🔹 Key Highlights:
- 29% Fewer Parameters: nyun-c2-llama3-50B comprises approximately 29% fewer parameters than the popular Llama-3-70B.
- Comparable Performance: Despite having far fewer parameters, this model undergoes minimal performance degredation.
- 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-llama3-50B | Meta-Llama3-70B | Meta-Llama2-70B | MBZUAI K2-65B |
|---|---|---|---|---|
| MMLU (5-shot) | 78.4 | 79.5 | 69.7 | 67.9 |
| Winogrande (5-shot) | 85.3 | 83.1 | 81.8 | 77.0 |
| BoolQ (0-shot) | 83.9 | 79.0 | 73.1 | 83.0 |
| Hellaswag (10-shot) | 85.4 | 88.0 | 86.9 | 85.5 |
| Arc Challenge (25-shot) | 65.4 | 68.8 | 67.2 | 64.8 |
| GSM8K (5-shot) | 64.7 | 76.9 | 52.6 | 50.2 |
| Average | 77.2 | 79.2 | 71.9 | 71.4 |