Code Generation on MBPP (val)
220Pass@1DeepSeekMoE
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeekMoE# Shot=3-shot, # Total Params=2.0B, # Activated Params=0.3B, FLOPs per 2K Tokens=4.3T, # Training Tokens=100B2024.01 | 220 | |
| Hash Layer# Shot=3-shot, # Total Params=2.0B, # Activated Params=0.2B, FLOPs per 2K Tokens=2.9T, # Training Tokens=100B2024.01 | 60 | |
| Switch# Shot=3-shot, # Total Params=2.0B, # Activated Params=0.2B, FLOPs per 2K Tokens=2.9T, # Training Tokens=100B2024.01 | 40 | |
| Dense# Shot=3-shot, # Total Params=0.2B, # Activated Params=0.2B, FLOPs per 2K Tokens=2.9T, # Training Tokens=100B2024.01 | 20 | |
| GShard# Shot=3-shot, # Total Params=2.0B, # Activated Params=0.3B, FLOPs per 2K Tokens=4.3T, # Training Tokens=100B2024.01 | 20 |