Code Generation on HumanEval (Accuracy, Retention Percentage)
100Retention (%)Llada-Instruct
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Llada-InstructLayers Skipped=02026.03 | 100 | 51 | |
| Llada-InstructLayers Skipped=22026.03 | 100 | — | |
| Llada-InstructLayers Skipped=32026.03 | 100 | — | |
| Dream-BaseLayers Skipped=02026.03 | 100 | 65 | |
| Dream-InstructLayers Skipped=02026.03 | 100 | 64 | |
| Qwen2.5Layers Skipped=02026.03 | 100 | 68 | |
| Llada-InstructLayers Skipped=42026.03 | 92.2 | — | |
| Llada-InstructLayers Skipped=62026.03 | 88.2 | — | |
| Llada-InstructLayers Skipped=52026.03 | 82.3 | — | |
| Dream-InstructLayers Skipped=22026.03 | 70.3 | — | |
| Dream-BaseLayers Skipped=22026.03 | 66.2 | — | |
| Qwen2.5Layers Skipped=22026.03 | 64.7 | — | |
| Dream-BaseLayers Skipped=32026.03 | 63.1 | — | |
| Llada-InstructLayers Skipped=82026.03 | 62.7 | — | |
| Dream-InstructLayers Skipped=32026.03 | 59.4 | — | |
| Qwen2.5Layers Skipped=32026.03 | 48.5 | — | |
| Dream-BaseLayers Skipped=42026.03 | 43.1 | — | |
| Qwen2.5Layers Skipped=42026.03 | 41.2 | — | |
| Dream-InstructLayers Skipped=42026.03 | 26.6 | — |