',
'max_seq_length': 4096,
'batch_size': 2,
'learning_rate': 5e-6,
'epochs': 3,
'dataset': 'OpenHermes-FR',
'trainer_type': 'SFTTrainer',
'hardware': 'GPU (H100/A100)',
'mixed_precision': True,
'gradient_checkpointing': True,
'flash_attention': True
}),
'artifacts': json.dumps([]),
'logs': json.dumps([
{
'timestamp': datetime.now().isoformat(),
'level': 'INFO',
'message': 'Training started successfully'
},
{
'timestamp': datetime.now().isoformat(),
'level': 'INFO',
'message': 'Model loaded and configured'
},
{
'timestamp': datetime.now().isoformat(),
'level': 'INFO',
'message': 'Dataset loaded and preprocessed'
}
]),
'last_updated': datetime.now().isoformat()
}
# Use dataset manager to safely add the experiment
success = dataset_manager.upsert_experiment(initial_experiment)
if success:
print(f"✅ Successfully added initial experiment data to {repo_id}")
final_count = len(dataset_manager.load_existing_experiments())
print(f"📊 Dataset now contains {final_count} total experiments")
else:
print(f"❌ Failed to add initial experiment data to {repo_id}")
return False
# Add README template
add_dataset_readme(repo_id, token)
return True
except Exception as e:
print(f"⚠️ Could not add initial experiment data: {e}")
return False
def add_dataset_readme(repo_id: str, token: str) -> bool:
"""
Add README template to the dataset repository.
Args:
repo_id (str): Dataset repository ID
token (str): HF token
Returns:
bool: True if successful, False otherwise
"""
try:
# Read the README template
template_path = os.path.join(os.path.dirname(__file__), '..', '..', 'templates', 'datasets', 'readme.md')
if os.path.exists(template_path):
with open(template_path, 'r', encoding='utf-8') as f:
readme_content = f.read()
else:
# Create a basic README if template doesn't exist
readme_content = f"""---
dataset_info:
features:
- name: experiment_id
dtype: string
- name: name
dtype: string
- name: description
dtype: string
- name: created_at
dtype: string
- name: status
dtype: string
- name: metrics
dtype: string
- name: parameters
dtype: string
- name: artifacts
dtype: string
- name: logs
dtype: string
- name: last_updated
dtype: string
tags:
- trackio
- experiment tracking
- smollm3
- fine-tuning
---
# Trackio Experiments Dataset
This dataset stores experiment tracking data for ML training runs, particularly focused on SmolLM3 fine-tuning experiments with comprehensive metrics tracking.
## Dataset Structure
The dataset contains the following columns:
- **experiment_id**: Unique identifier for each experiment
- **name**: Human-readable name for the experiment
- **description**: Detailed description of the experiment
- **created_at**: Timestamp when the experiment was created
- **status**: Current status (running, completed, failed, paused)
- **metrics**: JSON string containing training metrics over time
- **parameters**: JSON string containing experiment configuration
- **artifacts**: JSON string containing experiment artifacts
- **logs**: JSON string containing experiment logs
- **last_updated**: Timestamp of last update
## Usage
This dataset is automatically used by the Trackio monitoring system to store and retrieve experiment data. It provides persistent storage for experiment tracking across different training runs.
## Integration
The dataset is used by:
- Trackio Spaces for experiment visualization
- Training scripts for logging metrics and parameters
- Monitoring systems for experiment tracking
- SmolLM3 fine-tuning pipeline for comprehensive metrics capture
## Privacy
This dataset is public by default for easier sharing and collaboration. Only non-sensitive experiment data is stored.
## Examples
### Sample Experiment Entry
```json
{{
"experiment_id": "exp_20250720_130853",
"name": "smollm3-finetune-demo",
"description": "SmolLM3 fine-tuning experiment demo",
"created_at": "2025-07-20T13:08:53",
"status": "completed",
"metrics": "{{...}}",
"parameters": "{{...}}",
"artifacts": "[]",
"logs": "{{...}}",
"last_updated": "2025-07-20T13:08:53"
}}
```
This dataset is maintained by the Trackio monitoring system and automatically updated during training runs.
"""
# Upload README to the dataset repository
from huggingface_hub import upload_file
# Create a temporary file with the README content
import tempfile
with tempfile.NamedTemporaryFile(mode='w', suffix='.md', delete=False, encoding='utf-8') as f:
f.write(readme_content)
temp_file = f.name
try:
upload_file(
path_or_fileobj=temp_file,
path_in_repo="README.md",
repo_id=repo_id,
repo_type="dataset",
token=token,
commit_message="Add dataset README"
)
print(f"✅ Successfully added README to {repo_id}")
return True
finally:
# Clean up temporary file
if os.path.exists(temp_file):
os.unlink(temp_file)
except Exception as e:
print(f"⚠️ Could not add README to dataset: {e}")
return False
def main():
"""Main function to set up the dataset."""
# Get token from environment first
token = os.environ.get('HUGGING_FACE_HUB_TOKEN') or os.environ.get('HF_TOKEN')
# If no token in environment, try command line argument
if not token and len(sys.argv) > 1:
token = sys.argv[1]
if not token:
print("❌ No HF token found. Please set HUGGING_FACE_HUB_TOKEN environment variable or provide as argument.")
sys.exit(1)
# Get dataset name from command line or use default
dataset_name = None
if len(sys.argv) > 2:
dataset_name = sys.argv[2]
# Pass token to setup function
success = setup_trackio_dataset(dataset_name, token)
sys.exit(0 if success else 1)
if __name__ == "__main__":
main()