README.md

🏆 Transformer Hackathon Leaderboard

Build Your Own GPT • Train • Compete • Win!

📊 Current Standings

Lower perplexity = better model performance

Rank	Name	Team	Perplexity ⬇️	Loss	Tokens/sec	Timestamp
🥇	Kay sep	kay	1.9439	0.6647	67774	2026-02-07 21:29
🥈	Praneeth	Solo	1.9832	0.6847	91551	2026-02-07 21:35
🥉	James Faul, Vatsala	BA	1.9970	0.6917	149140	2026-02-07 20:18
4	ImageBear	Solo	1.9983	0.6923	67189	2026-02-07 20:10
5	Matt O	Solo	2.0216	0.7039	64286	2026-02-07 21:05
6	weldon	Solo	2.0280	0.7070	65689	2026-02-07 21:12
7	Nick	Solo	2.0281	0.7071	63951	2026-02-07 21:38
8	Megumi	Solo	2.0304	0.7082	65614	2026-02-07 21:40
9	Jack G. Conrad	Solo	2.0476	0.7167	66417	2026-02-07 21:39
10	Abhi	BlueBird	2.0494	0.7175	0	2026-02-07 07:43
11	Abhi_1	Redbird	2.0494	0.7175	0	2026-02-07 07:51
12	Allen	Solo	2.0531	0.7193	65446	2026-02-07 21:35
13	Mark Pedigo	Solo	2.0546	0.7201	65582	2026-02-07 21:32
14	Abhi	Red Team	2.0571	0.7213	64682	2026-02-07 21:35
15	DSM	Solo	2.0661	0.7256	63707	2026-02-07 21:35
16	Maheshwari	Solo	2.0674	0.7263	66315	2026-02-07 21:08
17	DSM	Solo	2.0693	0.7272	67373	2026-02-07 19:46
18	H100 5min D1024	Solo	2.1888	0.7833	89296	2026-02-07 21:00
19	H100 5min	Solo	2.2233	0.7990	143639	2026-02-07 20:18
20	Kara	Conrads	2.3491	0.8540	66174	2026-02-07 20:57
21	Scott M	Solo	2.9152	1.0699	67890	2026-02-07 20:23
22	MacbookPro 5min 2layer	Solo	3.4363	1.2344	11940	2026-02-07 21:09
23	MacBookPro 5min	Solo	4.7225	1.5523	4055	2026-02-07 20:13
24	TestUser	TestTeam	99.9900	4.5000	1000	2026-02-07 08:02
25	TestUser	TestTeam	99.9900	4.5000	1000	2026-02-07 08:03
26	Hedgehog	Solo	1337.0000	2.3457	1875	2026-02-07 19:34

---

🎯 Competition Categories

• Best Perplexity: Lowest perplexity score wins! 🏆
• Best Efficiency: Highest tokens/second during training ⚡
• Most Creative: Best generated text samples 🎨

🚀 How to Participate

# Clone the repository
git clone https://github.com/abhishekadile/Transformer_Repo-
cd Transformer_Repo-

# Install dependencies
pip install -r requirements.txt

# Run the hackathon pipeline
python run_hackathon.py

The script will:

1. Download the TinyStories dataset
2. Train your model for 45 minutes
3. Evaluate performance
4. Submit to this leaderboard!

📈 Leaderboard Schema

• name: Participant name
• team: Team name (optional)
• perplexity: Model perplexity - lower is better
• loss: Final validation cross-entropy loss
• tokens_per_sec: Training throughput
• timestamp: Submission time

🔧 Optimization Tips

• 🚀 Enable mixed precision: --use-amp
• 📦 Increase batch size if GPU allows
• 📈 Try different learning rates
• 🧠 Experiment with model size in config.py

📚 Resources

• GitHub Repository
• Google Colab Guide
• Attention Is All You Need Paper

Good luck and have fun! 🎉