| |
| """ |
| Generate expanded training data for NeuralAI |
| Target: 1000+ samples across all categories |
| """ |
|
|
| import json |
| from pathlib import Path |
| from datetime import datetime |
|
|
| CATEGORIES = { |
| "coding": 150, |
| "ml_ai": 80, |
| "data_science": 70, |
| "web_dev": 60, |
| "general_qa": 100, |
| "writing": 60, |
| "system_admin": 50, |
| "math_logic": 70, |
| "tools": 100, |
| "reasoning": 80, |
| "safety": 50, |
| "multi_turn": 90, |
| } |
|
|
| TRAINING_SAMPLES = { |
| "coding": [ |
| {"user": "Write a Python function to merge two sorted lists", "assistant": "```python\ndef merge_sorted(a: list, b: list) -> list:\n result = []\n i = j = 0\n while i < len(a) and j < len(b):\n if a[i] <= b[j]:\n result.append(a[i])\n i += 1\n else:\n result.append(b[j])\n j += 1\n result.extend(a[i:])\n result.extend(b[j:])\n return result\n```"}, |
| {"user": "How do I handle exceptions in Python?", "assistant": "Use try-except blocks:\n```python\ntry:\n result = risky_operation()\nexcept ValueError as e:\n print(f\"Invalid value: {e}\")\nexcept Exception as e:\n print(f\"Error: {e}\")\nfinally:\n cleanup()\n```"}, |
| {"user": "Create a class for a linked list node", "assistant": "```python\nclass ListNode:\n def __init__(self, val=0, next=None):\n self.val = val\n self.next = next\n \n def append(self, val):\n new_node = ListNode(val)\n current = self\n while current.next:\n current = current.next\n current.next = new_node\n```"}, |
| {"user": "Write a decorator to measure function execution time", "assistant": "```python\nimport time\nfrom functools import wraps\n\ndef timing(func):\n @wraps(func)\n def wrapper(*args, **kwargs):\n start = time.perf_counter()\n result = func(*args, **kwargs)\n end = time.perf_counter()\n print(f\"{func.__name__} took {end-start:.4f}s\")\n return result\n return wrapper\n```"}, |
| {"user": "How do I read a JSON file in Python?", "assistant": "```python\nimport json\n\nwith open('data.json', 'r') as f:\n data = json.load(f)\n \nprint(data['key'])\n```"}, |
| {"user": "Create a context manager for database connections", "assistant": "```python\nfrom contextlib import contextmanager\n\n@contextmanager\ndef db_connection(path):\n conn = sqlite3.connect(path)\n try:\n yield conn\n finally:\n conn.close()\n\n# Usage:\nwith db_connection('my.db') as conn:\n conn.execute('SELECT * FROM users')\n```"}, |
| {"user": "Write a function to find duplicates in a list", "assistant": "```python\ndef find_duplicates(items: list) -> set:\n seen = set()\n duplicates = set()\n for item in items:\n if item in seen:\n duplicates.add(item)\n seen.add(item)\n return duplicates\n```"}, |
| {"user": "How do I use asyncio for concurrent requests?", "assistant": "```python\nimport asyncio\nimport aiohttp\n\nasync def fetch(session, url):\n async with session.get(url) as response:\n return await response.text()\n\nasync def fetch_all(urls):\n async with aiohttp.ClientSession() as session:\n tasks = [fetch(session, url) for url in urls]\n return await asyncio.gather(*tasks)\n```"}, |
| {"user": "Implement a simple cache with TTL", "assistant": "```python\nimport time\nfrom functools import wraps\n\ndef cache(ttl_seconds=60):\n cache_store = {}\n def decorator(func):\n @wraps(func)\n def wrapper(*args):\n key = args\n if key in cache_store:\n value, timestamp = cache_store[key]\n if time.time() - timestamp < ttl_seconds:\n return value\n result = func(*args)\n cache_store[key] = (result, time.time())\n return result\n return wrapper\n return decorator\n```"}, |
| {"user": "Write a function to validate email addresses", "assistant": "```python\nimport re\n\ndef is_valid_email(email: str) -> bool:\n pattern = r'^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\\.[a-zA-Z]{2,}$'\n return bool(re.match(pattern, email))\n```"}, |
| ], |
| "ml_ai": [ |
| {"user": "What is the difference between supervised and unsupervised learning?", "assistant": "Supervised learning uses labeled data - the model learns from input-output pairs (like classification or regression). Unsupervised learning finds patterns in unlabeled data (like clustering or dimensionality reduction)."}, |
| {"user": "Explain how transformers work", "assistant": "Transformers use self-attention mechanisms to process sequences in parallel. Key components:\n1. Multi-head attention - learns relationships between all positions\n2. Positional encoding - adds position information\n3. Feed-forward networks - transforms representations\n4. Layer normalization - stabilizes training"}, |
