🔹 Key Highlights:
- 13% Fewer Parameters: nyun-c2-llama3-61B comprises approximately 13% fewer parameters than the popular Llama-3-70B.
- Better Performance: Despite having fewer parameters, this model performs better than Llama3-70B on multiple benchmarks.
- No Fine-Tuning Required: This model undergoes no fine-tuning, showcasing the raw potential of our optimization techniques.
Pipeline and Collaboration
For insights into the pipeline and the list of methods used to optimize these models, check out our PruneGPT repository (https://github.com/nyunAI/PruneGPT). We invite companies and organizations interested in joining forces with us to release more such open-source variants to reach out at contact@nyunai.com.
Model Performance
| Dataset | nyun-c2-llama3-61B | Meta-Llama3-70B | Meta-Llama2-70B | MBZUAI K2-65B |
|---|---|---|---|---|
| MMLU (5-shot) | 78.8 | 79.5 | 69.7 | 67.9 |
| Winogrande (5-shot) | 86.2 | 83.1 | 81.8 | 77.0 |
| BoolQ (0-shot) | 85.1 | 79.0 | 73.1 | 83.0 |
| Hellaswag (10-shot) | 87.4 | 88.0 | 86.9 | 85.5 |
| Arc Challenge (25-shot) | 67.6 | 68.8 | 67.2 | 64.8 |
| GSM8K (5-shot) | 79.4 | 76.9 | 52.6 | 50.2 |
| Average | 80.7 | 79.2 | 71.9 | 71.4 |