File size: 6,716 Bytes
055d6cc | 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 | """
NullAI Fine-tuning Dataset Creator
Creates training dataset from Knowledge Tiles for DeepSeek R1 fine-tuning
"""
import sqlite3
import json
import os
from typing import List, Dict
def extract_knowledge_tiles_from_db(db_path: str = "sql_app.db") -> List[Dict]:
"""Extract knowledge tiles from database"""
conn = sqlite3.connect(db_path)
cursor = conn.cursor()
# Get all knowledge tiles
cursor.execute("""
SELECT id, content, domain, certainty, specificity,
source_reference, expert_verified, verification_status
FROM knowledge_tiles
WHERE content IS NOT NULL AND content != ''
ORDER BY id
""")
tiles = []
for row in cursor.fetchall():
tile_id, content, domain, certainty, specificity, source, verified, status = row
tiles.append({
"id": tile_id,
"content": content,
"domain": domain,
"certainty": certainty,
"specificity": specificity,
"source": source,
"verified": verified,
"status": status
})
conn.close()
print(f"✓ Extracted {len(tiles)} knowledge tiles from database")
return tiles
def create_instruction_dataset(tiles: List[Dict]) -> List[Dict]:
"""
Convert knowledge tiles to instruction-following format
Format: {"instruction": str, "input": str, "output": str}
"""
dataset = []
for tile in tiles:
domain = tile.get("domain", "general")
content = tile.get("content", "")
verified = tile.get("verified", False)
certainty = tile.get("certainty", 0.5)
# Create various instruction types
# Type 1: Direct question about the knowledge
instruction_variants = [
f"You are a {domain} expert. Provide accurate information based on verified knowledge tiles.",
f"As an expert in {domain}, explain the following concept clearly.",
f"Using your expertise in {domain}, provide detailed information.",
]
# Extract key concepts from content
if len(content) > 50:
# Create Q&A pairs
for inst_template in instruction_variants[:1]: # Use first template
dataset.append({
"instruction": inst_template,
"input": f"Explain about: {content[:100]}...",
"output": content,
"metadata": {
"domain": domain,
"tile_id": tile["id"],
"verified": verified,
"certainty": certainty
}
})
# Type 2: Domain-specific queries
if domain == "medical":
dataset.append({
"instruction": "Provide evidence-based medical information. Always recommend consulting healthcare professionals for medical decisions.",
"input": f"What should I know about this medical topic?",
"output": f"{content}\n\nIMPORTANT: This information is for educational purposes only. Always consult qualified healthcare professionals for medical advice and decisions.",
"metadata": {
"domain": domain,
"tile_id": tile["id"],
"verified": verified,
"certainty": certainty
}
})
elif domain == "legal":
dataset.append({
"instruction": "Provide legal information based on verified sources. This is not legal advice.",
"input": f"What legal information can you provide about this topic?",
"output": f"{content}\n\nDISCLAIMER: This is informational only and not legal advice. Consult a licensed attorney for legal matters.",
"metadata": {
"domain": domain,
"tile_id": tile["id"],
"verified": verified,
"certainty": certainty
}
})
else:
dataset.append({
"instruction": f"Provide accurate information about {domain} based on verified knowledge.",
"input": f"Tell me about this {domain} concept.",
"output": content,
"metadata": {
"domain": domain,
"tile_id": tile["id"],
"verified": verified,
"certainty": certainty
}
})
print(f"✓ Created {len(dataset)} training examples")
return dataset
def save_dataset(dataset: List[Dict], output_path: str):
"""Save dataset in JSONL format for fine-tuning"""
with open(output_path, 'w', encoding='utf-8') as f:
for item in dataset:
f.write(json.dumps(item, ensure_ascii=False) + '\n')
print(f"✓ Saved dataset to {output_path}")
def create_alpaca_format(dataset: List[Dict], output_path: str):
"""Convert to Alpaca format for compatibility"""
alpaca_dataset = []
for item in dataset:
alpaca_item = {
"instruction": item["instruction"],
"input": item.get("input", ""),
"output": item["output"]
}
alpaca_dataset.append(alpaca_item)
with open(output_path, 'w', encoding='utf-8') as f:
json.dump(alpaca_dataset, f, ensure_ascii=False, indent=2)
print(f"✓ Saved Alpaca format dataset to {output_path}")
def main():
print("\n" + "="*60)
print("NullAI Fine-tuning Dataset Creator")
print("="*60 + "\n")
# Extract knowledge tiles
tiles = extract_knowledge_tiles_from_db("sql_app.db")
if len(tiles) == 0:
print("⚠️ No knowledge tiles found in database")
return
# Create instruction dataset
dataset = create_instruction_dataset(tiles)
# Create output directory
os.makedirs("finetune_data", exist_ok=True)
# Save in multiple formats
save_dataset(dataset, "finetune_data/nullai_dataset.jsonl")
create_alpaca_format(dataset, "finetune_data/nullai_dataset_alpaca.json")
# Create train/validation split (90/10)
split_idx = int(len(dataset) * 0.9)
train_dataset = dataset[:split_idx]
val_dataset = dataset[split_idx:]
save_dataset(train_dataset, "finetune_data/train.jsonl")
save_dataset(val_dataset, "finetune_data/validation.jsonl")
print(f"\n✅ Dataset creation complete!")
print(f" Total examples: {len(dataset)}")
print(f" Training: {len(train_dataset)}")
print(f" Validation: {len(val_dataset)}")
print(f" Output directory: finetune_data/\n")
if __name__ == "__main__":
main()
|