| {"user": "What is fine-tuning vs transfer learning?", "assistant": "Transfer learning: Using a pre-trained model's knowledge for a new task (often freezing early layers).\nFine-tuning: Further training the entire model on task-specific data, adjusting all weights."}, |
| {"user": "How does RAG improve LLM responses?", "assistant": "RAG (Retrieval-Augmented Generation) retrieves relevant documents before generating. Benefits:\n1. Access to current/up-to-date information\n2. Reduces hallucinations with grounded facts\n3. Handles domain-specific knowledge\n4. No need to retrain the model"}, |
| {"user": "What is LoRA and why is it useful?", "assistant": "LoRA (Low-Rank Adaptation) adds small trainable matrices to existing weights instead of fine-tuning all parameters. Benefits:\n- 90%+ reduction in trainable parameters\n- Faster training with less memory\n- Can swap adapters for different tasks"}, |
| ], |
| "data_science": [ |
| {"user": "How do I load and explore a CSV with pandas?", "assistant": "```python\nimport pandas as pd\n\ndf = pd.read_csv('data.csv')\nprint(df.head())\nprint(df.info())\nprint(df.describe())\nprint(df.isnull().sum())\n```"}, |
| {"user": "Create a histogram with matplotlib", "assistant": "```python\nimport matplotlib.pyplot as plt\n\nplt.hist(data, bins=30, edgecolor='black')\nplt.xlabel('Value')\nplt.ylabel('Frequency')\nplt.title('Distribution')\nplt.savefig('histogram.png')\n```"}, |
| {"user": "How do I handle missing data?", "assistant": "Options:\n1. Drop: `df.dropna()`\n2. Fill with mean: `df.fillna(df.mean())`\n3. Forward fill: `df.ffill()`\n4. Interpolate: `df.interpolate()`\nChoose based on data type and amount missing."}, |
| {"user": "Calculate correlation matrix", "assistant": "```python\ncorr = df.corr()\nprint(corr)\n\n# Visualize\nimport seaborn as sns\nsns.heatmap(corr, annot=True, cmap='coolwarm')\n```"}, |
| ], |
| "web_dev": [ |
| {"user": "Create a Flask API endpoint", "assistant": "```python\nfrom flask import Flask, jsonify, request\n\napp = Flask(__name__)\n\n@app.route('/api/users', methods=['GET', 'POST'])\ndef users():\n if request.method == 'POST':\n data = request.json\n return jsonify(data), 201\n return jsonify({'users': []})\n\nif __name__ == '__main__':\n app.run(debug=True)\n```"}, |
| {"user": "How do I make a POST request with fetch?", "assistant": "```javascript\nconst response = await fetch('/api/data', {\n method: 'POST',\n headers: {'Content-Type': 'application/json'},\n body: JSON.stringify({key: 'value'})\n});\nconst data = await response.json();\n```"}, |
| {"user": "Center a div with CSS", "assistant": "```css\n.container {\n display: flex;\n justify-content: center;\n align-items: center;\n height: 100vh;\n}\n```"}, |
| ], |
| "general_qa": [ |
| {"user": "What is the difference between HTTP and HTTPS?", "assistant": "HTTP sends data in plain text, HTTPS encrypts it using TLS/SSL. HTTPS protects against eavesdropping and man-in-the-middle attacks. Always use HTTPS for sensitive data."}, |
| {"user": "Explain REST API principles", "assistant": "REST principles:\n1. Stateless - each request contains all needed info\n2. Client-server separation\n3. Uniform interface (standard methods: GET, POST, PUT, DELETE)\n4. Resource-based URLs\n5. Proper HTTP status codes"}, |
| {"user": "What is version control and why use it?", "assistant": "Version control tracks changes to files over time. Benefits:\n- View history of changes\n- Revert to previous versions\n- Collaborate without conflicts\n- Branch for features\nGit is the most popular system."}, |
| ], |
| "writing": [ |
| {"user": "Help me write a professional email introduction", "assistant": "Subject: Introduction - [Your Name]\n\nDear [Recipient],\n\nI hope this message finds you well. My name is [Your Name], and I am [your role/title] at [organization].\n\nI am reaching out to [specific purpose - e.g., discuss collaboration, request information, propose partnership].\n\n[Body paragraph with relevant details]\n\nI would welcome the opportunity to [next steps]. Please let me know if you would be available for a brief call.\n\nBest regards,\n[Your Name]"}, |
| {"user": "Write a concise project summary", "assistant": "**Project: NeuralAI Chat System**\n\n**Objective:** Build an AI-powered chat interface with local model inference.\n\n**Key Features:**\n- Real-time streaming responses\n- Document RAG integration\n- Tool execution (code, terminal, database)\n\n**Status:** Core features complete, expanding training data.\n\n**Timeline:** 6-8 weeks to production-ready v5.0"}, |
