Attention Is All You Need
Paper
•
1706.03762
•
Published
•
109
entries
list | last_updated
null | version
string | schema
dict | competition
dict |
|---|---|---|---|---|
[] | null |
1.0
|
{
"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": "ISO 8601 submission timestamp",
"model_config": "Optional model configuration details"
}
|
{
"name": "Transformer Hackathon",
"training_time_minutes": 45,
"dataset": "TinyStories",
"github": "https://github.com/abhishekadile/Transformer_Repo-"
}
|
🏆 Transformer Hackathon Leaderboard
Build Your Own GPT • Train • Compete • Win!
📊 Current Standings
| Rank | Name | Team | Perplexity ⬇️ | Loss | Tokens/sec |
|---|---|---|---|---|---|
| 🥇 | Be the first! | - | - | - | - |
Lower perplexity = better model performance
🎯 Competition Categories
| Category | Metric | Prize |
|---|---|---|
| 🏆 Best Performance | Lowest Perplexity | 1st Place |
| ⚡ Most Efficient | Highest Tokens/sec | Speed Award |
| 📝 Best Generation | Text Quality | Creativity Award |
🚀 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:
- Download the TinyStories dataset
- Train your model for 45 minutes
- Evaluate performance
- Submit to this leaderboard!
📈 Leaderboard Schema
| Field | Type | Description |
|---|---|---|
name |
string | Participant name |
team |
string | Team name |
perplexity |
float | Model perplexity (lower = better) |
loss |
float | Final validation loss |
tokens_per_sec |
float | Training speed |
timestamp |
string | Submission time (UTC) |
🔧 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
Good luck and have fun! 🎉
- Downloads last month
- 20
Size of downloaded dataset files:
679 Bytes
Size of the auto-converted Parquet files:
7.52 kB
Number of rows:
1