Content Generation on CoGenesis human assessment and automated evaluation
50Tie RateSLM finetuned
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SLM finetunedModel and Setting=SLM finetuned, Base Model=Qwen-1.8B2024.03 | 50 | — | — | 2.07 | 13.95 | |
| Logits-based CoGenModel and Setting=Logits-based CoGen, LLM=Qwen-72B, SLM=Qwen-1.8B2024.03 | 15 | 32 | 53 | 2.3 | 14.18 | |
| Sketch-based CoGenModel and Setting=Sketch-based CoGen, LLM=Qwen-72B, SLM=Qwen-1.8B2024.03 | 13 | 27 | 60 | 1.81 | 12.98 | |
| LLM w/ contextModel and Setting=LLM w/ context, Base Model=Qwen-72B2024.03 | 12 | 38 | 50 | 2.61 | 14.66 | |
| LLM w/o contextModel and Setting=LLM w/o context, Base Model=Qwen-72B2024.03 | 10 | 3 | 87 | 1.51 | 13.54 |