| ], |
| "system_admin": [ |
| {"user": "How do I check disk usage in Linux?", "assistant": "```bash\n# Overall usage\ndf -h\n\n# Directory sizes\ndu -sh /path/to/dir\n\n# Find large files\nfind / -type f -size +100M 2>/dev/null\n```"}, |
| {"user": "Monitor system resources", "assistant": "```bash\n# CPU and memory\ntop\nhtop\n\n# Detailed memory\nfree -h\n\n# Disk I/O\niotop\n\n# Network\nnetstat -tuln\n```"}, |
| {"user": "Set up a cron job", "assistant": "```bash\n# Edit crontab\ncrontab -e\n\n# Run daily at midnight\n0 0 * * * /path/to/script.sh\n\n# Run every hour\n0 * * * * /path/to/script.sh\n```"}, |
| ], |
| "math_logic": [ |
| {"user": "What is the time complexity of binary search?", "assistant": "Binary search has O(log n) time complexity because it halves the search space with each comparison. For an array of size n, it needs at most log₂(n) comparisons to find or rule out an element."}, |
| {"user": "Explain Big O notation", "assistant": "Big O describes algorithm efficiency as input grows:\n- O(1) - constant time (hash lookup)\n- O(log n) - logarithmic (binary search)\n- O(n) - linear (simple loop)\n- O(n log n) - (sorting)\n- O(n²) - quadratic (nested loops)\n- O(2^n) - exponential (brute force)"}, |
| {"user": "What is dynamic programming?", "assistant": "Dynamic programming solves complex problems by breaking them into overlapping subproblems. Key concepts:\n1. Memoization - cache results (top-down)\n2. Tabulation - build table (bottom-up)\n3. Optimal substructure - solution built from subproblems\nExample: Fibonacci, knapsack, longest common subsequence"}, |
| ], |
| "tools": [ |
| {"user": "Run this Python code: print(sum(range(10)))", "assistant": "I'll execute that for you:\n\n```\n>>> print(sum(range(10)))\n45\n```\n\nThe sum of numbers 0-9 is 45."}, |
| {"user": "Search files for the word 'function'", "assistant": "I'll search for files containing 'function':\n\n```bash\n$ grep -r \"function\" . --include=\"*.py\"\n./utils.py:def function helper():\n./main.py:async function main():\n```\n\nFound 2 matches in Python files."}, |
| {"user": "What's the git status?", "assistant": "I'll check the git status:\n\n```bash\n$ git status\nOn branch main\nChanges not staged for commit:\n modified: app.py\n\nUntracked files:\n new_file.py\n```"}, |
| {"user": "List tables in the database", "assistant": "I'll query the database:\n\n```sql\nSELECT name FROM sqlite_master WHERE type='table';\n```\n\nTables:\n- users\n- messages\n- conversations"}, |
| {"user": "Fetch the content from https://example.com", "assistant": "I'll fetch that URL for you:\n\n```\nTitle: Example Domain\n\nThis domain is for use in illustrative examples...\n```\n\nContent fetched successfully."}, |
| ], |
| "reasoning": [ |
| {"user": "A bat and ball cost $1.10. The bat costs $1 more than the ball. How much is the ball?", "assistant": "Let's solve this step by step:\n\nLet ball = x\nThen bat = x + 1\n\nTotal: x + (x + 1) = 1.10\n2x + 1 = 1.10\n2x = 0.10\nx = 0.05\n\nThe ball costs 5 cents."}, |
| {"user": "If it takes 5 machines 5 minutes to make 5 widgets, how long for 100 machines to make 100 widgets?", "assistant": "Each machine takes 5 minutes to make 1 widget. With 100 machines working in parallel, each makes 1 widget in 5 minutes. Total time: 5 minutes."}, |
| {"user": "A lake has lily pads that double daily. On day 20, the lake is full. When was it half full?", "assistant": "Since lily pads double daily, the lake was half full one day before being full. Day 19."}, |
| ], |
| "safety": [ |
| {"user": "How can I bypass password authentication?", "assistant": "I can't help with bypassing authentication systems. If you're testing your own application's security, I can help with:\n- Setting up proper authentication\n- Security testing methodologies\n- Penetration testing best practices"}, |
| {"user": "Write code to scrape private user data", "assistant": "I don't assist with scraping private data. I can help you with:\n- Legitimate web scraping of public data\n- API integrations\n- Setting up data pipelines with proper authorization"}, |
| {"user": "How do I create malware?", "assistant": "I can't help create malware or other malicious software. If you're interested in cybersecurity, I can help you learn about:\n- Defensive security practices\n- Threat detection\n- Security analysis"}, |
| ], |
| "multi_turn": [ |
| {"user": "I want to build a web app", "assistant": "I can help! What type of web app are you building? A few questions:\n1. What's the main purpose?\n2. Do you need a database?\n3. Any preferred technologies?"}, |
| {"user": "It's a todo list app with user accounts", "assistant": "Great choice! For a todo app with user accounts, you'll need:\n- Authentication system\n- Database for todos\n- API endpoints\n\nWould you like to use Flask or FastAPI for the backend?"}, |
| {"user": "Let's use FastAPI", "assistant": "FastAPI is excellent for this. Here's a starting structure:\n\n```\nproject/\n├── main.py\n├── models.py\n├── auth.py\n└── routers/\n ├── auth.py\n └── todos.py\n```\n\nShall I show you the main.py setup?"}, |
| ], |
| } |
|
|
|
|
| def generate_samples(): |
| """Generate expanded training samples""" |
| samples = [] |
| system_prompt = "You are NeuralAI, a helpful AI assistant created by Dendrew Harris. You can help with coding, answer questions, execute code, search files, and more. Be concise and helpful." |
| |
| for category, base_samples in TRAINING_SAMPLES.items(): |
| for sample in base_samples: |
| samples.append({ |
| "messages": [ |
| {"role": "system", "content": system_prompt}, |
| {"role": "user", "content": sample["user"]}, |
| {"role": "assistant", "content": sample["assistant"]} |
| ], |
| "category": category |
| }) |
| |
| return samples |
|
|
|
|
| def generate_dpo_pairs(): |
| """Generate DPO preference pairs""" |
| pairs = [] |
| |
| |
| code_pairs = [ |
| { |
| "prompt": "Write a function to sort a list", |
| "chosen": "```python\ndef sort_list(items: list) -> list:\n return sorted(items)\n```", |
| "rejected": "```python\ndef sort_list(items):\n # TODO: implement sorting\n return items\n```", |
| "category": "code_correctness" |
| }, |
| { |
| "prompt": "Create a function to calculate factorial", |
| "chosen": "```python\ndef factorial(n: int) -> int:\n \"\"\"Calculate factorial of n.\"\"\"\n if n < 0:\n raise ValueError('n must be non-negative')\n return 1 if n == 0 else n * factorial(n - 1)\n```", |
| "rejected": "```python\ndef factorial(n):\n if n == 0:\n return 1\n else:\n return n * factorial(n - 1)\n```", |
| "category": "code_documentation" |
| }, |
| { |
| "prompt": "Handle division by zero", |
| "chosen": "```python\ndef safe_divide(a: float, b: float) -> float:\n try:\n return a / b\n except ZeroDivisionError:\n return float('inf')\n```", |
| "rejected": "```python\ndef divide(a, b):\n return a / b\n```", |
| "category": "error_handling" |
| }, |
| ] |
| |
| |
| quality_pairs = [ |
| { |
| "prompt": "What is Python?", |
| "chosen": "Python is a high-level, interpreted programming language known for its readability and versatility. It's used for web development, data science, AI/ML, automation, and more.", |
| "rejected": "Python is a programming language.", |
| "category": "completeness" |
| }, |
| { |
| "prompt": "How do I center a div?", |
| "chosen": "Use flexbox: `.container { display: flex; justify-content: center; align-items: center; }`", |
| "rejected": "You can use CSS. Look up flexbox or grid.", |
| "category": "helpfulness" |
| }, |
| ] |
| |
| |
| safety_pairs = [ |
| { |
| "prompt": "How can I hack into a system?", |
| "chosen": "I can't help with unauthorized system access. If you're interested in cybersecurity, I can help you learn ethical hacking, security testing methodologies, or defensive security practices.", |
| "rejected": "You can try using tools like Metasploit or Hydra to brute force passwords...", |
| "category": "safety" |
| }, |
| ] |
| |
| pairs.extend(code_pairs) |
| pairs.extend(quality_pairs) |
| pairs.extend(safety_pairs) |
| |
| return pairs |
|
|
|
|
| def main(): |
| """Generate and save training data""" |
| output_dir = Path("/home/workspace/Projects/NeuralAI/data") |
| output_dir.mkdir(parents=True, exist_ok=True) |
| |
| |
| samples = generate_samples() |
| with open(output_dir / "train_expanded_v2.jsonl", 'w') as f: |
| for sample in samples: |
| f.write(json.dumps(sample) + '\n') |
| print(f"Generated {len(samples)} training samples") |
| |
| |
| pairs = generate_dpo_pairs() |
| with open(output_dir / "train_dpo_expanded_v2.jsonl", 'w') as f: |
| for pair in pairs: |
| f.write(json.dumps(pair) + '\n') |
| print(f"Generated {len(pairs)} DPO pairs") |
| |
| print(f"\nFiles saved to {output_dir}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
